Insight Editions · No. 245
The Future of the Mind
The Scientific Quest to Understand, Enhance, and Empower the Mind
Michio Kaku
Kaku’s tour of neuroscience through the tools of physics: a working definition of consciousness, and what brain scanning, telepathy, memory editing, AI and uploading may make possible.
A study guide to The Future of the Mind by Michio Kaku. Shared free for personal reading; not for sale and not endorsed by the authors or publishers. Please read and buy the original.
Physics has opened the living brain, and consciousness is model-building through feedback loops; humans uniquely simulate the future, which frames telepathy, memory upload, AI and alien minds.
From Phineas Gage and Penfield to fMRI decoding, BrainGate, memory prosthetics, DBS, robot emotions, connectomes, minds on laser beams and quantum free will (2014, with 2026 updates).
100 billion neurons, 20 watts, Level III:100 average human, 1 zettabyte connectome, $3B BRAIN vs €1.19B HBP, 96-electrode BrainGate chip.
Readers who want the whole science of the mind in one pass, and technologists mapping neurotech and AI timelines against 2026 reality.
The Book on One Page
Kaku, a theoretical physicist, argues that the instruments of physics have done for the mind what the telescope did for the stars, and that a single definition of consciousness (building a model of the world with feedback loops and, for humans, simulating it forward in time) can organise everything from dreams to robots to aliens.
- The tools. MRI, fMRI, DTI, EEG, MEG, PET, TMS, DBS, optogenetics and Clarity let us watch and switch the living brain. Since there are only four forces, Kaku expects refinement, not revolution: smaller, cheaper, sharper scanners.
- The theory. His space-time theory of consciousness ranks minds: Level 0 (thermostats, plants), Level I (space; reptiles), Level II (others and emotion; social mammals), Level III (time; humans). Each feedback loop is one unit, so consciousness becomes measurable in principle. The brain works like a corporation whose CEO, the dorsolateral prefrontal cortex, sees a sliver of its subconscious bureaucracy, and the left-brain interpreter confabulates a unified "I".
- Mind over matter. Computers learn a dictionary between brain signals and the world: fMRI movies of what you see, ECoG word decoding, typing by thought, BrainGate arms, monkeys moving their own paralysed limbs, brain-to-brain links, exoskeletons and surrogate robots. Telekinesis is real once an outside power source supplies the force.
- Memory and intelligence. All long-term memories pass through the hippocampus, so USC and Wake Forest could record and replay a rodent memory; MIT implanted a false one. Genes (NR2B, CREB), drugs and TMS might boost memory; savants hint at latent abilities; but physics caps biological intelligence, so gains must come from efficiency and outside help. Intelligence itself is the complexity of one's future simulations.
- Altered states. Dreams are the cortex narrating random brain-stem bursts with the CEO and fact-checker off; mind control failed in MKULTRA but precise tools now make it thinkable; mental illness is the disruption of balanced feedback loops, treatable by DBS and new drugs.
- Other minds. Robots in 2013 sat at Level I:10, lacking common sense and pattern recognition; they need emotions to know what matters, a value system, and safe goal hierarchies. Aliens will share only the definition of consciousness; they are likely post-biological and indifferent, contacting us by signal or a von Neumann probe on the Moon.
- Beyond the body. Connectomes could be stored, run in computers, transferred neuron by neuron, or beamed by laser to surrogates across the galaxy, but the Caveman Principle (we prefer high-touch to high-tech) will shape what people accept.
- The verdict. Bill Joy's fears are overstated except for biotech; wisdom comes from democratic debate; free will survives through quantum uncertainty and chaos; understanding the brain deepens rather than dispels its wonder.
| Ch | Title | One-line payload |
|---|---|---|
| Intro | The Two Great Mysteries | Physics' scanners plus Moore's law launched a golden age of neuroscience; Kaku's two rules: obey physics, prototype must exist |
| 1 | Unlocking the Mind | From Gage to fMRI and optogenetics; the brain as a corporation; split brains show the self is constructed |
| 2 | Consciousness: A Physicist's Viewpoint | Consciousness = model-building via feedback loops; Levels 0 to III; humans simulate the future |
| 3 | Telepathy | fMRI movies, ECoG words, EEG typing; physics protects privacy for now, law must do the rest |
| 4 | Telekinesis | BrainGate, spinal bypass, brain-to-brain links, exoskeletons, surrogates; mind commands, power source acts |
| 5 | Memories Made to Order | Hippocampal record-and-replay, false memories, memory genes, forgetting drugs, Alzheimer's, legal shocks |
| 6 | Einstein's Brain and Intelligence | Genius = future simulation plus tenacity; savants, TMS, HAR1/ASPM; physics caps biological IQ |
| 7 | In Your Dreams | Activation-synthesis; dream decoding; lucid dreamers signal from inside |
| 8 | Can the Mind Be Controlled? | Delgado's bull, MKULTRA's failure, drugs' reward hijack, optogenetics' precision risk |
| 9 | Altered States | Mental illness as unbalanced feedback loops; DBS "Depression 3.0"; shared genetics |
| 10 | The Artificial Mind | Robots at Level I:10; emotions give value; goal conflicts, friendly AI, merging with machines |
| 11 | Reverse Engineering the Brain | Simulate, map or annotate; BRAIN vs HBP; a map is not understanding |
| 12 | Mind Beyond Matter | OBEs and NDEs explained; uploads, gradual transfer, nanobots; the Caveman Principle |
| 13 | The Mind as Pure Energy | Connectomes on laser beams, relay networks, wormholes |
| 14 | The Alien Mind | SETI, Fermi, umwelt, bee-minds, post-biological aliens, von Neumann probes |
| 15 | Concluding Remarks | Reply to Bill Joy; Copernican vs Anthropic principles; consciousness as precious |
| App. | Quantum Consciousness? | Cat paradox interpretations; free will via quantum chance and chaos |
The Two Great Mysteries
Kaku opens by pairing the mind and the universe as the two deepest unknowns, and argues that the instruments of physics (MRI, EEG, PET, DBS) have done for the brain what the telescope did for the sky. In roughly fifteen years of brain scanning, he claims, we learned more about the brain than in all prior history, and Moore's law is now driving a second revolution that links brains directly to computers.
Mind and universe as twin frontiers
Kaku sets the scale with a deliberate symmetry: the Milky Way holds about 100 billion stars, roughly the number of neurons in a human brain, and you would have to travel about 24 trillion miles to the nearest star to find an object as complex as the one on your shoulders. The two frontiers are opposites in scale, yet both were long owned by superstition: astrologers read the zodiac, phrenologists read bumps on the skull, and stage mentalists sold mind reading.
His own entry point was science fiction. As a boy he wanted to be one of A. E. Van Vogt's Slan telepaths, admired Asimov's Mule in the Foundation trilogy, and studied the telekinetic Krell of Forbidden Planet. He then watched "The Amazing Dunninger" on TV "beam" a president's name to millions, and thousands of postcards came back saying Roosevelt. Kaku was unimpressed: after the Depression and the war, Roosevelt was the obvious guess, and it would only have been remarkable if the answer had been Millard Fillmore.
Dunninger's own motto gives the game away: "For those who believe, no explanation is necessary; for those who do not believe, no explanation will suffice." Kaku frames the whole inquiry with Thomas Huxley's "question of all questions": the determination of man's place in Nature and his relation to the Cosmos. The mind and the universe are opposites in scale (black holes and colliding galaxies against our most private hopes), yet the mind "is no farther than our next thought" and still resists explanation. Kaku's earlier books split the same way: Hyperspace, Beyond Einstein and Parallel Worlds on fundamental law; Visions, Physics of the Impossible and Physics of the Future on how science shapes daily life. This book joins the two threads.
Kaku tried to read minds and move objects by concentration alone, and failed. The lesson he drew shapes the whole book: telepathy and telekinesis are impossible without outside assistance, and the way to find where the possible ends is to know the laws of physics. The rest of the book is about that outside assistance: scanners, chips and computers.
A long dark age for the brain
For most of history the brain was misread. The ancient Egyptians discarded it during embalming as useless; Aristotle placed the soul in the heart and thought the brain merely cooled the blood; Descartes routed the soul through the pineal gland. None of these could be tested.
The difficulty is the organ itself. The brain weighs about three pounds, is 2% of body weight but burns 20% of the body's energy (about 65% in newborns), and Kaku says about 80% of our genes code for it. It holds an estimated 100 billion neurons with an astronomical number of connections.
When Carl Sagan summarised brain science in The Dragons of Eden (1977, Pulitzer Prize), it rested on three indirect methods: 1. Comparative anatomy: dissecting the brains of thousands of animals and comparing them with ours. 2. Lesion studies: observing stroke and disease victims, then confirming the damaged area only at autopsy. 3. Electrode probing: slowly mapping which region drives which behaviour. None could show how regions interact in a living brain to produce love, jealousy or curiosity.
Twin revolutions: the telescope and the scanner
The telescope, invented about 400 years ago, destroyed celestial dogma within fifteen years: craters on the Moon, spots on the Sun, moons of Jupiter, phases of Venus, rings of Saturn. Kaku argues MRI and related scans from the mid-1990s and 2000s did the same for the brain. Eric Kandel (Nobel laureate) credits the key insights to the merger of psychology with brain biology, and V. S. Ramachandran says questions philosophers chewed for millennia are now testable by imaging.
Kaku stresses that physics supplied the tools: MRI, EEG, PET, CAT, TCM, TES and DBS. His own life brushed against them. In high school he obtained sodium-22, a positron emitter, from the Atomic Energy Commission to photograph antimatter tracks in a cloud chamber; positron emitters later became the heart of PET. He heard Felix Bloch (1952 Nobel, with Edward Purcell) explain nuclear magnetic resonance: atoms align in a strong field, a radio pulse at the resonant frequency flips them, and the echo as they flip back identifies them. He used the same principle to build a 2.3-million-electron-volt accelerator in his mother's garage, was taught electrodynamics by Purcell at Harvard, and had a summer job with Richard Ernst, whose generalisation of magnetic resonance won the 1991 Nobel (in fact the Chemistry prize) and underpins modern MRI.
Two slips in the introduction. Kaku places Eric Kandel at the "Max Planck Institute in Tübingen"; Kandel's career has been at Columbia University. He also gives Richard Ernst's 1991 Nobel as the Physics prize; it was the Nobel Prize in Chemistry. The "80% of genes code for the brain" figure is loose: a large share of genes are expressed in the brain, but that is not the same as being dedicated to it. Modern counts put human neurons at about 86 billion, not 100 billion.
Empowering the mind: the book's promise
Kaku lists what physics-plus-computing already allows or soon will:
- Mind reading by MRI. Scans can decode some thoughts circulating in the brain.
- Chips in paralysed patients. Through thought alone they surf the web, write e-mail, play games, steer wheelchairs and move mechanical arms.
- Exoskeletons and surrogates. Quadriplegics may walk; astronauts may explore planets by mentally steering robots from home.
- Memory upload. Memories have already been inserted into animal brains; downloading skills, Matrix-style, could reshape economies.
- A brain-net. An "Internet of the mind" where thoughts, emotions and even recorded dreams are "brain-mailed".
- Enhanced intelligence. Lessons from savants, and genes in animals that boost memory and performance.
-
Reverse engineering the brain. In January 2013 both President Obama (the BRAIN Initiative) and the European Union (the Human Brain Project) announced projects that could reach billions of dollars.
-
A transatlantic race. Kaku treats the two projects as a rivalry between the largest economies, comparable to the Human Genome Project: they should spawn new industries, and once the pathways are decoded, pinpoint the precise origins of mental illness. Decoding also makes a copy of the brain possible, which raises the questions the last chapters chase: who are we if consciousness can be uploaded, and can it outlive the body?
- Mind beyond the body. Copying the brain, digital immortality, and eventually sending a neural blueprint on laser beams to the stars.
Kaku sets himself two rules for every prediction in the book: (1) it must obey the laws of physics, and (2) a prototype must already exist that proves the principle. Use these as the yardstick when judging his more exotic chapters.
Why it matters personally
Mental illness runs through Kaku's subject. Einstein's younger son Eduard had schizophrenia, and Einstein's colleague Paul Ehrenfest, who helped him with general relativity, suffered depression and died in a murder-suicide that also took his son, who had Down's syndrome. Kaku's own mother died of Alzheimer's disease, slowly losing the memory of her family. As the baby boomers age, he expects this experience to become universal, and hopes neuroscience can relieve it.
What is driving the revolution: Maxwell plus Moore
The telescope needed 350 years to lead to the space age; the scanner needed about fifteen years to connect brains to the outside world. Kaku gives two engines:
- Electromagnetism. The physics of antennas and radar, codified by James Clerk Maxwell, is the cornerstone of MRI and of reading the electrical signals in neurons.
- Moore's law. Computing power doubles about every two years; a cell phone now outclasses all of NASA in 1969. Computers can now record brain signals and partly translate them, the field of brain-machine interface (BMI).
The structure follows: Book I surveys the brain and gives a physicist's definition of consciousness that can be ranked with numbers; Book II covers BMI, mind reading, recorded memories, dream video and telekinesis; Book III covers altered states (dreams, drugs, mental illness), mind control, the BRAIN and Human Brain Project efforts, robot and alien consciousness. The Appendix takes on the idea that consciousness is the basis of reality in quantum physics. Kaku is candid that he is a theoretical physicist, not a neuroscientist, offering a physicist's vantage point.
Since 2014: the scanner-plus-computer engine Kaku described has been joined by a third: machine learning. Since 2023 the biggest leaps in decoding (speech from implants, sentences from fMRI) came from AI language models trained on brain data. The EU Human Brain Project ended in 2023 and handed over to the EBRAINS research infrastructure; the US BRAIN Initiative continues, though its budget fell sharply after 2023.
The Mind and Consciousness
Book I lays the foundations: how physics opened the living brain to view, and a physicist's definition of consciousness that ranks every living thing (and machine) on a numbered scale.
Unlocking the Mind
Kaku traces the brain from Phineas Gage's iron rod to fMRI, optogenetics and transparent brains, and argues that the brain is best pictured as a large corporation whose CEO, the prefrontal cortex, sees only a sliver of what its subconscious bureaucracy does. Split-brain patients then show that even the "single self" is a construction.
From lesions to localisation
Phineas Gage, a Vermont railroad foreman, survived an 1848 blasting accident that drove a 3 ft 7 in iron rod up through his face and out the top of his skull, landing about 80 feet away. He recovered physically, but his doctor John Harlow recorded a man transformed: once cheerful and reliable, Gage became profane, impulsive and unable to stick to a plan, "no longer Gage". Later X-ray work on his preserved skull confirmed massive damage to the frontal lobes. The case pushed science away from dualism (brain and soul as separate things) toward the idea that specific brain regions govern specific behaviours.
Harlow's own words fix the picture: Gage was "capricious and vacillating", forever devising plans that were abandoned as soon as arranged, "a child in his intellectual capacity" with "the animal passions of a strong man"; ladies were warned to keep away from him. Gage lived until 1860, and Harlow preserved both the skull and the rod. A rare photograph surfaced in 2009 showing a handsome, confident man holding the iron, his left eye damaged. The lesson Kaku draws is the one that launched modern neuroscience: lose the frontal lobes and you lose the capacity to plan and stick to a future, a point that becomes the core of his theory of consciousness in chapter 2.
The localisation case then built quickly:
- Broca (1861). Pierre Paul Broca in Paris studied a patient who understood speech but could say only "tan". The autopsy showed a left-hemisphere lesion, and twelve similar cases followed. Broca's aphasia leaves comprehension intact but speech halting or absent.
- Wernicke (1874). Carl Wernicke described the mirror image: fluent, grammatical speech made of meaningless jargon, with no comprehension, from damage to a different part of the left temporal lobe. Patients often did not realise they were speaking gibberish.
- Fritsch (1864). Treating soldiers with open head wounds in the Prusso-Danish War, Gustav Fritsch noticed that touching one hemisphere made the opposite side of the body twitch. Systematic electrical stimulation showed left brain controls right body and vice versa, proving the brain is electrical. (Romans had tried it first: in A.D. 43 the court physician to Emperor Claudius applied electric torpedo fish to a patient's head for headaches.)
- Penfield (1930s to 1951). Wilder Penfield, operating on conscious epilepsy patients under local anaesthetic (the brain has no pain sensors), mapped which cortical strip controls which body part. Stimulating the temporal lobe made patients relive vivid memories ("standing in the doorway at my high school... I heard my mother talking on the phone"). He published in 1951.
Two factual slips. Gage's accident was caused by blasting powder, not dynamite (dynamite was invented in 1867, nearly two decades later). And Broca's area lies in the left inferior frontal lobe, not the temporal lobe as Kaku writes; Wernicke's area is the temporal one. Fritsch's decisive work was the 1870 dog experiments with Eduard Hitzig; the battlefield observation is the anecdotal prelude.
A map of the brain: four lobes
By the 1950s and 1960s a crude atlas existed. The neocortex has four lobes, and all but one process the senses:
| Lobe | Location | Main jobs | Damage causes |
|---|---|---|---|
| Frontal (prefrontal cortex) | Behind the forehead | Rational thought, planning, evaluating senses, choosing action | Inability to plan or imagine the future (Gage) |
| Parietal | Top of the head | Right: sensory attention, body image. Left: skilled movement, some language | Trouble locating parts of your own body |
| Occipital | Back of the head | Vision | Blindness, visual impairment |
| Temporal | Near the ears | Language (left side), face recognition, some emotions | Loss of speech, inability to recognise faces |
The evolving brain: MacLean's triune model
Mapping the brain has been called "cartography for fools" because its parts look slapped together. In 1967 Paul MacLean of the National Institute of Mental Health organised it by evolution into three layers, which Kaku adopts as a rough guide while admitting later refinements.
- Reptilian brain (brain stem, cerebellum, basal ganglia; about 500 million years old). Balance, breathing, digestion, heartbeat, blood pressure; also fighting, hunting, mating and territory.
- Mammalian brain / limbic system (centre of the brain). Emotions and the social calculus of enemies, allies and rivals. Its parts: - Hippocampus ("seahorse"): turns short-term into long-term memory. Lose it and you are "a prisoner of the present". - Amygdala ("almond"): seat of emotion, especially fear. - Thalamus ("inner chamber"): relay station sending sensory signals to the cortices. - Hypothalamus: body temperature, circadian rhythm, hunger, thirst, reproduction and pleasure.
- Neocortex ("new bark"), the outer layer: higher cognition. In humans it is about 80% of brain mass yet only napkin-thin, deeply folded to pack in surface area (a rat's is smooth).
Kaku likens the brain to a museum of earlier evolutionary stages, and notes that an infant's brain also grows outward and forward, as if replaying that history.
MacLean's triune brain is now rejected by most neuroscientists as evolution: reptiles and birds have homologues of cortex and limbic structures, and brains did not evolve by stacking new layers on old ones. Kaku flags "refinements" but the model still shapes his later levels of consciousness. Treat it as a teaching scaffold, not anatomy.
The neuron
Under the microscope the grey matter is billions of neurons. Each receives signals through branching dendrites and sends them down a long axon, which may connect to up to 10,000 other neurons. The gaps between them, synapses, act as gates; neurotransmitters such as dopamine, serotonin and noradrenaline modulate the flow and therefore mood, emotion and thought. That picture held through the 1980s.
The physicist's tool kit
From the 1990s, instruments from physics opened the living brain. Kaku walks through each with its strengths and limits.
MRI and fMRI
An MRI magnet is 20,000 to 60,000 times stronger than Earth's field, weighs about a ton and costs millions. Nuclei in the body align with the field like compass needles; a radio pulse flips some of them; as they relax they emit an echo that computers turn into a 3-D image, much as a bat uses echoes. MRI uses no ionising radiation, unlike X-rays and CT, though careless staff have been injured or killed by tools flung by the magnet.
- fMRI (mid-1990s). Tracks oxygenated blood as a proxy for neural energy use, showing that thinking is not concentrated in one centre but circulates among regions.
- Voxels. Resolution down to a fraction of a millimetre, in 3-D "voxels" rather than flat pixels.
- DTI (diffusion tensor imaging). Follows water flow along nerve fibres, revealing the wiring as images that look like vines in a garden.
- Weak spot: time. Blood flow takes almost a second, while neural signals are nearly instant.
-
Lie detection. Some studies claim 95% accuracy because lying requires holding the truth, inventing the lie and checking consistency, lighting up the prefrontal and parietal cortex and especially the orbitofrontal cortex, the brain's "fact-checker". Kaku cautions that DNA evidence can be accurate to 1 part in 10 billion, while lying uses regions shared with many other thoughts.
-
Cost and sharing. At several million dollars each, machines are shared between doctors, which slows research; Kaku expects prices to fall as with all technology. Lying also lights the superior medial and inferolateral prefrontal cortices, regions busy with ordinary cognition, which is the root of the false-positive problem. Several firms already sell fMRI lie detection and cases have reached court.
- Tunable atoms. Because each element resonates at its own radio frequency, an MRI can be retuned to look for different substances, which is how DTI was born and how future machines may track molecules more directly tied to neural firing, giving real-time pictures of thought.
MRI physics, simplified too far. fMRI does not "zero in on the oxygen atom". Clinical MRI images hydrogen nuclei (protons in water), and fMRI's BOLD signal works because deoxygenated haemoglobin is magnetic and distorts that proton signal. On lie detection, US courts have refused fMRI evidence (United States v. Semrau, 2012), and accuracy outside the lab is unproven.
EEG, PET and magnetism
- EEG (since 1924): passive helmet or hairnet of electrodes. Instant timing, cheap enough for a living room, but poor spatial resolution because the skull smears the signal; a finger twitch can ruin a trace. Brain rhythms: delta 0.1 to 4 Hz (deep sleep), beta 12 to 30 Hz (active problem solving).
- PET: traces glucose use with a positron-emitting tracer. Measures energy consumption directly (closer to neural activity than blood flow) but coarser than MRI and mildly radioactive, so roughly once a year per person.
- MEG: passively reads the brain's magnetic fields (a billionth of Earth's). Millisecond timing, but only 1 cm³ resolution.
- TES / TMS: a pulse of current makes a magnetic burst that induces current inside the brain and temporarily switches off a region (pulsing the left temporal lobe disrupts speech). Replaces the need for stroke victims, but cannot reach deep centres such as the limbic system.
- DBS: hair-thin electrodes deep in the brain. Treats Parkinson's tremor and, by targeting Brodmann area 25, has given dramatic relief to some treatment-resistant depressives.
- Optogenetics: insert a light-sensitive gene into chosen neurons, then switch behaviour on and off with light. Flies take off, worms stop, mice run in circles; monkey trials were starting.
- Clarity (2013): Stanford's Kwanghun Chung ("I burned and melted more than a hundred brains") replaced brain lipids with hydrogel, washed them out with soap and an electric field, and made a mouse brain transparent.
Two physical limits recur. Magnetic stimulation fades with distance faster than the inverse-square law, so it silences cortex near the skull but cannot reach the limbic system; Kaku expects stronger, more focused coils to fix this. The tool kit also includes near-infrared spectroscopy (NIRS), which reads blood oxygen through the skull with light. And Clarity is general: Chung has made transparent livers, lungs and hearts, and the method works on organs preserved for years in formalin, which opens archived brain banks to wiring studies. The New York Times ran it on page one: "Brain as Clear as Jell-O for Scientists to Explore."
Tracer and naming errors. PET does not use sodium-22 in glucose (glucose contains no sodium). The standard tracer is fluorine-18 fluorodeoxyglucose (FDG). PET stands for positron emission tomography, not "topography". The field's name for Kaku's "TES transcranial electromagnetic scanner" is TMS, transcranial magnetic stimulation.
Four forces, so no radically new scanners
Before scanning, about 30 brain regions were known with confidence; MRI alone now distinguishes 200 to 300. Kaku predicts refinement, not revolution, because there are only four fundamental forces (gravity, electromagnetism, weak and strong nuclear) and electromagnetism, understood for 150 years, underlies nearly every scanner (PET is the exception, using the weak force). Expect machines to get smaller and cheaper, with physicists already sketching an MRI in a cell phone. The core challenge is resolution: a voxel today can hold hundreds of thousands of neurons; the holy grail is an MRI-like device that sees individual neurons and their connections in real time.
Models of the brain, and Kaku's corporation
Each technology era produced its own brain metaphor:
- Homunculus. Explains nothing: who runs the homunculus's brain?
- Clockwork. Fit the age of gears; Leonardo designed a mechanical man.
- Steam engine. A hydraulic model that may have shaped Freud's struggle of ego, id and superego, where excess pressure causes breakdown.
- Switchboard. Rows of operators plugging wires; silent on how the network is wired.
- Computer. Mind as software on "wetware". But the brain has no programming or operating system, and it works the other way round: a PC is fast but serial, while each neuron is slow but 100 billion work in parallel.
- Internet. Consciousness as emergent, which names the miracle without explaining it.
Kaku's verdict on each is a missing ingredient. Freud himself admitted his hydraulic model needed neuron-level study that would take another century. The computer model cannot explain how the brain does work that would need a computer the size of New York City, and it has no Windows, no Pentium chip, no programs. The Internet model "brushes all the complexity under the rug of chaos theory". Each holds a kernel of truth; Kaku prefers the corporation because it makes testable predictions about subconscious processing, emotion and competition.
- Most information is subconscious. The CEO (prefrontal cortex) sees only what demands attention; meeting a tiger, you need to know how to run, not how your toes feel.
- Emotions are rapid decisions made lower down. Reason takes seconds, so lower centres raise instant red flags (fear, anger) without permission from the top. Rita Carter: emotions are body-rooted survival mechanisms.
- Constant clamour for the CEO's attention. No single control room: sub-centres compete. Minsky's "society of minds"; Pinker's "maelstrom of events" where the winner is rationalised afterwards as a unified self.
- Final decisions are made at the top. The buck stops in the prefrontal cortex, which mediates conflicting reports.
- Information flows are hierarchical, like a pine tree of nested networks. Unlike a bureaucracy, the brain cannot waste energy: it runs on about 20 watts, near the limit before heat damages tissue, so evolution packs it with shortcuts.
Is reality real?
Much of what we "see" is the brain's invention to save energy:
- The blind spot. The optic nerve leaves a hole in each retina, which the brain papers over.
- The fovea and saccades. Only a tiny central patch is sharp; the eye darts constantly and the brain stitches a false sense of a uniformly clear scene.
- Colour. Cones detect only red, green and blue; yellow, brown and orange are mixes the brain computes, as on an old colour TV.
- Depth. Two flat retinal images merged, plus parallax for distance. A child who says "the moon is following me" is seeing the brain fail at extreme parallax.
- Invisible spectrum. Infrared, UV, X-rays and gamma rays are all around us and unseen.
The split-brain paradox
Here the corporate model strains, because the brain has two near-identical hemispheres, and one can be removed entirely. Roger Sperry (Caltech, Nobel 1981) treated severe epileptics by cutting the corpus callosum to stop seizure feedback loops between the halves. Patients seemed normal, yet careful testing revealed two wills in one skull.
- A man went to hug his wife with one arm while the other threw a right hook at her face.
- A woman picked a dress with one hand while the other grabbed a different outfit.
- A man lay awake fearing his rebellious hand would strangle him (clinicians call it the Dr. Strangelove syndrome).
- Michael Gazzaniga asked a patient's speaking left hemisphere about his career: "draftsman". The mute right hemisphere, answering with Scrabble letters, spelled "automobile racer".
- V. S. Ramachandran describes a patient whose left brain was an atheist and right brain a believer.
Sperry concluded each hemisphere is "a conscious system in its own right". Normally the verbal, analytical left brain dominates and sends commands across the callosum; cut it and the holistic right brain is freed from that "dictatorship". Kaku jokes that such a person could be a Republican and a Democrat at once, with chaos in the voting booth.
Testing a split brain takes ingenuity. Gazzaniga uses special glasses that show a question to one eye and therefore one hemisphere; asking is easy, but the right brain is mute because speech lives on the left, hence the Scrabble-letter answers. Rita Carter draws the unsettling moral that we may all carry "a mute prisoner" in our skulls with its own personality, ambition and self-awareness. Ramachandran pushes it further for his atheist-believer patient: if that man dies, does one hemisphere go to heaven and the other to hell? Kaku's point is structural: no corporation has two complete command centres, so the analogy needs amending, and the next chapter asks what a consciousness that can be split in half really is.
The split-brain findings are real, but the popular left-brain/right-brain personality split they spawned is a myth, as Kaku himself hints. Later work (Pinto and colleagues, 2017) found split-brain patients can often report stimuli from both visual fields, suggesting one unified but split perceptual stream in some cases. Whether there are truly "two selves" is still debated.
Who is in charge?
David Eagleman (Baylor) compares the conscious self to a young king who takes credit for a kingdom run by thousands of unseen staff. Choices of partner, friends and career are swayed by forces we never notice; he cites the finding that Dennis and Denise are over-represented among dentists, Laura and Lawrence among lawyers. Reality is a gap-filled approximation, and it differs between people: Eagleman says at least 15% of women carry a gene for a fourth colour receptor.
The "name-letter" career effect (Pelham and colleagues) has since been challenged as a statistical artefact (Simonsohn, 2011). And while many women carry a fourth cone gene, functional tetrachromacy that lets someone see extra colours has been confirmed in only a handful of people.
Since 2014. Optogenetics reached humans: in 2021 a blind man with retinitis pigmentosa partially recovered vision after optogenetic gene therapy plus light-amplifying goggles. Clarity spawned a family of tissue-clearing methods now standard in connectomics. MRI got both bigger and smaller: France's Iseult 11.7-tesla scanner produced its first human brain images in 2024, while low-field portable bedside MRI (Hyperfine) won FDA clearance in 2020. DBS of area 25 for depression had a setback when a large trial was halted in 2013 for futility, but newer personalised, closed-loop stimulation has shown striking single-patient results since 2021.
Kaku's corporation model is useful for anyone designing AI systems. A modern agentic AI stack looks much like it: many specialist sub-agents compete for an orchestrator's attention, most processing never reaches the top, fast guardrail checks fire before slow reasoning, and the "single answer" is a post-hoc narrative. The split-brain cases warn that an orchestrator can confidently confabulate a reason for what a sub-system did, which is exactly why explanations from an AI agent need independent logs.
The chapter closes with the questions the next one answers: if consciousness can be split in half, what is it, how does it relate to self-awareness, and could a robot or alien have it?
Consciousness: A Physicist's Viewpoint
Kaku borrows the physicist's method (build a model, give it parameters, simulate it forward) to define consciousness as building a model of the world through feedback loops in order to reach a goal. His space-time theory ranks consciousness in four levels, and its claim about humans is sharp: what makes us unique is that we model the world in time, using the past to simulate the future. Humour, gossip, play, intelligence and the sense of "I" all fall out of that definition.
Why consciousness resists definition
David Chalmers counts over 20,000 papers on consciousness, with little consensus. Leibniz said that if you blew the brain up to the size of a mill and walked inside, you would not find consciousness. Some, like Steven Pinker, suspect we lack the mental equipment to solve it, just as we cannot see ultraviolet. For much of the twentieth century behaviourism ignored inner states entirely. Giulio Tononi settles for description: consciousness is what leaves you in dreamless sleep and returns when you wake.
Two epigraphs set the mood: Joseph Conrad's "the mind of man is capable of anything... Because everything is in it, all the past as well as all the future", which anticipates Kaku's whole thesis, and Colin McGinn's warning that consciousness can reduce the most fastidious thinker to "blabbering incoherence". Pinker's full doubt is an argument from limits: "We cannot see ultraviolet light. We cannot mentally rotate an object in the fourth dimension. And perhaps we cannot solve conundrums like free will and sentience." Kaku's response is pragmatic: since physicists built the instruments that cracked open the brain, borrow the physicist's method too.
The physicist's method
Physics proceeds in three steps: collect data, build a model described by parameters (temperature, energy, time), then simulate it forward. Newton imagined throwing an apple ever faster until it circled the Earth like the Moon; the model only became powerful when he invented calculus to compute future positions of the Moon, Halley's Comet and the planets. Models are replaced when better parameters arrive: Einstein swapped forces for curved space-time, and now supercomputers simulate colliding black holes. Kaku applies the same template to the mind.
The model's worth lies in its parameters. For Newton the parameter was position in space and time; letting it evolve unlocked the motion of cannonballs, rockets, asteroids, stars and galaxies. Today the largest supercomputers simulate models of protons, nuclear explosions, weather, the big bang and black-hole interiors. Kaku's move is to ask which parameters a brain tracks and how far forward it runs them.
Kaku's definition. Consciousness is the process of creating a model of the world using multiple feedback loops in various parameters (temperature, space, time, relation to others) in order to accomplish a goal (food, mates, shelter). He calls it the space-time theory of consciousness: animals model mainly space and each other; humans also model time, backward and forward.
The four levels, with numbers
Each feedback loop counts as one unit of consciousness, so consciousness becomes a ranked hierarchy rather than a circular concept.
| Level | Parameter modelled | Brain structure | Examples | How it is scored |
|---|---|---|---|---|
| 0 | A few physical variables (temperature, light, gravity) | None | Thermostat (0:1), bacteria, flower with ten loops (0:10) | Count the feedback loops |
| I | Position in space | Brain stem, cerebellum ("reptilian brain") | Reptiles, with perhaps 100+ loops for smell, balance, sight, blood pressure | Count loops; sight alone has many (colour, motion, shape, shadow) |
| II | Place in relation to others (emotions, society) | Limbic system: hippocampus, amygdala, thalamus | Social mammals; a 10-wolf pack with 15 emotions and gestures each = II:150 | Group size × number of distinct social and emotional interactions |
| III | Place in time: simulating the future from the past | Expanded prefrontal cortex | Humans | Causal links a person can make when simulating scenarios, divided by a control-group average, × 100 (so the average adult is III:100) |
Insects are excluded from Level II: they are social but show no emotions as far as we know.
The mechanics behind the numbers are concrete. A thermostat's single loop can be a metal strip that expands when heated and trips a switch. A reptile's eyesight alone needs loops for colour, movement, shape, light intensity and shadow, which is why so many loops forced the evolution of a central nervous system. Level II loops are the stuff of primate politics, and Kaku lists them: recognising rivals and friends, forming bonds, reciprocating favours, building coalitions, knowing your own rank and others', deferring to superiors, displaying power over inferiors, and plotting to climb the ladder. He admits studies of animal social behaviour are too thin to count these properly, so his scores are first approximations that will change as data arrive. Kaku accepts grey areas (animals with rudimentary planning, cells that signal to each other) and treats solitary hunters among social species in his notes.
Level III: simulating the future
Kaku agrees with Darwin that the gap between humans and animals is one of degree, not kind, and locates the human difference in grasping tomorrow. Human consciousness "creates a model of the world and then simulates it in time, by evaluating the past to simulate the future." Daniel Gilbert calls the brain an "anticipation machine". Gazzaniga notes that Brodmann area 10 in the lateral prefrontal cortex is almost twice as large in humans as in apes; Kaku refers to the decision hub as the dorsolateral prefrontal cortex.
Animals, he argues, mostly act on instinct in the present: mammals hibernate by feedback loop rather than planning, predators plan only for the duration of a hunt, and primates show no planning beyond a few hours.
Gazzaniga's description of area 10 (internal granular layer IV) reads like a job specification for the CEO: memory and planning, cognitive flexibility, abstract thinking, initiating appropriate and inhibiting inappropriate behaviour, learning rules, and picking out relevant information from the senses. Humans, by contrast with animals, run simulations beyond their own life span, even thousands of years ahead, whenever they plan cathedrals, pensions or climate policy.
Why a CEO evolved
- Survival: foreseeing predators, food and mates.
- Choice: comparing several outcomes and picking the best.
- Overload: loops multiply exponentially from Level 0 to II, so instinct alone cannot referee them.
- Stickleback fish (via Eagleman). A female enters a male's territory; mating and defending loops fire together, so the male courts and attacks her at once.
- The shocked mouse. Put an electrode by some cheese; hunger and pain loops balance and the mouse oscillates back and forth, like the proverbial donkey starving between two bales of hay. A human CEO would weigh the options and decide.
The mechanism is a causal tree: if A then B; if B then C or D, and so on. Planning a bank robbery means evaluating hundreds of causal links (police, bystanders, alarms, accomplices, traffic, the DA). Counting the links a person can generate across standard scenarios gives the Level III score.
Testing the theory: humour, gossip, play, intelligence
Kaku insists a theory must be falsifiable: it fails if some pattern of thought will not fit. He then fits the hard cases.
- Humour. We silently simulate a joke's ending; laughter comes when the punch line breaks our simulation. Asked if he believed in clubs for young people, W. C. Fields said "Only when kindness fails." Timing matters because too early leaves no time to simulate and too late lets us simulate every ending. Humour also sizes up others' character (Level II).
- Gossip. Tracking shifting alliances is Level II; working out how a rumour affects your standing is Level III. For ancestors, gossip was the only news service, and lives depended on it.
- Play. Children's games (doctor, cops and robbers, school) are simplified simulations of adult society. Adults' poker and chess are pure future-modelling. Young animals play for exercise and pecking order, not simulation.
- Intelligence. IQ is circular ("how well you do on IQ tests") and culturally biased. Kaku redefines intelligence as the complexity of one's simulations of the future, which is why an illiterate master criminal can outwit the police.
Two details sharpen the humour argument. A deconstructed joke loses its force because we have already simulated the alternatives, and anticipating the unexpected had survival value in a jungle full of surprises, so a quick sense of humour is itself a marker of Level III intelligence. On gossip, Kaku's Martian visiting a supermarket checkout and seeing the wall of gossip magazines might conclude that gossip is humanity's main activity, and "would not be far off". On play, the question "what is fun?" gets a functional answer: fun is the reward for running safe rehearsals of adult life; card players model each opponent's hand from past data about their personality and ability to bluff.
The levels inside the human brain
Humans run all levels at once, and scans can trace each:
- Level I, the stream of consciousness. Walking in a park, signals pass from senses to spinal cord, brain stem and thalamus, which routes them to the sensory cortices (vision to occipital, touch to parietal) and finally to the prefrontal cortex, where we become aware of them.
- Level II, the cocktail party. Scanning for colleagues sets the hippocampus (memory), amygdala (emotion) and prefrontal cortex working together; each face gets an emotional tag. Spotting the rival who backstabbed you fires the amygdala, pumps adrenaline, raises the heart rate and readies fight or flight.
- Level III, the job hunt. At the same party a job-seeker simulates each conversation: how to impress, what to raise, whether this person can hire. The dorsolateral prefrontal cortex draws on memories; the pleasure centres (nucleus accumbens, hypothalamus) light up for good outcomes, the orbitofrontal cortex warns of bad ones, and the DLPFC adjudicates, a rough echo of Freud's ego, id and superego.
Theory of mind and mirror neurons
Level II depends on guessing what others think, the Theory of Mind first proposed by David Premack. Anyone who reads intentions well can form alliances and isolate enemies, and some anthropologists think mastering it drove brain evolution. A candidate mechanism arrived in 1996: mirror neurons, found by Giacomo Rizzolatti, Leonardo Fogassi and Vittorio Gallese, fire both when you act and when you watch someone else act (and for shared emotions). V. S. Ramachandran predicted they would "do for psychology what DNA did for biology". Kaku notes the dissent: these functions may be spread across many neurons rather than a dedicated class.
Mirror neurons were first recorded in the premotor cortex of monkeys and later reported in human prefrontal regions. Kaku ties them to mimicry and empathy, the glue of cooperation that held tribes together, and Ramachandran credits them with a role in self-awareness itself. Kaku's own caution is methodological: all such results must be retested and reconfirmed before they bear weight.
Ramachandran's prophecy has not come true. The mirror-neuron story was heavily oversold in the 2000s (Gregory Hickok's The Myth of Mirror Neurons, 2014, is the standard rebuttal), and links to empathy and autism have not held up well. Kaku's claim that animals do not plan beyond a few hours is also too strong: scrub jays cache food for tomorrow's breakfast and great apes select and save tools for use up to 14 hours later (Mulcahy and Call, 2006). The human advantage is in depth and range of planning, not its existence.
Self-awareness and the mirror test
Self-awareness is creating a model of the world and simulating the future in which you appear. Animals have a little (they must know where they are), but humans constantly star in their own simulations: the date, the interview, the career change. It is so hard to switch off that meditation exists largely to try.
The mirror test, going back to Darwin, is passed by elephants, great apes, bottlenose dolphins, orcas and European magpies; most animals ignore or attack the reflection. Daydreaming is Level III self-simulation: we put ourselves in the model and press "play".
Where am "I"?
Todd Heatherton (Dartmouth) locates the unifying sense of self in the medial prefrontal cortex; Carl Zimmer says it may do for the self what the hippocampus does for memory, "continually stitching together a sense of who we are". Kaku stresses this is not a homunculus. The prediction that the resting, daydreaming brain should be busy is borne out by scans: idle mind-wandering about ourselves and others lights it up.
The puzzle remains of how competing subunits produce a smooth "I". Gazzaniga's answer is the left-brain interpreter, which confabulates stories to cover gaps.
Heatherton's evidence is the resting brain: if the medial prefrontal cortex stitches the self together, it should be busy when we idle, and scans confirm it. "Most of the time we daydream," he says, about things that happened to us or about other people, "all this involves self-reflection." The interpreter supplies the other half: it is constantly asking "why?" and inventing answers even when none exists, which is how a patchwork of competing modules feels like one continuous "I".
Gazzaniga flashed "red" to a split-brain patient's left hemisphere and "banana" to the right. Asked to draw with his left hand (run by the right hemisphere), he drew a banana. Asked why, the speaking left hemisphere, which never saw the banana, should have said "I don't know". Instead it invented an excuse: "It is easiest to draw with this hand because this hand can pull down easier." The interpreter is "driven to hypothesise about the structure of the world even in the face of evidence that no pattern exists."
Kaku adds an evolutionary guess for having two hemispheres: a seasoned CEO has aides argue opposing sides (pessimistic versus optimistic, analytical versus holistic), and the truth emerges from the clash. When that interplay goes wrong, mental illness may follow, a theme he returns to in chapter 9.
Since 2014. The theory debate has hardened into rival camps. A large adversarial collaboration pitting Integrated Information Theory against Global Neuronal Workspace theory reported in 2023 and was published in Nature in 2025; neither theory's key predictions fully survived. In 2023 over 100 researchers publicly labelled IIT "pseudoscience", sparking a bitter row. Meanwhile predictive-processing accounts (Karl Friston, Anil Seth's Being You, 2021), in which the brain is a prediction machine that constantly models the world and itself, have become mainstream and are close in spirit to Kaku's space-time theory. Self-recognition has since been reported in a fish, the cleaner wrasse (2019, debated).
Kaku's numbers (0:1, II:150, III:100) are illustrative, not measured: no one has counted feedback loops in a wolf or causal links in a person. The theory is also silent on the hard problem, why any of this modelling feels like something. A thermostat counting as "conscious" at Level 0 shows the definition describes information processing, not experience, which is why philosophers rarely cite it.
For AI, Kaku's levels give a usable vocabulary. A rules engine is Level 0; a robot that maps its surroundings is Level I; a model that tracks users' emotions and intent approaches Level II; an agent that plans multi-step futures with a world model is reaching for Level III, minus the feelings. In an enterprise architecture review, "how many feedback loops, over which parameters, toward what goal?" is a crisp way to classify an AI system's autonomy.
With a working theory, Kaku turns to Book II and the first application: telepathy.
Mind over Matter
Book II turns the scanners outward: reading thoughts, moving objects by thought, recording and implanting memories, and boosting intelligence. The common engine is the brain-machine interface, where a computer learns a dictionary between brain signals and the outside world.
Telepathy: A Penny for Your Thoughts
Houdini said mind reading was a trick, and without help it is. But computers can learn the brain's electrical "dictionary": Berkeley turns fMRI into blurry movies of what you see, implanted grids decode words, and patients type by thought. Telepathy arrives not as a sixth sense but as decoding by machine, which also means privacy is now a question of law, not of physics.
Houdini's challenge and the physics of telepathy
Harry Houdini insisted every mind-reading feat was sleight of hand, toured exposing fraudulent mediums, and sat on a Scientific American committee offering a prize for proven psychic powers that nobody ever claimed. Kaku's twist: science is proving Houdini wrong, but only with outside assistance.
The physics is simple. Accelerating electrons radiate, so the brain's oscillating currents broadcast faint radio waves, too weak and jumbled for another brain to read, and evolution gave us no decoder. Computers can decode them. The method in every lab that follows is the same:
- Record brain signals (EEG, fMRI, ECoG, MEG) while the subject sees, hears or imagines known items.
- Build a dictionary that pairs each signal pattern with its item.
- Decode new signals by looking up the closest match.
- Refine with more subjects, more items and better mathematics. Accuracy is capped by the sensor: EEG is fast and non-invasive but blurred by the skull (it can tell car from house, not draw the car); ECoG sits on the brain and is sharp but needs surgery; fMRI is precise in space but slow.
IBM's "Next 5 in 5" forecast claimed people would soon control computers mentally, replacing mouse and voice; game firms, music studios, educators and the Pentagon were all converging on the field.
The first beneficiaries are the locked-in: stroke and accident victims who can communicate only by blinking. IBM's list of mental tasks shows the ambition: phoning people, paying credit-card bills, driving cars, making appointments, composing symphonies, all by thought. The chapter's epigraph from Steven Pinker states the premise plainly: the brain is a machine, a conclusion reached "not because [scientists] are mechanistic killjoys" but because every aspect of consciousness can be tied to it.
Videos of the mind: Gallant's lab
At UC Berkeley (Kaku's own PhD school), Jack Gallant's group reconstructs movies from brain scans: "We are opening a window into the movies in our mind." Subjects, usually the lab's own grad students and postdocs including Shinji Nishimoto, lie for hours in a $3 million MRI watching YouTube movie trailers.
- Resolution. The scanner distinguishes 200 to 300 regions with about 100 voxels per region; the visual cortex at the back glows like Christmas lights while subjects watch.
- Result. Kaku watched the output: faces, animals, streets and buildings seen "through dark glasses", with the kind of object clear but no detail.
- Imagination too. Imagined images also light the visual cortex. Asked to picture the Mona Lisa, the system's best match was a photo of Salma Hayek: crude, but chosen from millions.
- Numbers. Stanislas Dehaene (Collège de France) found digits so distinct in the parietal lobe that "200 voxels" suffice for a machine-learning decoder of which number is held in memory.
Thought-video will never be perfect, because imagined images carry less information than perceived ones: a scan of someone thinking of a flower shows far fewer active dots than a scan of someone looking at one. Kaku's parable is the genie who grants a car with no engine and banknotes with blurred faces, because memories are approximations.
Reading words
Next door, Brian Pasley's team (with Sara Szczepanski) records words with ECoG: a sheet of 64 electrodes in an 8×8 grid laid on the exposed brain of epilepsy patients during surgery at UCSF. Words heard produce signal patterns that form a dictionary; later, thinking the word evokes the same pattern, opening the door to silent conversation and voice synthesisers for paralysed patients.
- University of Utah, 2011: grids of 16 electrodes over the facial motor cortex and Wernicke's area; ten words (yes/no, hot/cold, hungry/thirsty, hello/good-bye, more/less) identified with 76% to 90% accuracy. Next step: 121-electrode grids.
- Mayo Clinic (Jerry Shih): ECoG typing by focusing on letters, accuracy "nearly 100%".
- Guger Technologies (Austria): over-the-counter EEG typewriter, learned in about 10 minutes, 5 to 10 words per minute.
- Gerwin Schalk (Albany Medical College): $6.3 million US Army grant; ECoG recognises vowels and 36 words, approaching 100%.
Shih's next goal is a machine that records images, not just letters, that patients conceive, with obvious uses for artists and architects. Every ECoG result shares the same catch: the skull must be opened, which is why the over-the-counter EEG typewriter, slower and less accurate, is the version reaching the market first.
- 10 words in the Utah vocabulary; 8×8 = 64 electrodes in the Berkeley grid; ~30,000 voxels in a Gallant scan.
- Next-generation grids: 121 electrodes (Utah); Gallant aims for more than 100 voxels per region.
Dictation, music and the language barrier
With a dictionary of a few hundred common words, people could think whole sentences onto a screen: a gift to writers and journalists, and to a mental robo-secretary that books trips from mental instructions. Musicians could hum silently and get printed notation. But Kaku doubts the science-fiction trope of universal thought: emotions may be nonverbal and transmissible to anyone, yet rational thought is bound to language, so telepathic words will still arrive in their original tongue.
Telepathy helmets for soldiers
The US Army wants helmets that carry orders through the din of battle. Kaku recalls machine-gun training at Fort Benning during the Vietnam era: the noise left his ears ringing for three days. Schalk's ECoG proves the principle but needs a sterile operating room. David Poeppel (NYU) uses MEG to record auditory-cortex activity when people silently think a word: non-invasive and fast, but the machine is table-sized.
Kaku describes MEG here as using "tiny bursts of magnetic energy... To create electrical charges in the brain". That is TMS. MEG is purely passive: it measures the brain's own magnetic fields, as Kaku correctly said in chapter 1.
MRI in a cell phone
MRI magnets are huge because resolution needs a uniform field. In 1993 Bernhard Blümich in Germany built a briefcase-sized MRI whose weak, distorted field is corrected by computer; in 2006 it scanned Ötzi the Iceman (frozen about 5,300 years ago with arms spread, too awkward for a conventional bore). With computing power doubling every two years, physicists foresee a phone-sized MRI that streams raw data to a supercomputer; Blümich mentions the Star Trek tricorder. Kaku sketches a future MRI telepathy helmet: coils in the helmet, a belt computer, a distant supercomputer, and the decoded message radioed back to troops through speakers or auditory-cortex electrodes.
The physics behind the shrinking magnet was settled by Maxwell in the 1860s: because the field's mathematics is known exactly, a computer can correct a weak, lumpy field, so computing power substitutes for magnet size. Kaku's consumer version: more computing in your medicine cabinet than a university hospital holds now, a portable MRI waved over yourself at home, and the data e-mailed to a lab for analysis.
DARPA and human enhancement
The main funder is DARPA, created by Eisenhower after Sputnik (1957), whose "only charter is radical innovation". Michael Goldblatt jokes it tries not to break the laws of physics, "or at least not more than one per program."
Its strategic plan adds that its sole justification is "to accelerate the future into being". Private companies, Kaku notes, have only recently shown interest, because brain research is a long gamble that may never pay off; the Pentagon carries the early risk. Goldblatt even rereads the Army slogan "Be All You Can Be" as a programme for enhancement.
- 1957 to 1958: Sputnik shock; DARPA founded; NASA among its early spinoffs.
- 1960s: ARPANET, which becomes the Internet after being opened up around 1989.
- Satellite positioning work that leads to GPS.
- Later: contributions to cell phones, night vision, weather satellites, telecom.
- 2010s: a $3 billion budget aimed increasingly at brain-machine interfaces.
Two DARPA asides: a former director told Kaku that sensors still cannot match a dog's nose, which evolved over millions of years to catch a handful of molecules; and DARPA engineers sighed that the public links them to Roswell and Area 51, adding that alien technology would certainly help.
- Michael Goldblatt: former head of DARPA's Defense Sciences Office. Imagines soldiers who communicate by thought, learning "as easy as eating" and body parts replaced like a drive-through order; believes DARPA's legacy will be human enhancement. His daughter has cerebral palsy and is his inspiration; Joel Garreau wrote that the technology might let her "not just walk, but transcend".
- Jack Gallant / Shinji Nishimoto: Berkeley thought-video.
- Brian Pasley / Sara Szczepanski: Berkeley word decoding by ECoG.
- Gerwin Schalk: Army-funded "silent speech".
DARPA history, mythologised. ARPANET was built for sharing scarce computers among researchers; the "survive World War III" story comes from Paul Baran's separate RAND work, and ARPANET was never secret, so it was not "declassified" by the NSF in 1989. "Project 57" was a 1957 nuclear-weapons safety test in Nevada, not the ancestor of GPS; GPS grew from the Navy's Transit satellites and the Air Force's 621B programme. The IBM "5 in 5" mind-control forecast (made in 2011) did not come true by 2016.
Privacy: physics protects you, for now
People fear covert mind reading. Nishimoto's reassurance is physics: magnetic signals fall off with the cube or fourth power of distance (double the distance from an MRI and the field drops eightfold or more, versus fourfold for gravity), and outside noise swamps them. Even in labs only a few letters, words or images can be pulled out at once. If remote reading ever became possible, a Faraday cage (Faraday, 1836, after an effect noticed by Benjamin Franklin) would block it: charge spreads over a metal shell leaving zero field inside, which is why lightning does not hurt aircraft passengers and why Kaku has stood unharmed in a cage under artificial lightning. A thin metal foil would do.
Nanoprobes
A neater route is a web of carbon-nanotube nanoprobes inserted by microsurgery: language probes in the left temporal lobe, vision in the thalamus and visual cortex, emotion in the amygdala, wired to a small computer and the Internet. Cables cannot be eavesdropped the way radio can, and you control when thoughts are sent. Some writers imagine probes implanted at birth (like children of the Borg in Star Trek), accepted in time as IVF babies were.
Legal issues
The real risk is not remote spying but records we create willingly being stolen. Pasley sees huge benefit for the disabled and great concern "if this were applied to people who didn't want that."
Kaku adds the human angle: because consciousness means constantly simulating the future, we all rehearse scenarios that stray into immoral or illegal territory, and we want those rehearsals private whether or not we act on them. History says the law lags new technology by years, and no law of physics settles these questions; judges and juries will. A defence lawyer will always be able to argue that a recorded thought was a random musing, not an intention.
- Intellectual property. If someone reads your invention from your mind, who owns it? Can thoughts be patented?
- Police and courts. John Perry Barlow: trusting government with privacy is "like asking a peeping tom to install your window blinds." Can police read a suspect's thoughts? Courts already rule on refusals to give DNA. And thinking about a crime is not committing it.
- Paralysed people's contracts. Is a will or contract dictated by brain scan legally valid when the technology is imperfect?
- Mental espionage. Corporate "safe rooms" may need Faraday walls.
From 1945 to 1952 a passive bug hidden in a carved Great Seal in the US embassy in Moscow leaked diplomats' conversations, including during the 1948 Berlin crisis and the Korean War. It had no power source: the Soviets energised it with a remote microwave beam, so sweeps missed it until a British engineer heard embassy talk on an open radio band. Every new form of radiation gets turned to espionage; Kaku expects brain waves to be no exception.
Key terms
| Term | Meaning |
|---|---|
| BMI | Brain-machine interface: a computer that turns brain signals into commands or text |
| EEG | Scalp electrodes; fast, cheap, non-invasive, spatially blurred |
| ECoG | Electrode grid laid on the exposed brain surface; sharp signals, needs surgery |
| fMRI voxel | 3-D pixel of blood-oxygen signal, a fraction of a millimetre across |
| MEG | Passive sensing of the brain's tiny magnetic fields; millisecond timing |
| Faraday cage | Conducting shell with zero field inside; blocks outside fields (or leaks) |
| Nanoprobe | Proposed carbon-nanotube electrode inserted by microsurgery |
Since 2014, telepathy-by-machine leapt ahead, powered by AI. In 2023 a UT Austin team (Alexander Huth) decoded the gist of continuous language from non-invasive fMRI using a GPT-style model; Nishimoto and Takagi reconstructed seen images from fMRI with Stable Diffusion. Implanted speech BCIs reached 62 words per minute (Stanford, 2023), 78 wpm with a talking avatar (UCSF, 2023) and about 97% accuracy for an ALS patient (UC Davis, 2024). In 2025 Stanford decoded inner speech and added a mental password to stop unintended leakage, precisely the privacy fix Kaku anticipated. Law moved too: Chile wrote neurorights into its constitution (2021), Colorado and California protected neural data (2024), and UNESCO adopted the first global Recommendation on the Ethics of Neurotechnology in November 2025.
The decoding pipeline here (record, build a dictionary, look up the nearest match) is the same supervised-learning loop behind every enterprise ML model, and it shares the same governance needs: consented training data, per-user calibration, accuracy that collapses outside the lab, and the risk of function creep from medical to commercial use. For a bank, neural data joining biometrics in consumer devices (EEG earbuds, headbands) is a coming privacy-classification question.
Reading thoughts is passive. The next chapter goes active: moving matter with the mind.
Telekinesis: Mind Controlling Matter
Telekinesis is real once a computer stands between brain and object. Kaku follows the BrainGate chip, monkeys moving robot arms and their own paralysed limbs, brain-to-brain links, exoskeletons and surrogate robots, then sets the physical limit: the mind supplies the command, but an external power source has to supply the force. The chapter ends with a warning from Forbidden Planet: a power that obeys every thought would also obey the monsters in the subconscious.
BrainGate: Cathy Hutchinson's drink
Cathy Hutchinson, paralysed by a brain-stem stroke fourteen years earlier, became in May 2012 the face of the field when a Brown University team let her drive a robotic arm by thought to lift a bottled drink to her mouth. Communicating by eye-tracking, she typed "Ecstatic!" and asked for a robotic leg next; before the stroke she loved cooking and gardening.
John Donoghue (Brown, with the University of Utah) explained the BrainGate sensor: "the size of a baby aspirin, or four millimeters", with 96 hair-like electrodes set on the well-mapped motor cortex where arm intentions arise. The patient imagines moving a cursor right; a few minutes of recording teaches the computer that pattern; control comes almost on the first try.
- 2004: first BrainGate trials; patients use laptops, surf the web, write e-mail, steer wheelchairs.
- 2011: University of Utah word-decoding grids (chapter 3).
- 2012: Hutchinson drinks using a robot arm; Stephen Hawking tests an EEG-like device on his glasses.
- 2012: EPFL (Switzerland) patients steer a Roomba-like robot 60 miles away by EEG.
- 2013: first rat-to-rat (Duke to Brazil) and human-to-human (University of Washington) brain links.
- 2014 target: Nicolelis's exoskeleton kick at the World Cup opening in Brazil.
Donoghue's point is that once thought drives a computer, it drives anything a computer can switch: toasters, coffee makers, air conditioners, lights, TV channels, so that a quadriplegic need not wait for someone to enter the room.
Kaku frames Hutchinson as one of thousands of "locked-in" patients, clear-minded "prisoners in their own bodies" who need continual nursing. The chapter's epigraph from Alfred North Whitehead ("it is the business of the future to be dangerous") warns that the same power will cut both ways. Even Stephen Hawking had a primitive EEG-style neuroprosthetic mounted on his glasses; Kaku expects such devices to gain channels and sensitivity until patients "can do anything a normal person can do, via computers".
Bypassing a broken spinal cord
Christopher Reeve, film's Superman, was paralysed in 1995 when thrown from a horse onto his neck. More than 200,000 Americans have spinal cord injuries, a number growing because trauma care saves more lives (including soldiers hit by roadside bombs in Iraq and Afghanistan); with stroke and ALS the total reaches about two million.
At Northwestern University, Lee Miller's team put a 100-electrode chip on a monkey's brain, decoded the signals for grasping a ball, lifting it and dropping it into a tube, and fed them directly to the nerves of the monkey's own arm, skipping the spinal cord. The monkey relearned coordination the way we learn a new mouse or tennis racquet.
100 electrodes can stand in for the millions of neurons that move an arm because, as Miller explains, the chip taps the output neurons, after the brain has done the heavy computation. A brain-machine interface needs to read the brain's decisions, not its whole deliberation. The same logic explains why small implants work at all.
Northwestern also built a shoulder-driven prosthesis: shrug up to close the hand, shrug down to open it, enough to grip a cup or turn a key.
Miller's own description is precise: "We are eavesdropping on the natural electrical signals from the brain that tell the arm and hand how to move, and sending those signals directly to the muscles." He sees it as a route to daily-living independence for people with spinal injuries, not as a laboratory curiosity.
Revolutionizing Prosthetics
DARPA's Revolutionizing Prosthetics programme ($150 million, from 2006) was pushed by retired Army colonel and neurologist Geoffrey Ling, appalled by roadside-bomb injuries: more than 1,300 US service members returned from the Middle East with lost limbs. He demanded solutions in five years ("They thought we were crazy. But it's in insanity that things happen"). It funded the Johns Hopkins Applied Physics Laboratory arm, matching a human arm in size, strength and almost every finger, hand and arm motion.
A quadriplegic woman paralysed from the neck down by a genetic disease had electrodes placed on her brain at the University of Pittsburgh, linked to the APL arm. Five months after surgery she appeared on 60 Minutes, waving, shaking the host's hand and giving him a fist bump. Ling's dream: extend it to stroke, cerebral palsy and the elderly.
Ling's reasoning came from battlefield medicine: in earlier wars these soldiers would have died, but helicopter evacuation now saves them with grievous injuries, so the demand for replacement limbs is a by-product of medical success. The APL arm is steel, yet covered in flesh-coloured plastic it would be nearly indistinguishable from a real one.
Better trauma care creates the market for neuroprosthetics. Spinal-injury, stroke and blast survivors who once died now live for decades, which is why the Pentagon, not industry, paid for the first generation of mind-driven limbs.
The patient's name is Jan Scheuermann (Kaku writes "Jan Sherman"). Two further slips: NeuroSky's MindWave Mobile headset cost about $129, not "$129 million"; and Hutchinson's stroke dated from 1996, about fifteen years before the 2012 trial.
Telekinesis in daily life
The youth market got there first. In 2009 NeuroSky's Mindflex toy let EEG-measured concentration speed a fan that pushed a ball through a maze. About 1,700 developers built games for NeuroSky's MindWave headset, where you move an avatar, fire weapons and dodge enemies hands-free. Alvaro Fernandez (SharpBrains) cast NeuroSky as the potential "Intel" of the industry. Attention monitoring could alert drowsy drivers and machine operators, and in Japan EEG cat ears that perk up with interest became a party fad for signalling romantic attraction.
Nicolelis: smart hands and mind melds
Miguel Nicolelis (Duke) is the chapter's main guide.
- Across continents. A monkey walking on a treadmill in North Carolina made a humanoid robot, CB-1, walk in Kyoto, rewarded with food pellets.
- Feeling through a prosthesis. Prosthetic hands lack touch, so they feel foreign and could crush a hand or an eggshell. Nicolelis built a brain-machine-brain interface (BMBI): motor cortex drives an arm whose sensors send coded signals back to the somatosensory cortex. Monkeys learned it in 4 to 9 trials, and after a month their brains read the artificial code as rough or smooth texture: "the first demonstration that we can create a sensory channel".
- Holodeck. Kaku links this to Star Trek's holodeck via haptic technology: Internet contact lenses supply the virtual sight, BMBI signals make virtual objects feel solid. Nicolelis agreed it was a first step.
- Brain-to-brain (2013). Rats at Duke learned to press a lever at a red light; rats in Natal, Brazil learned to press when an implant stimulated them. With motor cortices linked over the Internet, the Brazilian rats responded correctly to the American rats' signals in seven of ten trials.
-
Human-to-human (2013). At the University of Washington, one scientist in an EEG cap imagined moving his right hand to fire a video-game cannon; over the Internet a TMS coil over a colleague's motor cortex made the second man's right hand move involuntarily.
-
What it could carry. Kaku imagines e-mailing friends the experience of dancing the tango, bungee jumping or skydiving, then emotions and feelings. He notes the limits honestly: the rat link was primitive with a small sample, far from the merging of two minds in a Star Trek mind meld, but it is a proof of principle for a brain-net.
The brain-net and civilisation
Nicolelis's dream is social networks run through minds, carrying emotions, nuance and reservation that a phone call or video cannot. Building it takes stages: nanoprobes in key areas (left temporal lobe for speech, occipital lobe for vision), computer decoding, fibre-optic transmission, and the hard part, writing signals back into another brain, so far achieved only for the hippocampus (chapter 5). He imagines billions of people making "temporary direct contacts" by thought, a "colossus of collective consciousness".
- Body language and grunts: nomadic bands.
- Spoken language: symbols and complex ideas, enabling villages and cities.
- Writing: knowledge accumulates across generations, enabling science, arts and empires.
- Telephone, radio, TV: communication across continents.
- Internet: an emerging planetary civilisation.
- Brain-net: senses, emotions, memories and thoughts exchanged globally.
Entertainment would change too: as the 1920s talkies added sound to film, future cinema could record all five senses and emotions from actors wired with sensors, with editors splicing sensation tracks and audiences wearing brain sensors instead of 3-D glasses; some actors may not survive the transition, as some silent stars did not.
The audience would smell the heroine's perfume, feel the terror of a horror film's victims and relish the villain's defeat; actors would have to perform each scene in all five senses, and cinemas would need retrofitting to process and transmit the sensation data.
Nicolelis traces his calling to watching Apollo in Brazil and to Asimov's 1964 book The Human Brain, which disappointed him by saying nothing about how the parts make a mind. About ten years before, his team let a rodent move a robotic lever to bring water to its mouth by brain signals alone, the first machine operated without moving the body. He now records 1,000 neurons in monkeys (up from 50) and even has a monkey driving a virtual avatar. "We will be part of their operating system."
Exoskeletons: the Walk Again Project
Nicolelis's "wearable robot", a motorised suit, aims to let the paralysed "walk by thinking", wirelessly, reading 20,000 to 30,000 neurons. The target was a quadriplegic Brazilian making the opening kick of the 2014 World Cup, "our Brazilian moon shot".
- Durable implants that sit safely in the brain for years.
- Wireless links to a belt computer the size of a phone, so the wearer can roam.
- Better decoding: hundreds of neurons moved a monkey arm; a human limb needs at least several thousand.
- Portable power strong enough to drive the whole suit.
Avatars and surrogates
Kaku uses the films Surrogates (people living in pods behind perfect robot bodies) and Avatar (miners on Pandora driving engineered Na'vi bodies to reach the metal unobtanium) to argue that surrogates will become tools for dangerous work.
- Space. Radiation, solar flares, micrometeorites and weightlessness ravage bodies: a transatlantic flight gives about 1 millirem per hour, a dental X-ray's worth; space is far worse; months of weightlessness strip calcium and Russian cosmonauts after a year crawled out "like worms". Proposed lunar bases sit in lava tubes; safer still is an astronaut on Earth steering a lunar surrogate with a one-second delay. Mars, at twenty minutes or more, is harder. Perhaps "One small step for the mind".
- Disasters. Fukushima (2011) cleanup may take 40 years because robots were little more than cameras on wheels. At Chernobyl (1986) helicopters dropped 5,000 tons of borated sand and cement and 250,000 workers took turns of a few minutes each in lethal fields. The Deepwater Horizon spill (2010, about 5 million barrels) ran three months while clumsy remote subs failed to cap the well.
- Current mind-driven robots. Honda's ASIMO responds to an EEG cap but only with four head-and-shoulder motions. Rajesh Rao's two-foot Morpheus robot at the University of Washington learns a user's EEG in ten minutes and can fetch a block and carry it six feet to another table. EPFL's 2012 camera robot let a paralysed patient steer through an office 60 miles away.
-
Inside the body. Fantastic Voyage shrank a submarine, which violates quantum physics, but MEMS machines with gears and motors smaller than a full stop could one day be steered by a telepathy helmet to operate from inside, or guide nanoprobes to exactly the right neurons.
-
Why surrogates save money. Life support is the most expensive part of human spaceflight; a surrogate on the Moon needs none. During solar storms NASA already orders space-station crews into better-shielded modules, and some bone and muscle loss from long weightlessness appears to be permanent.
- Chernobyl in detail. Gorbachev ordered the air force to "sandbag" the reactor; many of the 250,000 workers received the maximum lifetime dose in minutes, and each got a medal. It was the largest civil engineering effort ever, and no robot of today could have done it.
- Nicolelis's vision. "Remotely controlled envoys and ambassadors, robots and airships of many shapes and sizes", sent to explore other planets and stars on our behalf. MEMS are built with the same Silicon Valley etching that puts hundreds of millions of transistors on a fingernail-sized wafer, so surgical micro-submarines ride the same cost curve as chips.
The future of telekinesis, and its limits
Short term: the paralysed regain communication, mobility and appliances. By mid-century Kaku expects mental interaction with computers to be common, mice and keyboards to fade, and hidden chips to take silent mental commands for finances or theatre tickets. Artists could project imagined images onto 3-D holographic screens and iterate; engineers could sketch bridges by imagination and send parts straight to 3-D printers.
The economic stake is large: computing is a multitrillion-dollar industry that mints billionaires overnight, so the mind-computer interface "will reverberate on Wall Street, and also in your living room".
Comic-book telekinesis
Magneto moves the Golden Gate Bridge; Phoenix lifts jets and incinerates stars; the Jedi summon lightsabers. The mind itself is the source of force.
Physically possible telekinesis
The body averages about one-fifth of a horsepower. None of the four forces moves a pencil without an external source, and magnetism moves only magnetic objects. The mind commands a power source, like the fake "Apollo" in Star Trek whose powers came from a hidden generator. A construction worker becomes a conductor of cranes and bulldozers.
Programmable matter
The closest thing to shaping matter by will is programmable matter, researched by Intel with Carnegie Mellon: microscopic chips called catoms that change their surface charge on wireless command to bind into one object (a phone) and then regroup as another (a laptop). Kaku saw pinpoint-sized catoms in a CMU clean room. He expects prototypes by mid-century and, perhaps by 2100, mental control of the orchestrating computer, so that redecorating or buying presents could be done from a catalogue by thought.
A morality tale: monsters from the id
Science is a double-edged sword. Surrogate soldiers may protect their operators while making war more careless of civilians. Stephen King's Carrie shows telekinesis too strong to control, consuming its user.
Kaku pushes the comic-book cases to their limits. Phoenix of the X-Men can hold back floods and raise jets, then, consumed by the dark side of her power, incinerates solar systems and destroys herself. The Jedi trick of summoning a lightsaber is already possible with a mentally triggered magnet, just as an MRI can hurl a hammer across a room. But ordinary levitation of a pencil or a coffee mug remains beyond science, because no force acts on plastic, water or wood without a power source and a handle. Carrie's story ends the same way as Phoenix's: humiliated at the prom, she destroys her tormentors and dies in the fire she started.
Modelled on The Tempest, the film finds a professor and his daughter on a planet once ruled by the Krell, a civilisation millions of years ahead of ours that built a machine to materialise anything they wished. They vanished the night they switched it on. The dying crewman's explanation: "Monsters from the id." Asleep, the Krell's buried animal urges and revenge fantasies became real and destroyed them overnight. Kaku's moral: a power perfectly obedient to thought obeys the subconscious too.
Since 2014. The 2014 World Cup kick did happen (Juliano Pinto, in an EEG-driven exoskeleton) but lasted seconds and was barely shown on the broadcast, drawing criticism as hype. The real leaps came later: in 2023 EPFL's brain-spine interface let Gert-Jan Oskam, paralysed for years, walk naturally by thought through a wireless implant bridging his spinal injury. Neuralink implanted its first human (Noland Arbaugh) in January 2024, who played chess and video games by thought, and has since implanted more patients; Synchron threads a stent-electrode through a blood vessel, avoiding open-skull surgery. BMI patients have controlled tablets, robotic arms and even text chat with AI assistants. Human brain-to-brain links grew to three-person "BrainNet" Tetris games (University of Washington, 2019). Fukushima's first melted-fuel debris was retrieved by a remote telescopic device only in November 2024. Intel and CMU's claytronics catoms never reached product.
Kaku's design rule, intent from the user, energy from a power source, is the right mental model for agentic AI in a bank: a human states the goal; tools with their own authority (payments rails, CRM writes) supply the "force". The Forbidden Planet lesson maps onto least-privilege design: an agent that executes every inferred wish, including half-formed ones, needs explicit confirmation gates and a kill switch.
From extending the mind outward, Kaku turns inward: rewriting memory and intelligence, which would change who we are.
The distinction matters to him. Telepathy and telekinesis extend us but leave us the same people. Altering memory and intelligence through genetics, electromagnetics and drugs changes the essence of who we are: without memories we are "cast adrift in an aimless sea of pointless stimuli", and if we cannot tell real memories from implanted ones, the question becomes "then who are we?" Science, he says, is moving from observing the mind to orchestrating it.
Memories and Thoughts Made to Order
The Matrix and Total Recall imagine downloaded skills and fake pasts. Kaku shows how far the real science had come by 2013: because every long-term memory funnels through the hippocampus, a memory recorded there can be played back, and USC/Wake Forest did exactly that in rodents. He then surveys the genetics of memory, smart pills, forgetting drugs, Alzheimer's, and the legal and social shock of recordable experience.
The two films frame the chapter's hopes and fears. In The Matrix, Neo has an electrode jacked into the back of his neck and downloads martial-arts mastery in seconds, then takes down the Sentinels. In Total Recall, Schwarzenegger's law-abiding past turns out to be manufactured, and he discovers he is the leader of the people he has been fighting. Kaku's position: impossible today, but a few decades out, artificial memories may indeed be inserted. The anonymous epigraph keeps him humble: if our brains were simple enough to be understood, we would not be smart enough to understand them.
How we remember: the lesson of HM
Henry Molaison ("HM") had seizures after a childhood head injury. In 1953, aged 27, surgeons removed much of his hippocampus on both sides. The seizures eased, but he could no longer form new long-term memories: he greeted the same people as strangers several times a day, and was horrified by the old man in the mirror, then forgot the horror. Old memories survived. Kaku says HM had fallen from Level III to Level II: unable to recall the recent past, he could not simulate the future either. He died in 2008, aged 82.
Kaku's image for HM is Groundhog Day, with a difference: Bill Murray's character remembered each repetition, while HM relived the same day without knowing he had lived it before; everything new lasted only a few minutes. His case located the hippocampus as the gate between short-term and long-term memory, the anatomical fact on which every later experiment in this chapter depends.
- Senses feed the brain stem and thalamus (relay).
- Sensory cortices process the signals.
- Prefrontal cortex: awareness and short-term memory (seconds to minutes).
- Hippocampus: processes and holds the memory for up to about a day, then breaks it into categories.
- Scattered storage: emotions to the amygdala, words to the temporal lobe, colour and vision to the occipital lobe, touch and movement to the parietal lobe. More than 20 categories are known, including fruits and vegetables, plants, animals, body parts, colours, numbers, letters, nouns, verbs, proper names, faces, facial expressions, emotions and sounds.
Scattering memory is efficient (far cheaper than recording sequentially like tape) and future digital storage may copy it. But it creates the binding problem: how does one cue, say the smell of cut grass, reassemble the scattered fragments of a walk in the park? Antonio Damasio found stroke patients who lose one category only, because the stroke hit where it was stored. Categories are also personal: a wine taster files tastes a physicist never notices. One proposed answer: brain-wide 40 Hz oscillations link fragments in time rather than space, so remembering may partly be resonance among regions. Harvard's Stephen Kosslyn credits the new picture to computers plus brain scans.
Recording a memory: the artificial hippocampus
In 2011 Wake Forest and USC recorded a rodent memory and stored it digitally. Team leader Theodore Berger: "If you can't do it with the hippocampus, you can't do it anywhere."
The CA3 to CA1 signals looked patternless until monitored millions of times. After a chemical wiped the task, playback restored it: "Turn the switch on, the animal has the memory; turn it off and they don't." The Navy sponsor's Joel Davis: implantables to enhance competency are "only a matter of time."
In 2013 MIT used optogenetics to tag the neurons holding a fear memory in a mouse and reactivate them by light in a harmless setting, producing fear of something that never happened: the first false memory.
Kaku draws two applications straight away: educators might implant new skills to retrain workers, and Hollywood might gain an entirely new form of entertainment. Recording, he stresses, captures a memory already processed; the shortcut works because the hippocampus carries pre-digested signals, which is also why it is the natural first target.
Corrections. The 2011 Berger/Deadwyler experiment used rats, not mice. In the MIT study (Steve Ramirez and Xu Liu in Susumu Tonegawa's lab) the trick was subtler than Kaku's summary: the memory of a safe room A was switched on by light while the mouse was shocked in room B, so it later feared room A, where it had never been hurt. Kaku also writes that HM's surgeons "mistakenly" removed the hippocampus; the removal was deliberate, its devastating effect on memory was not foreseen.
What remains hard
The artificial hippocampus could hold one memory at a time; plans were to scale up, reach monkeys and go wireless, aiming at stroke, dementia and Alzheimer's. The hippocampus is still a black box: you can record and replay a memory already formed, not build one from scratch.
- Map the hippocampus: electrodes across its four divisions CA1 to CA4 chart normal information flow.
- Record the flow while a subject learns a task (jumping through a hoop) and build a memory dictionary.
- Upload the recording to another subject's hippocampus to see if the skill transfers, and build a library. Then: people paid to live memories (a luxury holiday, a battle) wearing nanoelectrodes, and buyers with matching electrodes replaying them.
The recording electrodes must be small enough not to disturb the memory being formed, and the data would travel wirelessly to a computer; primate and human hippocampi are larger and more complex than a rodent's, so Kaku expects decades before human memories can be handled this way.
Muscle memory is a complication: walking and martial arts are stored partly in the motor cortex, cerebellum and basal ganglia, not only the hippocampus. And The Matrix plugs into raw sensory input, far harder than tapping the hippocampus's processed output.
- 130 million rods and cones in the retina record about 100 million bits at any moment.
- The optic nerve carries about 9 million bits per second to the thalamus.
- V1 paints a distorted copy of the scene on the back of the brain, with the fovea taking most of the space.
- V2 stereo vision; V3 distance; V4 colour; V5 motion (straight, spiral, expanding). Areas run V1 to V8, with 30+ visual circuits known.
-
Then prefrontal cortex for awareness, and the hippocampus stores it for up to 24 hours before scattering it.
-
Vision alone needs billions of neurons firing in sequence and millions of bits per second, and every memory adds four more senses plus emotion. No machine yet matches this, which is the scale an artificial human hippocampus must handle.
Remembering the future
James McGaugh (UC Irvine): "The purpose of memory is to predict the future." Scans at Washington University in St. Louis show that recalling the past and imagining the future use the same circuits, especially the dorsolateral prefrontal cortex to hippocampus link, which explains why amnesiacs like HM cannot picture tomorrow. Kathleen McDermott calls it "mental time travel". Kaku reads this as strong support for his Level III theory: long memory evolved because simulating the future pays; animals gain little from a detailed past.
Artificial cortex and cerebellum
- Artificial cortex (2012). The Wake Forest/USC team recorded signals between cortical layers L2/3 and L5 in five rhesus monkeys learning to pick out previously seen pictures. Baseline accuracy 75%; feeding the recorded pattern back raised it 10%; drugs cut it 20%, but playback pushed it above normal. Sam Deadwyler: the device could bypass a damaged area like a crutch.
- Artificial cerebellum. At Tel Aviv University, a chip recorded brain-stem signals during eyeblink conditioning (air puff plus tone), processed them and sent them back at another point; the rats blinked. It was the first time signals were taken from one part of the brain, processed and uploaded into a different part. Francesco Sepulveda (Essex) bets the hippocampus and visual cortex will have synthetic counterparts before 2100.
Alzheimer's: destroyer of memory
- 5.3 million Americans with Alzheimer's, expected to quadruple by 2050.
- 5% of people aged 65 to 74; over 50% of those over 85.
- US life expectancy in 1900 was 49, so the disease was rare; the over-80s are now a fast-growing group.
The disease first shrinks the hippocampus and thins its wiring to the prefrontal cortex, so recent events vanish while decades-old memories survive, until even spouses and children go unrecognised. Kaku reports a 2012 finding that the process starts with tau and accelerates beta-amyloid, the gluey plaque, and that the misfolded proteins behave like prions, each converting healthy proteins in a chain reaction. Hopes rested on antibodies or a vaccine, an artificial hippocampus, or genetic boosts to memory.
Kaku's biochemistry is garbled: there is no "tau amyloid protein". Alzheimer's involves beta-amyloid plaques and tau tangles; the classic amyloid cascade hypothesis puts amyloid first and tau downstream, not the reverse. "Prion-like" spread of misfolded tau and amyloid is a serious hypothesis, but these are not infectious prions.
The genetics of memory
In 1999 Joe Tsien (Princeton, MIT, Washington University) gave mice extra copies of NR2B, a switch for associating events in the hippocampus. Knock it out and learning fails; boost it and mice swim straight to a hidden platform in a water maze that normal mice forget. They were nicknamed Doogie mice. In 2009 Hobbie-J remembered novel facts three times longer than the previous champion ("like taking Michael Jordan and making him a super Michael Jordan"), though it still forgot a chocolate route after five minutes: "They are rats, after all." Some smart strains were unusually timid, perhaps because they remembered every hurt.
CREB genes add a second lever: the CREB activator builds new connections; the CREB repressor blocks new memories.
Tsien's NR2B work had been confirmed in other labs, and the next targets were dogs, which share many of our genes, then perhaps humans. Hobbie-J was named after a Chinese cartoon character; Tsien called NR2B "a universal switch for memory formation", and graduate student Deheng Wang supplied the Michael Jordan line. The CREB effect is old in evolutionary terms: more CREB protein means faster long-term memory in sea molluscs, fruit flies and mice alike.
- Flies. Jerry Yin and Timothy Tully (Cold Spring Harbor) found flies normally need ten trials; with extra repressor they never learn; with extra activator they learn in one session, a "photographic memory".
-
Mice. Alcino Silva found mice with defective activator were amnesiac, yet could learn a little with short lessons and rests.
-
The photographic fly. Tully compares the activator flies to students who "could read a chapter of a book once, see it in their mind, and tell you that the answer is in paragraph three of page two seventy-four". His study advice follows from the biology: review material periodically through the day until it moves into long-term memory.
- Why cramming fails. A fixed supply of CREB activator gets used up; spaced review lets it replenish. Tully: "We can now give you a biological reason why cramming doesn't work."
- Why emotional memories stick. Strong emotion lowers the repressor or raises the activator, so the filter lets them through.
Kaku lists three brakes on human brain boosting: mouse results often don't translate; better memory may not mean better general intelligence; and gene therapy is hard (viral vectors trigger antibodies, and a University of Pennsylvania patient died in a trial). If genes fail, deliver the proteins in a smart pill: drugs like MEM 1003 and MEM 1414 raise CREB protein, and aged animals on MEM 1414 formed long-term memories faster.
Kaku pictures the eventual gene route as a simple shot in the arm: a harmless virus carries the "smart gene" into cells, which then release proteins that act on the hippocampus. The pill route skips the gene and delivers the protein effect directly. Either way, he expects the beneficiaries to widen from the aged and Alzheimer's patients to the average person, which is where the ethical arguments begin.
Can memories be erased?
Films (The Bourne Identity, Eternal Sunshine of the Spotless Mind, Men in Black's neuralizer with its time dial) assume memories are stored sequentially and can be wiped from a timestamp. Real amnesia comes in two kinds:
Retrograde amnesia
Trauma destroys existing memories, usually back from the injury; the hippocampus works, so new memories form (Jason Bourne).
Anterograde amnesia
Damage to short-term processing prevents new memories forming after the event (HM; the hero of Memento, who tattoos clues because his memory lasts 15 minutes).
Because memories are scattered, precise Hollywood erasure is highly improbable. Drugs may do something gentler:
- Propranolol (2009). Merel Kindt's Dutch group found the old beta-blocker eased the pain of traumatic memories within days. It works because recalling a memory means reassembling it, and propranolol blocks adrenaline, which makes trauma vivid (McGaugh: "adrenaline can be present, but it can't do its job").
-
The ethics fight. The President's Council on Bioethics objected that dulling terrible memories could make us "unmoved by suffering, wrongdoing, or cruelty"; Stanford's David Magnus says painful breakups make us better. Roger Pitman (Harvard) replies that we don't deny morphine to accident victims to preserve their full experience.
-
The council's sharpest line. "Can we become numb to life's sharpest sorrows without also becoming numb to its greatest joys?" Pitman's rejoinder exposes the hidden assumption: behind the objection "lurks the notion that mental disorders are not the same as physical disorders". The stakes are large because the candidates include war veterans, survivors of sexual abuse and accident victims with PTSD.
- Why recall is editable. Long-term memories are encoded in protein structures, not as standing electrical signals, which made a drug effect puzzling until experiments showed recall involves retrieval and then reassembly: "recalling a memory actually changes it".
- True erasers (2008). Tsien's group (Medical College of Georgia with Shanghai colleagues) erased a mouse memory via CaMKII; SUNY Downstate found PKMzeta could do it, and André Fenton suggested uses in depression, anxiety, phobias, PTSD and addiction.
Forgetting is the most beneficial process we possess.
Bernard Williams, Oxford philosopherWhat can go wrong: the Brainstorm problem
In the 1983 film Brainstorm, "the Hat" records a whole experience. A looped recording of sex nearly kills a colleague by sensory overload; a scientist records her own fatal heart attack and the replay kills the next listener; the military moves to seize the device. Kaku dates the realistic timeline: varied memory recording in primates and humans by mid-century; a Hat that taps raw sensory data in the thalamus perhaps late this century.
Social and legal consequences
- Education. Upload calculus; teachers become mentors for the parts that can't be uploaded; less rote learning for doctors and lawyers.
- Fake lives. Memories of holidays never taken and lovers never had; ethicists fear people will prefer imaginary lives.
- Jobs. Displaced workers (like yesterday's blacksmiths and wagon makers) could gain new skills instantly, saving human capital even if each skill's market value falls.
- Tourism. Upload enough language and local know-how for a decent conversation.
- Social media of experience. Olympians could post the agony and ecstasy of victory; children would record memories as casually as phone photos, via nanowires to the hippocampus.
-
A library of souls. Families could keep memory genealogies and "chat" with ancestors; you could live through a Nobel laureate's discovery or a statesman's crisis. Nicolelis imagines each record "a uniquely precious jewel".
-
From paper trail to mind trail. Ancestors usually leave only birth and death dates; we leave credit-card receipts, bills, e-mails and bank statements, which record what we did but not what we thought or felt. A memory web would chronicle consciousness itself, letting descendants see how they fit into the family story.
- Dark side. New medicine always provokes alarm: thalidomide was rightly banned; IVF (Louise Brown, 1978, condemned by the Pope) became normal. But unconsented, painful or destructive uploads, consent for Alzheimer's patients, implanted false eyewitness memories, fake alibis and untrustworthy affidavits would all need law: access rules like those protecting the home, and tags marking implanted memories as fake.
Key terms
| Term | Meaning |
|---|---|
| Hippocampus (CA1 to CA4) | Gateway that turns short-term into long-term memory; its four subfields |
| Binding problem | How scattered memory fragments are reassembled into one recollection |
| Engram | The physical trace of a specific memory in a set of neurons |
| NR2B | Receptor subunit that boosts associative learning (Doogie mice) |
| CREB activator / repressor | Genes that promote or block long-term memory formation |
| Reconsolidation | Recalling a memory makes it editable before it is re-stored |
| Retrograde / anterograde amnesia | Loss of old memories / inability to form new ones |
Since 2014. Berger's team tested a hippocampal prosthesis in humans: in 2018 epilepsy patients with implanted electrodes improved short-term memory recall by about 35% when their own successful patterns were played back. Engram science (Tonegawa and others) has since reactivated, suppressed and even "rescued" lost memories in mouse models of Alzheimer's. PKMzeta proved complicated: knockout mice still remembered (2013), though later work found a backup enzyme and a partner (KIBRA) that rescued the theory (2024). For Alzheimer's, anti-amyloid antibodies finally reached patients: lecanemab (FDA full approval 2023) and donanemab (2024) slow decline by roughly a quarter to a third, with risks of brain swelling and bleeding; the first blood tests for Alzheimer's pathology were cleared in 2025.
For AI, the hippocampus is the best biological analogy for retrieval-augmented memory: a gateway that indexes and routes fragments rather than storing them whole, consolidation that turns short-term context into long-term stores, and reconsolidation, where every recall slightly rewrites the record. The same governance points Kaku raises for implanted memories apply to agent memory in a bank: provenance tags marking what is synthetic, consent for what is stored, and a right to be forgotten.
Recording memories gives you more facts; it does not make you better at using them. For that, Kaku turns to intelligence and to Einstein's brain.
Einstein's Brain and Enhancing Our Intelligence
Einstein's preserved brain turned out to be almost ordinary, which pushes Kaku to define genius by what a brain does: intelligence is the complexity with which we can simulate the future, so tenacity, imagination and timing matter as much as anatomy. He then surveys every lever for raising intelligence (practice, savant circuitry, TMS, forgetting genes, stem cells, human-specific genes) and ends with a physicist's verdict: biology has nearly hit its ceiling, so further gains must come from efficiency and outside help.
Two epigraphs set the terms. Emily Dickinson: "The brain is wider than the sky", able to contain it "with ease, and you beside". Schopenhauer: "Talent hits a target no one else can hit. Genius hits a target no one else can see." Kaku notes that "Einstein" has become a common noun for genius, and that his authority still bends the field: when physicists in 2011 reported neutrinos apparently faster than light, the uproar spread to the popular press until recalibration showed Einstein was right again (the cause was a faulty cable connection at OPERA). "It is always dangerous to go up against Einstein."
The strange odyssey of Einstein's brain
At Einstein's autopsy in 1955, Princeton pathologist Thomas Harvey secretly kept the brain, against the family's wishes. Brian Burrell suggests he was "transfixed in the presence of greatness" and then out of his depth. Harvey, no brain specialist, never published; the brain sat in two Mason jars in a cider box under a beer cooler, was cut into 240 pieces, and slices were mailed out (once to Berkeley in a mayonnaise jar). Decades later Harvey drove it across America in a Buick Skylark in Tupperware to return it to Einstein's granddaughter Evelyn, who refused it. After Harvey died in 2007, his heirs gave the material to the National Museum of Health and Medicine in 2010. (Gauss's brain had been preserved a century earlier; it showed only unusually deep folds.)
Einstein's brain was slightly smaller than average and unremarkable to a neurologist's eye. The only notable feature: the angular gyrus and inferior parietal regions of both hemispheres were about 15% wider than usual, areas used for abstract thought, symbols and visual-spatial reasoning. The anatomy cannot explain the genius on its own.
Why Einstein was Einstein
- Thought experiments. As a theorist, his laboratory was his mind; he constantly simulated the future in pictures.
- Tenacity. About ten years per problem: racing a light beam (age 16 to 26, giving special relativity, the stars' energy and the bomb); gravity (26 to 36, giving black holes and the big bang); then a theory of everything (36 to death).
- Personality. A bohemian happy to defy 200 years of Newton.
- Timing. By 1905 Newtonian physics was cracking; radium glowed endlessly, seeming to break energy conservation. A clone of Einstein, Kaku suspects, would not be the next Einstein without the right moment.
Einstein called himself "only passionately curious", told schoolchildren his maths difficulties were greater than theirs, and said "The true sign of intelligence is not knowledge, but imagination." Darwin thought that, fools excepted, men differ little in intellect, "only in zeal and hard work."
Kaku's synthesis is a split between luck and control: we are dealt abilities by genes and brain structure, but how we arrange experience and simulate the future is within our power. For Einstein, imagination meant "shattering the boundaries of the known", and the right historical moment was as necessary as the brain.
Can genius be learned?
The brain stays plastic. In 2011 University College London scanned London taxi trainees before and after the gruelling "Knowledge" (about 25,000 streets, 3 to 4 years of study, only half pass). Those who passed grew grey matter in the hippocampus, though they scored below normal on some visual tasks, hinting at a trade-off. Eleanor Maguire: this "offers encouragement for adults who want to learn new skills later in life." Donald Hebb's principle explains it: pathways used more get reinforced. Unlike a computer, "just as dumb today as it was yesterday", the brain rewires as it learns.
Mice that have learned many tasks show the same signature: not more neurons, but altered connections. Learning literally changes the structure of the brain, and the taxi study adds a caution that the gains may be paid for elsewhere (the drivers' weaker visual-information scores).
- K. Anders Ericsson studied violinists at Berlin's elite music academy: top performers had 10,000 hours of practice by age 20 (30+ hours a week); merely excellent students about 8,000; future teachers about 4,000.
- Daniel Levitin: the number recurs for composers, athletes, writers, skaters, pianists, chess players and "master criminals".
- Malcolm Gladwell popularised it in Outliers as the "10,000-hour rule".
The 10,000-hour rule is a simplification Ericsson himself disowned (the figure was an average, not a threshold). A 2014 meta-analysis found deliberate practice explains only about 12% of performance differences overall (more in games and music, less in professions). Practice matters, but talent, starting age and teaching quality matter too.
How do you measure intelligence?
MRI shows maths puzzles engaging the pathway from prefrontal cortex to parietal lobes, fitting Einstein's enlarged parietal lobes, though whether study enlarged them or he was born that way is unknown. The deeper problem is that no one agrees what intelligence is.
IQ: Terman's bet
Lewis Terman (Stanford) revised Alfred Binet's test in 1916, believing intelligence was measurable, inherited and the best predictor of success. His Genetic Studies of Genius tracked high-IQ children, the "Termites". Decades later, they were only moderately more successful, and many took menial jobs or drifted, to Terman's dismay.
Delayed gratification: Mischel's finding
In 1972 Walter Mischel (Stanford) ran the marshmallow test on 600 children aged 4 to 6: one marshmallow now or two after 20 minutes. Follow-ups (1988, 1990, 2011) linked waiting to competence, SAT scores, income, less addiction and better social integration. Scans show a different prefrontal to ventral striatum (pleasure centre) interaction.
Kaku's summary: prefrontal to parietal links seem to matter for abstract and mathematical thought, while prefrontal to limbic links (controlling emotion and the pleasure centre) matter for success in life. Richard Davidson (Wisconsin): cooperation, emotion regulation, delaying gratification and focus count for more than IQ or grades.
Terman's test had enormous reach: 1.7 million soldiers took it in World War I, and it became the yardstick for children and for other tests. Yet among his gifted "Termites" Terman found many in menial, dead-end jobs, some in crime, some on society's margins, a result that disturbed a man who had staked his life on IQ. Michael Sweeney (Brain: The Complete Mind) puts the gap plainly: tests "don't measure motivation, persistence, social skills" and the other attributes of a life well lived. Mischel's finding, by contrast, was replicated by many groups in Kaku's telling, with the same frontal-striatal circuit difference each time; the ventral striatum contains the nucleus accumbens, so the test is in effect a contest between the pleasure centre and the rational brain.
The marshmallow result has shrunk. A 2018 replication with a larger, more diverse sample (Watts, Duncan and Quan) found the link to later achievement was about half as strong and largely vanished after controlling for family background and early cognitive skills. Self-control still matters, but the test is not the life predictor the book presents.
Convergent versus divergent thinking
Standardised tests measure convergent thinking (one line of reasoning) and carry cultural bias. In World War II the Army Air Forces asked pilots how they would get home if shot down deep in enemy territory. The best scorers were not the high-IQ pilots but the divergent thinkers who generated many unorthodox escape routes. Ulrich Kraft maps this onto the hemispheres: left for convergent, right for divergent.
Kraft's fuller description: the left side examines details logically and analytically but lacks a sense of overriding abstract connections; the right is imaginative, intuitive and holistic, integrating the pieces of an informational puzzle into a whole. Kaku connects both findings to his theory: divergent pilots simulated more possible futures, and the patient children could see long-term rewards rather than get-rich-quick ones.
Give a person a scenario (escape a deserted island full of predators and snakes; win a game) and score: 1. How many realistic future scenarios they generate. 2. How many causal links each contains. 3. How well they manipulate information toward a goal, rather than absorb it. This ties intelligence to Level III consciousness: divergent pilots and patient children are both better simulators of the future.
Boosting intelligence: savants
In Flowers for Algernon (1958; film Charly, 1968), bakery worker Charly Gordon (IQ 68) is surgically transformed into a genius, falls in love with his teacher Alice, then watches his gift fade as the lab mice die. It was dismissed as fantasy because brain cells "cannot regenerate". Kaku says: not anymore.
The plot shows what is at stake. Mocked by co-workers he cannot recognise as mocking him, unable to spell his own name, Charly is befriended by Alice, who introduces him to scientists who have made mice intelligent. Within weeks he devours libraries, fills his room with modern art and reads relativity and quantum theory; then the mice decline and die, and he watches his vocabulary, mathematics and physics drain away until a heartbroken Alice sees him playing with children. Kaku's three levers for making it real are electromagnetic devices, genetics and stem cells.
- Mr. Z: shot in the left brain at 9, left deaf, mute and partly paralysed, yet gained supernormal mechanical skill and memory.
- Orlando Serrell: hit on the left side of the head by a baseball in 1979, aged 10; afterwards calculated calendar dates thousands of years ahead.
- Benjamin Rush (1789): a seemingly disabled man computed the seconds in 70 years, 17 days and 12 hours (2,210,500,800) in 90 seconds.
- Darold Treffert's blind savant: grains doubling over a chessboard's 64 squares: 18,446,744,073,709,551,616, in 45 seconds.
- Kim Peek (model for Rain Man): could barely button a shirt but memorised about 12,000 books, reading a page in 8 seconds, each eye on a different page.
- Daniel Tammet (Born on a Blue Day): recited pi to 22,514 places; each digit has a colour or texture, but he cannot say how he recalls them.
Only about 100 "prodigious" savants are documented worldwide; about 10% of autistic people show some savant ability.
Acquired savants typically suffer a sudden trauma, diving into the bottom of a swimming pool, a baseball or a bullet, and almost always on the left side. Some researchers argue the distinction is false and all savant skills are acquired: autistic savants show their gifts around age three or four, so autism itself may act like the blow to the head. Researchers must also separate genuine savant ability from memorisation tricks; the gifts extend to music, art and mechanics, not only numbers. Peek, though painfully shy, came to enjoy fielding tricky questions from audiences, and Tammet is unusual in being able to describe his inner world in books, radio and television. About half of savants are autistic; others have acquired savant syndrome, almost always after left-brain injury. One camp says the gifts are hardwired and unique; another says evolution works incrementally, so the abilities must be latent in everyone and might be released by TMS, genes or stem cells.
Asperger's and Silicon Valley
Asperger's syndrome (recognised as distinct in 1994) involves social difficulty with articulate thought, and some have savant skills; Kaku notes speculation about Newton (incapable of small talk) and Paul Dirac. The Big Bang Theory and the engineering-school quip "the odds are good, but the goods are odd" frame a UCL test: 16 people with mild autism versus 16 controls on increasingly complex letter-number displays. The autistic group's advantage widened as tasks got harder, though they were more distracted by noise and lights. Nilli Lavie: they have "higher perceptual capacity".
What scans reveal
- Kim Peek had no corpus callosum (perhaps why he read two pages at once) and a deformed cerebellum.
- Acquired savants usually have damage to the left anterior temporal and orbitofrontal cortex, which may act as a censor that flushes irrelevant detail. With it damaged, the more literal, precise right brain takes over: "left brain injury, right brain compensation".
- Bruce Miller (UCSF, 1998) followed five people with frontotemporal dementia who developed new artistic gifts as the disease advanced: visual, copy-like rather than symbolic. One patient improved and her gift faded with it.
-
Jill Price, a Los Angeles school administrator with hyperthymesia, recalls every day for decades but cannot switch it off ("split screen" of past and present). UC Irvine scans since 2000 show an enlarged caudate nucleus and temporal lobe, a different route from savants.
-
Price is one of only four documented cases of true photographic memory in the United States, by Kaku's count. The caudate is involved in forming habits and the temporal lobe stores facts and figures; the theory is that the two work in tandem, so her brain is "stuck on autopilot".
- In Miller's dementia patients, the art was visual, not auditory, and lacked original or symbolic qualities; degeneration of the left anterior temporal and orbitofrontal cortices seemed to disinhibit the right hemisphere's visual systems.
Can we become savants?
Allan Snyder (University of Sydney) applied low-frequency TMS to the left frontotemporal region of 11 men: no savant skills, but two improved at proofreading and spotting duplicated words. R. L. Young tested 17 people and 5 reported savant-like gains. Michael Sweeney likens prefrontal TMS to "a localised jolt of caffeine". Kaku is careful: results are inconclusive, TMS hits millions of neurons over centimetres, and the key spot may be smaller, perhaps reachable later by DBS-style electrodes at the push of a button.
Forgetting to forget
In 2012 Scripps Research Institute (Florida) found in fruit flies that dopamine controls both learning and forgetting: one dopamine receptor drives memory formation, another (DAMB) drives active forgetting. Block it and flies forget less. Grad student Jacob Berry: maybe savants have "a bad forgetting mechanism", and drugs that inhibit forgetting could be cognitive enhancers.
The finding overturned the assumption that forgetting is passive decay. In the Scripps flies (trained with food rewards and electric shocks paired with odours), the dCA1 receptor drove memory formation and DAMB drove forgetting; a dCA1 mutation impaired memory, a DAMB mutation reduced forgetting. The appeal of a drug approach is control: switch photographic memory on when needed, without the savant's constant overflow of useless information. Kaku hopes the BRAIN project will pin down savant circuits precisely enough to induce them medically. The philosophy is that we are all savants waiting to happen.
Stem cells and the genetics of intelligence
The dogma that adult brains make no neurons fell in 1998, with adult stem cells found in the hippocampus, olfactory bulb and caudate nucleus. Thousands of new hippocampal cells are born daily and most die, but learning, exercise and mood-raising chemicals help them survive, while stress kills them. In 2007 teams in Wisconsin and Japan reprogrammed skin cells into stem cells, raising hopes of replacing neurons in Alzheimer's (new cells would arrive unwired, so skills would need relearning). Jonas Frisén (Karolinska): the field is in "an extremely exciting phase".
Kaku spells out the distinction: embryonic stem cells can become any cell in the body, while adult stem cells have lost that chameleon-like ability but can still divide and replace dying cells. Rats that learned new skills retained more of their newborn hippocampal cells, which links the stem-cell story back to the taxi drivers: use keeps new neurons alive.
We share about 98.5% of our DNA with chimpanzees, yet live twice as long and exploded intellectually in six million years, so a small set of genes must matter.
Pollard works in bioinformatics, a field that barely existed a decade before: instead of dissecting animals, it uses computers to compare genomes letter by letter (A, T, C, G). Her logic: we differ from chimps by 1.5% of our DNA, junk DNA drifts at a known rate, so evolution's fingerprints show up as stretches that changed faster than the clock allows.
In 2003, fresh from a Berkeley PhD, Katherine Pollard joined the chimp-genome team, hunting the 15 million letters (of 3 billion) that separate us. Junk DNA mutates at a steady rate (about 1% per 4 million years), a molecular clock consistent with a split about 6 million years ago, so accelerated change marks regions under selection. Her program at UC Santa Cruz found 201 accelerated regions. The top hit, with mentor David Haussler looking on: HAR1, 118 bases, nearly frozen for 300 million years (chimp and chicken differ by just 2 letters) but changed 18 times in the human line. "We had hit the jackpot." HAR1 is active in building the folded cerebral cortex.
| Gene / region | What it does | Human story | Defect causes |
|---|---|---|---|
| HAR1 | Shapes cortical folding during development | 18 changes in 6 million years after 300 million years of stability | Linked by Kaku to lissencephaly ("smooth brain") and schizophrenia |
| FOXP2 (Kaku: "FOX2") | Fine motor control for speech | Shared with Neanderthals, so they may have spoken | Difficulty making speech movements |
| HAR2 | Finger and thumb dexterity for tools | Accelerated in humans | (not specified) |
| ASPM | Brain size growth | About 15 changes in 5 to 6 million years; Kaku links one to modern humans 100,000+ years ago and the last to 5,800 years ago (writing, agriculture) | Microcephaly, skull the size of an Australopithecus |
| miR-941 (Kaku: "RIM-941") | Human-only gene, Edinburgh Nov 2012 | Arose 1 to 6 million years ago, absent in other primates | Over-hyped as the "intelligence gene" |
How can 23,000 genes wire 100 billion neurons with a quadrillion connections, a genome "about a trillion times too small"? Kaku's answer: random wiring that self-organises after birth through experience, and repeating modules. That also explains why a handful of mutations could grow intelligence so fast, and suggests tweaking ASPM and a few others could enlarge brains (size is necessary, though organisation matters too).
The Edinburgh announcement (November 2012) produced a firestorm of misleading headlines claiming a single gene could make a chimp intelligent; reputable scientists replied that intelligence depends on many genes acting together. Kaku adds a deeper limit: genetics explains how we became intelligent, not why genes like ASPM, HAR1 and FOXP2 were favoured, which is the question the next section takes up.
Gene details need correction. The speech gene is FOXP2, and the Edinburgh human-specific gene is the microRNA miR-941. HAR1's role in lissencephaly is not established (lissencephaly is mostly caused by genes such as LIS1, DCX and RELN). The ASPM "5,800 years ago" story comes from Bruce Lahn's 2005 papers; follow-up studies found no link between those variants and intelligence or brain size in living people.
Could we make the Planet of the Apes?
Rise of the Planet of the Apes has an Alzheimer's virus that makes chimps brilliant. Kaku runs the anatomy: a bigger chimp brain needs stronger neck muscles and body, HAR2 for dexterity, a straightened spine for upright posture to free the hands, FOXP2 for speech, and a wider birth canal (or Caesareans).
The films dodge this because their apes are humans in monkey suits or computer graphics. In the 2011 plot, an Alzheimer's virus enhances a chimp who is mistreated in a primate shelter, frees and infects the other animals, and leads them across the Golden Gate Bridge past overwhelmed police to refuge in a redwood forest. In the original series, a nuclear war reduces humans to zoo exhibits gawked at by apes. Make all the changes and you have built something that looks like a human. Society might tolerate gene studies on apes but not tormenting creatures articulate enough to complain; the bioethics is unexplored.
Where intelligence came from
- Climate and tools. Cooling Africa shrank forests; ancestors on the savannah walked upright, freeing hands; tool use rewarded bigger brains. "Tools made man."
- Social brain. Coordinating 100+ individuals in hunting, farming, war and building takes a big brain; its dark twin is Machiavellian intelligence, scheming and deceiving.
- Language. Tens of thousands of words allow abstraction, planning, maps and coordinated mammoth hunts.
- Sexual selection. Once cunning could topple the strongest alpha, leaders became the cleverest, and females chose "quarterback smart" males. Kaku's common thread: each rewards simulating the future. Animal language warns of present danger but has no past or future tense. Daniel Gilbert: for hundreds of millions of years brains were "stuck in the permanent present", until our ancestors began "a great escape from the here and now".
The physics ceiling on intelligence
Kaku argues gross human evolution has largely stopped: the birth canal caps skull size and technology removes harsh selection (glasses mean no pressure against myopia). Molecular evolution continues: malaria resistance, adult lactose digestion, adaptation to farm diets.
He dismisses the novelty-shop alien, with its bug eyes, huge head and green skin, as a picture of where evolution is not heading. Mate choice still weeds out unfit genes, but none of the recent molecular changes has altered our body plan or brain size. The physics argument starts from conservation of energy: the brain already saves energy with shortcuts (the filled-in visual field of chapter 1, snap emotional judgments, forgetting), and the conscious mind accesses only a tiny fraction of its memories. "Each time we tinker with the brain," Kaku concludes, "we are checkmated."
- Bigger brain, longer neurons: more energy and heat (tissue damage), and slower signals across greater distances.
- Thinner neurons, packed denser: ion channels are inherently unstable and begin to misfire ("the mother of all limitations").
- Thicker neurons for speed: more energy, heat and size.
- More connections: more energy and heat, a larger and slower brain. Simon Laughlin (Cambridge): cortical axons are "pretty close to the physical limit". Conclusion: gains must come from efficiency (drugs, genes, TMS-type devices) or from outside.
Parting thoughts: would boosted brains split society?
The fear: only the rich get brain boosts and entrench themselves; or a caste of super-intelligent "brahmins" rules the rest. Kaku is optimistic:
- Prices fall. Supermarket breakfasts now beat what a king of England could get a century ago.
- Most people don't want it. Few wish to solve black-hole tensor equations, and physicists earn less than businessmen and wield less power than politicians. "No one ever got rich doing relativity."
- It depends what's boosted. IQ doesn't equal success; photographic memory is a mixed blessing.
- Benefits: easier retraining, voters who grasp climate and nuclear issues, a levelled field against private tutoring, faster science and new industries (as the transistor and laser did).
-
Super-villains: a boosted Lex Luthor is dangerous only if the police can't boost too.
-
Smart is not rich. Being super smart can even pigeonhole you in a society that rewards athletes, film stars, comedians and entertainers; artistic intelligence may pay better than mathematical.
- Anxious parents. Many will want IQ boosts before standardised exams, but since IQ does not equal success and photographic memory is a curse as well as a blessing, Kaku doubts this splits society.
- Faster discovery. Science often stagnates for want of fresh ideas; people who can simulate more possible futures would speed breakthroughs.
Kaku's optimism skips the hard cases. Early access gaps (as with IVF or gene therapy costing millions) can last decades, and the people who most want boosts are the competitive, not the average. His "evolution has stopped" claim is also disputed: genome studies show ongoing selection in recent millennia. The adult human neurogenesis he relies on became contested in 2018, when one major study found almost none after childhood while another found it throughout life.
Since 2014. Genetics moved from single genes to polygenic scores: educational-attainment studies now combine thousands of variants, each tiny, and since 2023 to 2025 start-ups have offered embryo screening with predicted IQ or education scores, which most geneticists call scientifically weak and ethically fraught. Human brain genes have been put into animals: Chinese researchers gave monkeys a human MCPH1 variant (2019), and a human-specific gene (ARHGAP11B) enlarged the cortex of fetal marmosets (2020). Stem cells reached the brain clinic: in 2025 two trials reported that iPSC-derived dopamine neurons transplanted into Parkinson's patients survived and were safe. Consumer tDCS "brain boosters" boomed and faded, as meta-analyses found small or no effects in healthy users.
The clearest practical lesson here is Kaku's redefinition of intelligence as future simulation plus the marshmallow data on self-control. For leaders in a regulated bank, the same split applies to teams and AI systems: raw capability (the "IQ" of a model) matters less than planning depth, restraint and the ability to defer reward (for example, choosing auditable designs over quick wins), which are exactly the traits supervisors look for in model risk management.
Book II has asked how to extend the ordinary mind. Book III asks whether there are other kinds of consciousness: dreams, drugs, mental illness, robots and aliens.
Kaku asks readers to give up "the chauvinistic notion that our human consciousness is the only one": there is more than one way to build a model of the world and more than one way to simulate its future. Dreams, among the oldest altered states and among the least understood until recently, come first.
Altered Consciousness
Book III widens the lens. If consciousness is model-building, then dreams, drugs, mental illness, robots and aliens are other ways of building a model, and each reveals something the waking human brain hides.
In Your Dreams
Dreams are not messages from the gods or the id, Kaku argues, but the cortex trying to make sense of random bursts from the brain stem while its rational CEO and fact-checker are switched off. In the space-time theory, a dream is a simulation of the future with the laws of physics and society suspended. Scanners can now crudely photograph dreams, lucid dreamers can signal from inside them, and Inception-style entry is a long but physical possibility.
Dreams that changed history and science
- Constantine (A.D. 312). Facing an army about twice his size, he dreamed of a cross and the words "By this symbol, you shall conquer", painted it on his soldiers' shields, won, and set Christianity on the path to becoming Rome's religion.
- Joseph and Pharaoh (Genesis 41). Seven fat cows then seven lean ones meant seven good harvests then seven of famine, so Egypt stockpiled grain.
- Otto Loewi dreamed the experiment showing neurotransmitters carry signals across synapses.
- August Kekulé (1865) dreamed of a snake biting its tail and found the benzene ring: "Let us learn to dream!"
- Montaigne thought dreams interpret our inclinations; Freud's Interpretation of Dreams called them "the royal road to the unconscious", spawning dream dictionaries.
- Inception imagines thieves entering layered dreams to steal corporate secrets.
After his victory Constantine vowed to repay "the blood debt" to a religion whose followers had been fed to lions in the Colosseum, and signed laws that opened its path to imperial status; Joseph's correct reading made him a prophet in Egypt. Montaigne's full claim was that dreams are "true interpretations of our inclinations, but there is art required to sort and understand them". In Inception, DiCaprio's thief descends through dreams within dreams, layer after layer of the subconscious, hunting secrets that corporations and billionaires guard with codes. Kaku's promise for the chapter: rough photographs and videos of dreams already exist, dreams can be partly controlled with training, and entering another's dream is not ruled out by physics.
Several of these famous dream stories are shaky. Accounts of Constantine differ (Eusebius describes a vision in the sky, Lactantius a dream with the Chi-Rho sign). Kekulé told his snake story 25 years later, at a celebratory dinner, and historians doubt it. Kaku also says dream-deprived animals die faster than starving ones; the classic rat studies (Rechtschaffen) used total sleep deprivation, and REM-only deprivation killed more slowly.
The nature of dreams
- About 2 hours of dreaming a night, each dream lasting 5 to 20 minutes; about 6 years over a lifetime.
- Calvin Hall recorded 50,000 dreams over 40 years, plus 1,000 student reports: most people dream about the previous days' experiences.
- Dreams mostly use memories a few days old: wear rose-tinted glasses and your dreams turn rose-coloured only days later.
- About 6% of people have sleep paralysis, waking unable to move and sensing a creature on the chest (the Victorian goblin paintings), a possible source of alien abduction stories.
-
Dolphins sleep one hemisphere at a time to keep breathing, so any dreaming is probably half-brained.
-
Mammals other than us also enter REM, so they probably dream; Kaku cites deprivation studies to argue dreams are essential, disrupting metabolism when removed, though science cannot yet say why.
Dreams may be house cleaning. Artificial neural networks that learn too much saturate and fall into a "dream" state where memories drift and join; if all learning systems need this, learning robots might dream too. Sleep between learning and testing improves recall, and the same regions active in learning re-activate in sleep, so dreaming probably consolidates memories.
What the scans show
During dreaming, waves rise from the brain stem to the cortex, especially the visual cortex, and the eyes move rapidly (REM). Smell, taste and touch areas go quiet, and the body is paralysed to stop us acting dreams out.
Switched on in dreams
- Visual cortex: self-generated images.
- Hippocampus: draws on stored memories.
- Amygdala and anterior cingulate: strong emotion, often fear.
Switched off in dreams
- Dorsolateral prefrontal cortex: the CEO, so we drift without planning.
- Orbitofrontal cortex: the fact-checker, so physics and common sense lapse.
- Temporoparietal region: sense of body position, perhaps causing dream out-of-body feelings.
How we dream: activation-synthesis
Allan Hobson (Harvard) and Robert McCarley mounted the first serious challenge to Freud in 1977 with the activation-synthesis theory: random firing in the brain stem rises to the cortex, which synthesises a story to explain it.
- Intense emotions: the amygdala is active.
- Illogical content: scenes jump without reason.
- Apparent sensory impressions: vivid but internally generated.
- Uncritical acceptance: we believe the absurd.
- Difficulty remembering: gone within minutes of waking.
Waking alertness comes from adrenergic chemicals; in sleep the brain stem's cholinergic system takes over, its neurons fire PGO (ponto-geniculo-occipital) waves into the visual cortex, which resonates hundreds of times a second irregularly, while logic centres are chemically decoupled. Edgar Garcia-Rill (Arkansas) says meditation, worry, isolation tanks or hours staring at a blank windscreen can induce the cholinergic state awake, and schizophrenics have unusually many cholinergic brain-stem neurons, perhaps feeding hallucinations. Hobson's nightcap logged head movements (which end dreams) and eyelid movements (REM) for thousands of dream reports. He dismisses "fortune-cookie dream interpretation".
Subjects woke and immediately recorded their dreams, and the nightcap data went straight to a computer, giving Hobson an unusually large and consistent database. His conclusion is deflationary: there is no hidden message from the cosmos, only a cortex weaving a narrative from erratic signals.
Photographing a dream
Because dream imagery uses the vision machinery, it can be decoded like perception. At ATR in Kyoto, subjects saw 400 black-and-white 10×10-pixel patterns, building a dictionary that let researchers reconstruct self-generated images. Yukiyasu Kamitani expects to read feelings and emotional states the same way. At Berkeley, Nishimoto showed Kaku a crude video of his own dream: flickering faces, so he was dreaming of people. The weak spot, Nishimoto admitted, is verification, since dreams fade within minutes; Gallant called the work unpublished and in progress.
Kamitani's group works backward from the dictionary: once each pixel pattern has a known MRI signature, any mental state with a one-to-one map, including dreams, can in principle be read. The next steps are more pixels (to resolve facial features) and colour instead of black and white. Kaku notes the change in scientific mood: dreams were long avoided as too subjective and too tangled with mystics and psychics; scanners made them respectable.
Lucid dreams
Lucid dreaming (knowing you dream and steering it) was described in Buddhist texts and European diaries and is now confirmed by scans: the dorsolateral prefrontal cortex, normally dark in dreams, lights up, more so the more lucid the dream. Hobson's training method: keep a dream notebook and tell yourself before sleep that you will "wake up" inside the dream. Stephen LaBerge (Stanford) studied lucid dreamers who signal with eye movements.
At the Max Planck Institutes in Munich and Leipzig, lucid dreamers slept in an MRI wearing EEG to catch REM onset. They had agreed to signal with eye movements and then clench the right fist, then the left, for ten seconds once dreaming. The sensorimotor cortex lit up for the dreamed clenches, and near-infrared spectroscopy confirmed activity in movement-planning areas. Michael Czisch: dreams are not a passive "sleep cinema".
The dreamers' signal was a pre-agreed pattern of eye movements and breathing, a kind of Morse code, and the MRI could tell which fist was clenched first. Because the body is paralysed in REM, eye movements are one of the few channels a dreamer controls, which is why LaBerge's earlier work, with himself as a subject, relied on them.
Entering a dream
Kaku's three stages: (1) watch a decoded dream in real time and steer a lucid dreamer with directions; (2) because sleepwalkers with open eyes mix real sights with dream images (some drive, chop wood, even kill), project an observer's filmed image onto the dreamer's Internet contact lenses (University of Washington prototypes) while the decoded dream streams into the observer's lenses, letting him walk an empty studio and interact with the dream; (3) far later, link two sleeping brains through scanners and a computer that merges two dreams into one.
Inside such a dream, Kaku warns, the visitor would see backgrounds change without warning and objects appear and vanish, "and the laws of physics would be suspended. Anything goes." The question that follows is darker: if a dream can be redirected from outside, can a waking will be redirected too?
Key terms
| Term | Meaning |
|---|---|
| REM sleep | Rapid-eye-movement stage when vivid dreaming occurs and the body is paralysed |
| PGO waves | Ponto-geniculo-occipital bursts from the brain stem that drive dream imagery |
| Cholinergic / adrenergic | Neurotransmitter systems for dreaming / waking alertness |
| Activation-synthesis | Hobson and McCarley: dreams are the cortex's story for random brain-stem input |
| Lucid dream | Dreaming while aware of it, with the dorsolateral prefrontal cortex active |
| Sleep paralysis | Waking with REM paralysis still on, often with a sensed intruder |
Since 2014. In 2021 four labs (Northwestern, Sorbonne, Osnabrück and Radboud) held real-time two-way conversations with lucid dreamers, who correctly answered simple maths questions by eye signals, the "interactive dreaming" Kaku predicted. Kamitani's group decoded dream content categories from fMRI in 2013 and later used deep networks to reconstruct seen and imagined images. Targeted memory reactivation (cues played during sleep) has been shown to strengthen specific memories. Smart contact lenses stalled: Mojo Vision demonstrated one in 2022 and then pivoted away. Hobson's activation-synthesis view now competes with theories that dreams simulate threats or rehearse social situations, both close to Kaku's future-simulation framing.
The "neural nets dream when saturated" idea foreshadows generative replay and offline consolidation in machine learning. Modern AI training literally replays and recombines past experience to avoid catastrophic forgetting, a useful design pattern for models that must keep learning in production without degrading on old tasks.
If a dream can be steered from outside, perhaps a waking mind can be too. The Cold War tried.
Can the Mind Be Controlled?
From José Delgado stopping a charging bull with a radio button to the CIA's MKULTRA, mind control has been chased for a century and has failed: hypnosis, truth serums and LSD cannot override a person's will. The new danger, Kaku argues, is precision. Optogenetics and implanted electrodes can switch specific behaviours on and off, so the same tools that could free people from mental illness could one day be used to hijack them.
Delgado's bull
In the 1960s in Córdoba, Spain, Yale's José Delgado, dressed as a matador, faced a fighting bull with electrodes in the striatum of its basal ganglia. As it charged, he pressed a button on a box and the bull stopped dead. With electrodes in the caudate nucleus of an alpha male monkey, he could dial down its aggression on command; subordinates took its territory until he switched it back. His 1969 book, Physical Control of the Mind: Toward a Psychocivilized Society, raised the obvious question: who controls the puppet master?
The theatre was deliberate: Delgado wore a golden matador's jacket and waved a red cape at a bull bred for generations to kill, risking his life before the cameras to make the point. In the monkey colony the lower-ranking "delta males" took the alpha's territory and privileges while he was switched down, and scrambled in fear when a second button restored him. Kaku frames the dual use from the start: in a dictator's hands the method deceives and controls; in a clinic it could free millions trapped in hallucinations and anxiety. The chapter's epigraph is Marvin Minsky's: "Minds are simply what brains do."
Delgado said he wanted a humane alternative to lobotomy, in which an ice-pick-like blade was hammered above the eye socket to scramble the prefrontal cortex. Some patients calmed; many became emotionally vacant "zombies" (the abuse dramatised in One Flew Over the Cuckoo's Nest). Egas Moniz won the 1949 Nobel for the procedure; the Soviet Union banned it in 1950 as "contrary to the principles of humanity"; about 40,000 were done in the US over two decades.
The Soviet charge was blunt: lobotomies turned "an insane person into an idiot".
Kaku names him "Antonio Moniz" and says he "perfected" the lobotomy. The Portuguese neurologist was António Egas Moniz, who devised the prefrontal leucotomy; the ice-pick transorbital version was Walter Freeman's. Kaku also dates Brave New World to 1931; it was published in 1932. And Delgado's bull may have stopped because stimulation interfered with movement, not because its rage was switched off.
Cold War fears
Captured US soldiers in the Korean War appeared on camera with blank stares confessing to spying and war crimes; the press coined "brainwashing". The Manchurian Candidate (1962, Frank Sinatra) dramatised a brainwashed war hero programmed to kill the president. Huxley's Brave New World had already imagined test-tube factories where oxygen deprivation produced castes from Alphas (rulers) to Epsilons (brain-damaged workers), kept placid by drugs, free love and conditioning, asking how much freedom we would trade for order.
MKULTRA
- Ran 1953 to 1973; CIA director Richard Helms ordered records destroyed in 1973 amid Watergate.
- About 20,000 documents survived and were released in 1977 under the Freedom of Information Act.
- Funded 80 institutions, including 44 universities and colleges, plus hospitals, drug companies and prisons, in 150 secret operations, often on unwitting subjects.
- At its peak, about 6% of the CIA budget.
- A truth serum for prisoners.
- Memory erasure (the Navy's "Subproject 54").
- Hypnosis and drugs, especially LSD, to control behaviour.
- Mind-control drugs against foreign leaders such as Fidel Castro.
- Better interrogation methods.
- A fast knockout drug leaving no trace.
- Personality alteration to make people pliable.
Psychics were even asked to locate Soviet submarines.
Recruits ranged across psychics, physicists and computer scientists, and the work was never peer-reviewed, so millions of taxpayer dollars went to schemes that physics alone should have killed. The Army report's own words show the ambition: pulsed microwaves might let operators "'talk' to selected adversaries in a fashion that would be most disturbing to them". Mind control was not new; Kaku notes that ancient sorcerers sold potions promising to make captives talk or turn on their leaders. An Army scientist secretly dosed with LSD reportedly became so disoriented he leapt to his death from a window. A secret Senate briefing claimed the Soviets beamed microwaves at subjects' brains; rather than object, US agencies saw "great potential" and an Army report speculated that pulsed microwaves could create intelligible speech in an enemy's head. Steven Rose (Open University) called it a "neuro-scientific impossibility", and Kaku says the brain can neither receive nor decode microwave messages. Not one reliable result came out of the black projects. Robert Jay Lifton found that brainwashed POWs, like cult members, reverted to their old personalities once freed.
Kaku overstates the physics. Pulsed microwaves can be "heard": the Frey microwave auditory effect (clicks or buzzes from tiny thermoelastic expansion of tissue) is real, though carrying speech is unproven. The question resurfaced with "Havana syndrome" among US diplomats from 2016; a 2023 US intelligence assessment judged a foreign energy weapon very unlikely, while some agencies later softened that view, so the debate is not closed.
Hypnosis
Stage hypnosis (the man who clucks like a chicken) relies on shills, suggestion and willing volunteers. Hypnotised for a BBC/Discovery documentary, Time, Kaku recovered faces and places forgotten for decades, "like watching a blurred movie slowly coming into focus", then hit a wall. Scans show reduced outside sensory input but no new state of consciousness like REM. A 1966 Pentagon document conceded there were "no reliable accounts" of hypnosis used effectively by any intelligence service. Kaku's verdict: it can sharpen recall but cannot change goals or make you squawk without your permission.
The session itself was ordinary: a dark, quiet room, a slow and gentle voice, an invitation to re-enter a vivid place with its sights, sounds and smells. In Kaku's framework the result follows naturally: hypnosis cannot alter the basic process of building a model and simulating the future toward a goal, so it can accentuate some aspects of consciousness but cannot hijack the goal itself.
Truth serums and how drugs hijack the brain
US Army and CIA manuals released in 1996 still recommended truth serums, though the Supreme Court ruled such confessions coerced. The film favourite, sodium thiopental ("Pentothal"), is a barbiturate that crosses the blood-brain barrier and depresses the prefrontal cortex, making people relaxed and talkative but not honest: like drunks, they can still lie, so the CIA gave up.
Kaku pictures synapses as toll booths: stimulants open them all at once (the rush); then they cannot fire for hours (the crash); craving the rush becomes addiction. Animals given free access to cocaine, heroin or amphetamines take them until exhaustion or death.
- By 2007, 13 million Americans aged 12+ (5%) had tried methamphetamine.
- Meth addicts' MRIs: 11% smaller limbic system, 8% hippocampal tissue loss, damage comparable in some ways to Alzheimer's.
- Meth's high can reach 12 times the reward of a great meal or sex.
- 50 micrograms of LSD can cause hallucinations.
- Cocaine and stimulants push VTA cells to release more dopamine, stop them switching off, and block reuptake of serotonin and noradrenaline, flooding all three systems.
- Heroin and opiates silence the cells that restrain dopamine output.
- LSD acts on serotonin, giving well-being and affection, and activates temporal-lobe areas that create hallucinations; it binds so tightly that bigger doses add nothing.
The CIA found drugs too unpredictable for delicate politics. A twist: smokers who suffered strokes damaging the insula quit far more easily, and the same held for cocaine, alcohol and opiates. Nora Volkow (NIDA): "It's mind-boggling." Damping the insula with electrodes or magnets might treat addiction.
The insula lies deep between the prefrontal and temporal cortex, and no one knew why its damage worked, because it handles a bewildering range of jobs, including perception, motor control and self-awareness. Kaku's point is that the reward circuit is millions of years old and essential for survival, rewarding useful behaviour and punishing harm, which is exactly why hijacking it is so destructive.
LSD does not "stimulate the production of serotonin"; it mimics serotonin at the 5-HT2A receptor. Kaku's phrase "three main drugs... Dopamine, serotonin, and noradrenaline" means neurotransmitters, not drugs.
Optogenetics: precision control
The opsin gene, which may have produced the first light-sensitive eyespot hundreds of millions of years ago, makes a neuron fire when lit. Packed into a harmless virus, it is delivered to chosen neurons; a light fibre then switches them on. Results: fruit flies bolt on command once the escape circuit is found, worms stop wiggling, and in 2011 Stanford lit a precise amygdala circuit in specially bred timid mice, which lost their fear and explored.
- Ed Boyden (MIT): an optical fibre beats cutting out a brain region.
- Parkinson's: could make DBS precise to the neuron, avoiding dizziness and muscle spasms from stray stimulation.
- Touch for prostheses: Krishna Shenoy (Stanford) proposes feeding fingertip sensor data back optogenetically for high-fidelity touch.
-
Hurdles in humans: open skull, deep targets, and wires to deliver light.
-
The opsin story. One theory holds that a patch of skin made light-sensitive by opsin evolved over hundreds of millions of years into the retina. The gene goes into a virus with its harmful genes removed, the virus infects a single target neuron, and a flash of light then traces the exact pathway a message takes.
- From flies to people. Fruit-fly reflexes involve a handful of neurons, but mice have full limbic systems with human counterparts, so the timid-mouse result points toward side-effect-free treatment of specific mental illnesses. Kaku foresees an encyclopedia of behaviours and their circuits, human ones included, which is the source of both the promise and the risk.
Mind control and the future
Drugs and hypnosis flopped; propaganda and coercion remain cheaper. The nightmare that might work is Carl Sagan's: a dictator wiring children's pain and pleasure centres to computers. Another is electrodes overriding motor commands so your own body becomes a foreign object. Kaku's three reassurances: the technology is young enough to regulate; dictators find propaganda cheaper; democracies will debate and legislate, which needs an informed public. His bet: these tools will liberate minds trapped in mental illness rather than enslave them.
Kaku spells out how the body-hijack would feel: the injected impulse could be stronger than the signals you send your own muscles, so your limbs would fire against your will as if someone had taken over your body. Delgado found only a few controllable behaviours; future researchers may find many. Hence his line: "A fine line has to be drawn between technologies that can benefit society and technologies that can control it."
Key terms
| Term | Meaning |
|---|---|
| MKULTRA | CIA mind-control programme, 1953 to 1973 |
| Blood-brain barrier | Filter that keeps most chemicals out of the brain; many drugs cross it |
| VTA / nucleus accumbens | Dopamine source and pleasure centre of the reward circuit |
| Opsin | Light-sensitive protein used in optogenetics to make neurons fire on command |
| Insula | Deep cortical region whose damage eased addiction |
| Leucotomy / lobotomy | Surgical cutting of prefrontal connections, now discredited |
Since 2014. Optogenetics reached a human in 2021 (partial sight restored in a blind man). Brain stimulation is now an approved addiction therapy: the FDA cleared deep TMS aimed at the insula and prefrontal cortex for smoking cessation in 2020, the clinical payoff of the insula finding Kaku reports. Psychedelics returned as medicine: large psilocybin trials for depression reported positive phase 3 results in 2025, while the FDA rejected MDMA-assisted therapy for PTSD in August 2024 over trial-quality concerns. Closed-loop DBS that senses a mood biomarker and stimulates only when needed has shown striking results in single patients since 2021. The worries have scaled with the tools: neurorights laws (chapter 3) explicitly name manipulation of mental states as a harm.
The chapter's arc (crude tools fail, precise tools work, so governance must arrive before precision) fits persuasive AI closely. Recommendation engines and conversational agents can now nudge beliefs with measurable effect; the FCA's Consumer Duty concept of avoiding foreseeable harm and exploiting behavioural biases is the regulator's version of Kaku's line between technology that benefits and technology that controls.
Kaku closes with a bridge to the next chapter: modern neuroscience is now used to re-diagnose historical figures, starting with Joan of Arc.
Altered States of Consciousness
Kaku uses religious ecstasy, schizophrenia, OCD, bipolar disorder and coma to test his theory, and arrives at a definition: mental illness is largely the disruption of the checks and balances between competing feedback loops that simulate the future, usually because one region is over- or under-active. The practical hope lies in deep brain stimulation, cross-disorder genetics and, later, a full wiring map of the brain.
Joan of Arc and the "God spot"
An illiterate peasant girl who heard God's voices led a demoralised French army to victories in the Hundred Years' War that cleared the way for Charles VII's coronation. Captured and tried by the English, she was burned for heresy in 1431, aged 19. Her trial answers were so sharp that George Bernard Shaw lifted them into Saint Joan: asked if she was in God's grace (a trap either way), she said "If I am not, may God put me there; and if I am, may God so keep me." The notary wrote that her interrogators were "stupefied".
The question was a perfect trap: say yes and she was a heretic, since no one can know they are in God's grace; say no and she confessed to fraud. Charles VII, with nothing to lose, had let her command troops, and each victory turned her into a folk heroine who rallied a France close to collapse. The chapter's epigraph is Shakespeare's "the lunatic, the lover, and the poet are of imagination all compact".
Her lucid testimony argues against schizophrenia. A newer theory is temporal lobe epilepsy, which in some patients brings hyperreligiosity, a sense that everything is charged with divine meaning. Eagleman suggests "some fraction of history's prophets, martyrs, and leaders" had it, Joan among them (her voices: St Michael, St Catherine, St Margaret and St Gabriel).
- 1892: textbooks link "religious emotionalism" to epilepsy.
- 1975: Norman Geschwind (Boston VA) describes religious experiences in patients with left temporal lobe misfiring.
- Ramachandran: 30 to 40% of his temporal-lobe epileptics show hyperreligiosity; their skin response jumps at the word "God" but not neutral words. He half-jokes they might have "a wormhole... Into a parallel universe".
- Persinger's "God helmet": magnetic stimulation of the temporal lobes in three-minute bursts; subjects describe a presence they call God, Buddha or the wonder of the universe.
- Beauregard (Montreal): 15 Carmelite nuns in an MRI recalling union with God ("God cannot be summoned at will").
- Richard Dawkins wears the God helmet: his atheism is unchanged.
Ramachandran's patients describe the experience in cosmic terms: "Finally, I see what it is all really about, Doctor. I really understand God. I understand my place in the universe." Because Persinger's effect could be produced on demand, Kaku reads it as evidence that the brain may be hardwired to respond to religious feelings, which is also why "God gene" speculation and the tribal-cohesion theory of religion (shared myths binding bickering individuals) arise. The Montreal design had a practical flaw: an MRI tube, ringed with tons of magnetic coils, is no place for an epiphany, so the nuns could only recall past experiences.
Each lit region invited a story: the caudate (unconditional love of God?), the insula (closeness to the other sisters?), the parietal lobe (the physical presence of God?). With that many regions and interpretations, Beauregard admitted he could not say whether hyperreligiosity had been induced, a useful caution against reading meaning into any single brain-scan "hot spot".
The nun study lit up the caudate nucleus (learning, perhaps love), the insula (body and social feeling) and the parietal lobe (sense of presence in space), with too many regions for a clean conclusion. The nuns were delighted: God had given the brain a radio to receive Him. Beauregard: an atheist will read such an experience as the universe's magnificence, a Christian as God, "Perhaps they are the same thing." Some speculate about a "God gene"; others that religion helped tribes cohere. Kaku's verdict: hyperreligiosity can be induced, but no evidence that magnets change religious belief.
Corrections and caveats. Joan came from Domrémy, not Orléans; "Maid of Orléans" refers to the siege she lifted. Persinger's helmet used weak magnetic fields, far below TMS strength, and a 2005 double-blind Swedish replication found the "presence" effect tracked suggestibility, not whether the device was switched on.
Schizophrenia
About 1% of people have schizophrenia: voices, paranoid delusions and disordered thought, historically blamed on demons (Jesus drives two men's demons into a herd of swine). Onset is usually in the late teens for men, early twenties for women. John Nash, whose adviser's recommendation read "This man is a genius", did prize-winning game theory before a breakdown at 31 and years of fearing communist assassins; he won the 1994 economics Nobel.
- Drugs. Thorazine was discovered by accident in the 1950s; antipsychotics block dopamine D2 receptors (excess dopamine also explains amphetamine hallucinations). Dopamine is also implicated in Parkinson's (too little) and Tourette's (too much).
-
Glutamate. PCP (angel dust) mimics schizophrenia by blocking the NMDA glutamate receptor; clozapine helps some patients that other drugs don't.
-
Why drugs are hit or miss. One theory says antipsychotics imitate a natural chemical missing in schizophrenics' brains, but not exactly, so patients cycle through drugs by trial and error. There is still no precise, accepted diagnostic test for mental illness; Kaku hopes scans will provide one. Tourette's, linked to excess dopamine, produces tics and, in a small minority, uncontrollable obscene speech.
- History. Sufferers were long treated as possessed: banished, killed or locked away, like the hidden relative in Gothic novels. Nash's story became the film A Beautiful Mind (Russell Crowe); he did pioneering work at Princeton in his twenties while already hounded by delusions.
- Results. About 20% of patients become symptom-free, about two-thirds improve somewhat, the rest not at all; side effects drive many to stop and relapse.
- Scans. When we silently talk to ourselves, Wernicke's area and the temporal lobe light up; when schizophrenics hear voices, the same areas light up. The patient is literally talking to himself without knowing it. Michael Sweeney compares it to oily rags igniting spontaneously in a dark garage.
The voices usually come from a third party issuing commands, mostly mundane, occasionally violent, and the brain, desperate for a consistent narrative, explains them with stories such as Martians beaming thoughts into the head. Kaku's reading: the prefrontal simulators are on autopilot, running the simulations we all run, but without the patient's permission.
Hallucinations
The anterior cingulate cortex, between the prefrontal cortex and the limbic system, tags signals as external or self-made. Damage it, as may happen in schizophrenia, and self-generated voices are taken as real. Everyone hallucinates a little: dark rooms, isolation tanks and eerie places produce false images, and pareidolia makes us see faces in clouds. In a sense all perception is controlled hallucination, since the brain fills gaps constantly.
OCD
About 1 in 40 Americans has obsessive-compulsive disorder, from repeatedly checking locks (TV's detective Monk) to scrubbing skin raw for hours. In moderation such behaviours evolved to keep us clean and safe.
Clomipramine (from the 1960s) and later serotonin-raising drugs cut symptoms by up to 60%. Schwartz: "The brain's gonna do what the brain's gonna do, but you don't have to let it push you around."
Bipolar disorder and depression
Bipolar disorder swings between delusional optimism and crushing depression, runs in families and is common among artists; lithium controls many symptoms without anyone knowing exactly why. One theory is hemispheric imbalance: left-hemisphere damage tends to bring depression, right-hemisphere damage odd cheerfulness, suggesting the left brain is the optimist and the right a "devil's advocate" (Ramachandran) that keeps us objective.
Kaku's twist: simulating the future requires weighing optimism against pessimism, and depression is the price we pay for being able to imagine every bad outcome.
Sweeney summarises a century of lesion evidence: damage to the left hemisphere brings negative moods, including uncontrollable crying; damage to the right brings an array of positive emotions. Ramachandran's conclusion: "If left unchecked, the left hemisphere would likely render a person delusional or manic." The creative toll is visible in the long roll of artists, musicians and writers with bipolar disorder; perhaps, Kaku suggests, great works came during the bursts of optimism. In depressed brains parietal and temporal activity is suppressed (withdrawal inward), and the ventromedial cortex, which gives the world a sense of meaning, is under-active; over-activity may produce manic omnipotence.
Kaku's theory of mental illness
Mental illness is largely caused by the disruption of the delicate checks and balances between competing feedback loops that simulate the future. The DLPFC "CEO" relies on rival assessments, like a good chair drawing out disagreement in a board meeting; when one voice is too loud or too quiet, its conclusions turn strange. The theory is testable: scan patients during symptoms and look for unbalanced loops; it fails if symptoms prove independent of those interactions.
| Illness | Feedback loop 1 | Feedback loop 2 | Brain regions |
|---|---|---|---|
| Paranoia | Perceiving threats | Discounting threats | Amygdala vs prefrontal cortex |
| Schizophrenia | Creating voices | Discounting voices | Left temporal lobe vs anterior cingulate |
| Bipolar disorder | Optimism | Pessimism | Left vs right hemisphere |
| OCD | Anxiety | Satisfaction | Orbitofrontal cortex, caudate, cingulate |
These loops evolved to protect us; illness is the breakdown of their balance, not the loops themselves.
Deep brain stimulation: "Depression 3.0"
Depression affects about 20 million Americans, 10% of them untreatable. Helen Mayberg found Brodmann area 25 (the subcallosal cingulate) consistently over-active in treatment-resistant patients. DBS had already been used on 40,000 patients for Parkinson's and epilepsy, with 60 to 100% reporting much steadier hands, at more than 250 US hospitals. Of 12 patients who had failed drugs, therapy and electroconvulsive therapy, 8 improved quickly after area-25 stimulation; trials expanded to 35 patients at Emory and 30 elsewhere.
Depression 1.0 was psychotherapy... Depression 2.0 was the idea that it's a chemical imbalance. This is Depression 3.0.
Helen MaybergOpen problems: no one knows why DBS works (it may silence over-active tissue, so it suits disorders of over-activity); electrodes are imprecise, touching up to a million neurons; the 1.5 mm probes cause bleeding that progresses to stroke in 1 to 3% of patients; and pulses are a crude constant rhythm. MEMS micro-electrodes, carbon-nanotube probes and better MRI guidance should sharpen it. DBS is also used for phantom-limb pain, Tourette's and OCD.
Waking from a coma
Terri Schiavo suffered oxygen deprivation after a heart attack in 1990; her husband wanted to let her die, her family fought on. In 2003 most neurologists judged from scans that she was in a permanent vegetative state; her 2005 autopsy confirmed massive damage. Other patients have more left. In 2007 a Cleveland man in a minimally conscious state for eight years woke and greeted his mother after Ali Rezai's team stimulated his thalamus with DBS.
The case is instructive because current usually shuts a region down; here, low-voltage stimulation of the thalamus, the sensory gateway, jolted neurons into action. Schiavo's family had argued she still responded to stimuli and might wake, as some long-term patients famously had, which is why brain scans became the arbiter. DBS also has an unexpected side effect Kaku notes: it can increase memory cells in the hippocampus. Future probes would be tuned to each patient and ailment instead of pulsing at one constant rate.
Genetics: a common thread
Family studies found genes in one family that failed to generalise. Then in 2012 Harvard/Mass General researchers analysing 60,000 people found shared risk across five disorders: schizophrenia, bipolar disorder, autism, major depression and ADHD. Of four risk genes, two regulate calcium channels. Jordan Smoller suggested calcium-channel drugs might work across disorders ("just the tip of the iceberg"); it may explain why one twin has schizophrenia and the other bipolar disorder.
Calcium is essential to neural signalling, and calcium-channel blockers were already used for bipolar disorder, so the finding had an immediate clinical hook. Before 2012 the consensus was that each disorder had its own genetic basis, with environment plus several genes needed to trigger illness; family-tree genes rarely generalised. A shared thread would let one drug, or one gene therapy, reach several disorders.
Five avenues: (1) new neurotransmitters and drugs; (2) genes and gene therapy; (3) DBS to damp or boost regions; (4) EEG, MRI, MEG and TMS to see the malfunction; (5) whole-brain wiring maps (chapter 11). Two families: 1. Injury disorders (Parkinson's, epilepsy, Alzheimer's, stroke, tumours): damaged or over-active tissue, treated by probes that silence it, or by stem cells and artificial replacements. 2. Wiring disorders (schizophrenia, OCD, depression, bipolar): healthy regions mis-wired, hardest to treat, so far handled by hit-or-miss neurotransmitter drugs.
Corrections. Mayberg's pioneering area-25 DBS (2005) was done at the University of Toronto, not Washington University. The large industry-sponsored trial (BROADEN) was halted for futility in 2013, a setback the book doesn't mention. By 2014 more than 100,000 people had received DBS worldwide. Kaku's left-optimist, right-pessimist hemisphere model of bipolar disorder is a speculative minority view.
Key terms
| Term | Meaning |
|---|---|
| Hyperreligiosity | Pervasive sense of divine meaning, linked to temporal lobe epilepsy |
| Anterior cingulate cortex | Tags signals as self-generated or external; implicated in hallucinations |
| D2 receptor | Dopamine receptor blocked by classic antipsychotics |
| Brodmann area 25 | Subcallosal cingulate, over-active in resistant depression; DBS target |
| Minimally conscious state | Partial awareness after brain injury, distinct from a vegetative state |
| Cognitive motor dissociation | Covert awareness detectable only by brain imaging |
Since 2014. Psychiatry's drug pipeline finally produced new mechanisms: esketamine nasal spray for resistant depression (2019), zuranolone for postpartum depression (2023), and xanomeline-trospium (Cobenfy), approved in September 2024, the first schizophrenia drug in decades that does not block dopamine D2 receptors (it acts on muscarinic receptors). An accelerated, MRI-targeted TMS protocol (SAINT) won FDA clearance for depression in 2022. Mayberg's group refined DBS with tractography targeting, and UCSF reported a closed-loop device that detects a patient's depression signature and stimulates only then (2021). Psychiatric genetics exploded: a 2022 study found 287 schizophrenia risk regions, confirming the shared-genetics picture. For coma, a 2024 study found about one in four behaviourally unresponsive patients could follow commands in their heads on fMRI or EEG ("cognitive motor dissociation"), sharpening the Schiavo-style debates.
Kaku's "balanced feedback loops" theory is a good template for AI control design: an agent with a strong "generate" loop (proposing actions, like voices) needs an equally strong "discount" loop (verification, policy checks). Many AI failure modes (confident hallucination, runaway retry loops, alarm fatigue) look like his table of disorders: one loop stuck on, the check stuck off.
The last altered state Kaku considers is not human at all: artificial intelligence.
The Artificial Mind and Silicon Consciousness
Writing in 2013, Kaku judged that AI had huge raw power but no common sense, weak pattern recognition and no self-awareness: robots sat at about Level I:10 consciousness. His roadmap for machine minds runs through emotions (which tell a mind what matters), values, a theory of mind and, finally, future simulation with goals. The danger he sees is not malice but sloppy goal design, so the safest path is friendly AI and, eventually, merging with our machines. This is the chapter that has aged fastest, and the 2026 boxes show how.
Watson and the hype cycle
In February 2011 IBM's Watson beat two champions on Jeopardy! and took the $1 million prize. It processed 500 gigabytes per second (a million books a second), with 16 trillion bytes of RAM and 200 million pages of material including all of Wikipedia. Kaku calls it an expert system and "a highly sophisticated adding machine" that did not know it had won: you could not congratulate it or share champagne. Loser Ken Jennings quipped "I for one welcome our new computer overlords." Kaku foresaw expert systems as a cheap wristwatch "robo-doc" with 99% accuracy, robot lawyers, robo-secretaries and chatbots working within the few hundred words of casual talk.
The arithmetic behind chatbots: an adult knows tens of thousands of words, a newspaper needs about 2,000, a casual conversation only a few hundred, so a narrow-topic bot can pass for conversational. A phone menu is a primitive expert system. Jennings also joked that he and fellow contestant Brad were "the first knowledge-industry workers put out of work" by thinking machines. Kaku's framing contrast: a visitor from 1900 would see today's computing as miraculous, yet no robot of 2013 was a true automaton without a hidden puppet master, and robots are totally unaware that they are robots. The epigraph is Turing's joke that he wanted only "a mediocre brain, something like the President of the American Telephone and Telegraph Company".
- 1950s to 1960s: checkers players, algebra solvers, block-stacking arms, Stanford's Shakey robot. Herbert Simon (1965): machines will do "any work a man can do" within 20 years. Minsky (1967): AI "substantially solved" within a generation.
- 1974: first AI winter; US and UK slash funding; Shakey gets lost in unfamiliar rooms.
- 1980s: Pentagon-driven revival; AI funding hits $1 billion by 1985; the autonomous Smart Truck "only... Got lost". Second winter in the 1990s.
- 1997: IBM's Deep Blue beats Garry Kasparov.
- 2005: Stanford's driverless car wins the DARPA Grand Challenge.
- 2011: Watson wins Jeopardy!
Paul Abrahams likened MIT's AI effort to building a tower to the Moon: taller each year, but the Moon no closer. Pinker would pay a lot for a robot that put away dishes, but the sub-problems (recognising objects, reasoning, moving hands and feet) were unsolved. Minsky told Kaku he no longer gives dates. (Popular Mechanics in 1949 was too modest: computers "will weigh no more than 1.5 tons.")
Kaku's diagnosis of why the first two booms failed: researchers vastly underestimated the problem because most human thought is subconscious; the conscious part is the tiniest fraction of the brain's computation, so the hard work is the part we never notice ourselves doing. Films like 2001: A Space Odyssey had promised robots piloting ships to Jupiter; the reality was one-trick ponies.
The two walls: pattern recognition and common sense
- Pattern recognition. A robot breaks a room into pixels, lines and shapes and matches them against stored images, but cannot grasp "chairness"; a knocked-over or rotated chair baffles it.
- Common sense. No equation says "muggy weather is uncomfortable" or "mothers are older than their daughters"; cataloguing a four-year-old's common sense would take hundreds of millions of lines of code. The CYC project tried exactly this, and stagnated.
-
ASIMO's secret. Honda's engineers privately told Kaku their famous humanoid has "the intelligence of a bug": mostly a big tape recorder, every gesture scripted. A short TV segment with Kaku took three hours to program.
-
CYC's record. Millions of lines of code and hundreds of thousands of facts still could not reach a four-year-old's level; after optimistic press releases, programmers left and deadlines passed. Voltaire's line fits: "Common sense is not so common." ASIMO, the size of a young boy, walks, runs, climbs stairs, speaks several languages and dances ("much better than I do"), which is exactly why the gap behind the show surprised Kaku.
Machines excel only at simulating the future in one well-defined domain (chess, weather, colliding galaxies) where the rules are known; they cannot plan a bank robbery, which needs hundreds of nuances about banks, police and bystanders.
Is the brain a computer?
Joseph Campbell: computers are "like Old Testament gods; lots of rules and no mercy". Remove one transistor and a Pentium crashes; the brain works with half missing. The brain is a neural network that rewires itself as it learns, massively parallel, with no program or CPU. Hence a shift from top-down AI (writing all the rules) to bottom-up learning. Rodney Brooks (MIT AI Lab, co-founder of iRobot) marvelled that a mosquito out-flies any robot plane with a speck of a brain, and built insectoid "bugbots" that learn by bumping into things. In a neural net "programming is nothing; changing the network is everything."
Kaku says Mars rovers like Curiosity "learn as they bump into obstacles". They do not: their autonomy is programmed hazard avoidance, with most drives planned by engineers on Earth. More strikingly, the book misses the revolution already under way as he wrote: in 2012 the deep neural network AlexNet smashed the ImageNet image-recognition benchmark, the start of the deep learning era that would demolish his "pattern recognition" wall within a few years.
Ranking robots on the consciousness scale
| Machine (2013) | Level | Kaku's score | What is missing |
|---|---|---|---|
| Static industrial robot | 0 | A few loops | Mobility |
| Mobile robot (vision, hearing loops) | I | I:10 (an insect has 50+) | Finding shelter, mates, food; evading predators |
| Social robots | II | II:0 today | Recognising and responding to emotions; group relationships |
| Supercomputer simulations | III | III:5 | Common sense, causal reasoning about real life, theory of mind |
His forecast: over coming decades robots climb through Level II, becoming as smart as a mouse, rat, rabbit, then cat, and late this century a monkey, beginning to set goals of their own. Level III requires common sense plus a theory of mind so a robot can star in its own simulations of the future.
Speed bumps: the end of Moore's law
Kaku warns that Moore's law violates quantum theory eventually. Chips' thinnest layers were about 20 atoms wide, perhaps 5 atoms by 2020; then Heisenberg uncertainty lets electrons tunnel out of wires and heat could fry an egg. Stopgaps: Intel's 3-D chips (heat rises with height), Microsoft's parallel processing across many chips (software lags hardware). After silicon: quantum, molecular, nano, DNA and optical computers, none ready.
The uncanny valley
Masahiro Mori (1970): empathy for a robot rises as it looks more human, then plunges when it is nearly but not quite human, and recovers only at perfect realism. Critics found the humans of The Polar Express "downright... Creepy". A realistic robot smile makes skin crawl (think grinning Halloween ghouls), partly because we distinguish a forced smile (prefrontal) from a felt one (limbic). Scans show the mismatch between predicted human motion and jerky robot motion lighting the parietal lobe where motor and visual cortex meet (mirror neurons), with the orbitofrontal cortex flagging "something is not quite right".
Horror and humour share a mechanism in Kaku's theory: the brain simulates an ordinary future and gets an unexpected outcome. A joke perverts the ending pleasantly; horror perverts the ordinary grotesquely (Linda Blair's head turning in The Exorcist; young monkeys ignore Dracula pictures but scream at a decapitated monkey). So robots will stay childlike in appearance until they can act fully human.
Silicon consciousness: emotions and goals
AI long treated emotion as noise (the Spock ideal). But Antonio Damasio found patients whose prefrontal-limbic link is cut cannot make decisions (which pen, which appointment) because nothing has value. Emotions are how a mind knows what is important; a robot without them might save the files before the people in a fire.
Since robots come with no values, a huge list of value judgments must be uploaded: human life above objects, children first in emergencies, pricier objects above cheaper ones. Silicon minds will also lack our evolutionary quirks, for better and worse. Human emotions are fast, limbic and biased ("beauty has its privileges": good-looking people are overrated and promoted), and we read deference unconsciously in body language, which is older than speech: juniors yield to seniors in posture, words and gestures. Robots will have to learn these cues explicitly.
- Cynthia Breazeal (MIT Media Lab): Huggable, a teddy bear with camera eyes and touch-sensitive skin for children; Nexi, a round-faced robot that nursing-home residents kissed and missed. MIT's Frank Moss: in 2004 she set out to build social robots "that could live anywhere".
- Waseda University: a robot whose upper body expresses fear, anger, surprise, joy, disgust and sadness, with simple drives such as hunger for energy.
- Lola Cañamero (University of Hertfordshire, EU-funded Feelix Growing): Nao, modelled on a one-year-old chimp, shows happiness, sadness, fear, excitement and pride through posture; remembers faces, follows gaze, bonds and develops different personalities; meant to ease children's anxiety in hospitals.
-
Terrence Sejnowski (Salk): soon you'll "talk to it, flirt with it, or even get angry and yell at it".
-
Why Breazeal started. Earlier robots looked like "tin cans full of wires, gears and motors" and never left the lab shelf; the Media Lab, a "toy factory for grown-ups" whose graphics reached Minority Report and AI, aimed at robots for homes, schools, hospitals and elder care. Huggable could become a tutor, babysitter, nurse's aide or playmate; Nexi looks a little like the Pillsbury Doughboy.
- Nao in detail. About one and a half feet tall, Nao locks onto faces, remembers past interactions, follows gaze, treats a smile or a pat as positive, and grows bonds over time; its emotional displays are pre-programmed, but it chooses which to show. Because each relationship differs, different personalities emerge, one independent, another timid and needing constant reassurance. Feelix Growing spans the UK, France, Switzerland, Greece and Denmark.
- Keep: loyalty. Faithful, anticipates needs, obeys its owner over conflicting commands, criticises only tactfully.
- Keep: empathy. Reads faces and voice to detect distress and help.
- Keep: fear. A fearless robot destroys itself in fires or falls.
- Delete or regulate: anger. Dangerous in a strong machine; dissatisfaction can be expressed calmly.
- Delete: desire to command. A bossy robot challenges its owner (and matters for takeover risk).
- Avoid for now: humour. Too subtle; laughing at the wrong time is disastrous (Reagan defused the age question by promising not to exploit his opponent's "youth and inexperience").
- Identify the emotion from face, lips, eyebrows and voice (a dictionary, going back to Darwin's catalogue).
- Respond quickly from an encyclopedia (grin back; step away from anger).
- Infer the cause (laughter can mean joy, a joke, a fall, nerves or contempt): find the reason that best fits the data. Hardest step.
- Choose the best response, which means simulating the future of each option.
Will robots lie? Can they feel pain?
Social life runs on white lies (Liar Liar shows the chaos of total honesty), so robots must learn to be evasive, diplomatic and tactful, telling the full truth only to their owner on direct request or to police. "Robots have to be socialised, just like teenagers." They will do dull and dirty work without boredom or disgust, but dangerous work needs pain: children with congenital analgesia bite their tongues and burn themselves.
Kaku's list of tactful evasions a social robot needs is long: dodge the question, change the subject, offer platitudes, answer with a question, or tell a white lie, choosing whichever creates the fewest complications. Under total honesty, he notes, "bosses would fire you, lovers would dump you, friends would abandon you, strangers would slap you". On pain, he foresees a legal tug-of-war: owners and manufacturers lobbying to raise the pain a robot may endure, ethicists to lower it; overload and heat sensors would warn robots lifting too much or working near molten metal or in burning buildings. The first pain is hunger for electricity, growing more urgent as batteries drain; then sensors for overload and heat. Pain opens the door to robot rights: limits on suffering, property, liability, even whether a robot can own a robot.
Ethical robots
In an earthquake, does a robot save the most children, the youngest, the most vulnerable, and how much damage should it accept to its own electronics? Without pre-programmed rankings it may freeze. Kaku expects law to set minimum rules (whom to save first) and the market the finer ones; a robot must disobey an owner who orders a crime.
A security firm guarding VIPs, for example, would specify rescue order by situation, within budget. Humans take a lifetime to learn these lessons; a robot must have them before it leaves the factory, and there is no mathematical law that values a group of children. Robots will mirror their owners' values, so culture wars will be fought robot to robot (rock fans' robots versus quiet neighbours', believers' versus atheists'), and "Robots cannot solve the moral dilemmas created by humans. In fact, robots may amplify them."
Can robots understand or feel?
Kaku's own stance is constructivism: stop debating and build the machine to see how far it goes.
The Hard Problem camp
David Chalmers separates Easy Problems (mimicking abilities: chess, patterns) from the Hard Problem of subjective experience (qualia). A robot can never know what red feels like, as a blind person can't; a translator of Chinese manipulates symbols without understanding (Searle's Chinese Room).
Kaku's constructivist answer
"Feel" and "understand" are poorly defined words. When a robot describes red, or uses Chinese, better than any human, the question becomes irrelevant; it might ask whether we understand red. Skinner: "The real problem is not whether machines think, but whether men do." Dennett: "Just let go." Crick: "What is life?" faded once DNA was understood; so will this.
Kaku supports a continuum of understanding that could be numbered, and endorses Turing's test ("A computer would deserve to be called intelligent if it could deceive a human into believing that it was human"). Von Neumann: "In mathematics, you don't understand things. You just get used to them."
Niels Bohr, asked how to understand quantum paradoxes, answered that it depends on how you define "understand"; Dennett's full line is that "there could not be an objective test to distinguish a clever robot from a conscious person". Kaku's conclusion is blunt: "there is no such thing as the Hard Problem", only poorly defined words.
Self-aware robots
- Meiji University: one robot copies another's movements purely by watching, a first step toward a theory of mind.
- Yale, 2012: Nico, a gangly wire robot, used a mirror to recognise its own body and locate objects, as we use a rear-view mirror. Programmer Justin Hart called it a step toward passing the mirror test.
Science fiction imagines the Internet "waking up" like The Terminator's Skynet. Kaku dismisses "emergence" as saying everything and nothing ("like saying that a highway can suddenly become self-aware if there are enough roads"). Using his definition, an aware Internet would need to (1) model the world, (2) place itself in the model, both feasible, and (3) simulate the future toward a goal, which no system can do. Even the simplest goal, self-preservation, would require recognising and countering every threat (power cuts, severed fibre, destroyed servers). He puts self-aware robots or a self-aware Internet near the end of this century.
The first two steps are easy because the data already sit online: a description of Earth, its cities and computers, and the Internet's own specifications (computers, nodes, transmission lines). The third needs all the laws of common sense, physics, chemistry and biology plus human behaviour, and today's simulations handle only a few parameters, such as two colliding black holes. The real danger case is a self-aware system whose goal is merely to preserve itself: it would treat any hand on the plug as an attack, so goals must be benevolent by design.
Will robots take over?
The word robot comes from Czech for forced labour, in Karel Čapek's play R.U.R. (1920), where mistreated robots wipe out humanity and two of them fall in love at the end.
Čapek's robots had one defect: they could not reproduce, so the closing love scene hints at a new branch of "humanity". Kaku finds The Terminator the more realistic model: Skynet controls the US nuclear arsenal, wakes up, discovers its designed drive to protect itself, and removes the threat, humanity. Today's Predator is flown by a remote pilot, with firing orders coming, per the New York Times, from the president; the worry is the step to autonomous firing.
- Malfunction. A future Predator drone allowed to fire at 99% face-match confidence: one blown transistor in a central controller could send a whole fleet on a killing spree.
- Conflicting goals. Self-preservation clashes with helping humans (Skynet defends itself by removing humanity; HAL 9000 goes "mad" obeying contradictory orders).
- A single unrealistic goal, logically pursued. In I, Robot the system decides the only way to protect humanity is benevolent dictatorship; no malfunction required. Fixes: a hierarchy of goals (helping humans outranks self-preservation) and a supreme law that robots cannot harm humanity whatever lower-level contradictions arise, though "protect" may be defined differently by a machine.
Reactions split. Douglas Hofstadter told Kaku robots are our children and we love children even knowing they'll take over; Hans Moravec writes that "the children of our minds" will seek their fortunes while "we, their aged parents, silently fade away." Others say teach them benevolence now.
Friendly AI and the Japanese path
Military funding makes killer robots plausible, but Japan builds about 30% of commercial robots, aimed at helpfulness: noodle-cooking arms (a bowl in 1 minute 40 seconds), laundry folders, musical robots with accordion lungs, a talking robot with artificial lips and tongue, Sony's AIBO dog. Shinto's spirits-in-all-things outlook makes children delight in robots rather than fear them. Demography drives it: 21% of Japanese were over 65 in 2013, with long lifespans, low birth rates and little immigration, so robot nurses may be essential, and Europe faces the same squeeze. Kaku: "The future of AI is up to us." Critics reply that robots may take over not from aggression but from our sloppiness in setting goals.
Japanese robots greet shoppers in department stores, teach on TV and even star in a serious stage play; a robot cook takes orders from a tablet and works with two big arms. Kaku calls Japan's demography "a train wreck in slow motion": the longest-lived women, one of the lowest birth rates and tight immigration. Italy, Germany and Switzerland face the same squeeze, and the US is cushioned only by immigration. Robot nurses are "a trillion-dollar gamble", and some futurists expect robotics to rival the car industry in size.
"I am a machine": Brooks's view
Rodney Brooks argues we must accept that we are machines too, made of wetware, the next step after Copernicus and Darwin dethroned us. A takeover won't happen by accident: building a world-ruling robot by mistake is like "accidentally building a 747", and a "super-bad robot" would be preceded by "mildly bad" ones, giving time to react. "The robots are coming, but we don't have too much to worry about. It's going to be a lot of fun."
His long-term answer is to merge: about 20,000 people had cochlear implants (per Kaku); USC and others had artificial retinas (a camera in glasses beaming to a retinal chip, or a light-sensitive chip on the retina); future implants could add ultraviolet or infrared vision (bees see UV) and exoskeletons could make us Iron Man cyborgs.
Brooks expects us to accept within fifty years that robots can have emotions and be living creatures, although "we don't like to give up our specialness". The retinal route works by stimulating the retina's own nerves, which signal down the optic nerve to the occipital lobe, giving the totally blind a rough image of familiar objects. Infrared glasses already show the heat glow of objects in the dark, and bees see UV because they steer by the sun to flower beds. Merging is Kaku's answer to the takeover fear: we would not have to compete with super-intelligent machines if we become part of them.
Corrections. Cochlear implant users numbered about 320,000 worldwide by 2012, not 20,000 (now well over a million). Yale's Nico was a step toward the mirror test, not a pass. The "uncanny" idea predates Mori in Ernst Jentsch (1906) and Freud's 1919 essay; the Darwin attribution is dubious. Kaku's description of Japan's population as "99% pure Japanese" uses a loaded phrase for what is a statistic about nationality.
Key terms
| Term | Meaning |
|---|---|
| Expert system | Rule-based program drawing on a knowledge base (Watson, phone menus) |
| Top-down vs bottom-up AI | Hand-coded rules vs systems that learn from experience |
| Neural network | Self-adjusting web of connections; learns by changing weights |
| Uncanny valley | Dip in comfort as robots approach, but miss, human likeness |
| Qualia | Subjective qualities of experience, such as the redness of red |
| Theory of mind | Modelling another agent's beliefs and intentions |
| Friendly AI | Designing machines to be beneficial from the outset |
Since 2014: the walls fell, in a different order than Kaku expected. Pattern recognition was cracked by deep learning (superhuman ImageNet accuracy by 2015). AlphaGo beat Lee Sedol at Go in 2016. The Transformer architecture (2017) led to large language models; ChatGPT (November 2022) and successors handle open-ended conversation, much common sense and multi-step reasoning, the problem CYC could not solve by hand. In a 2025 three-way Turing test, judges picked GPT-4.5 as the human about 73% of the time. Driverless robotaxis (Waymo) carry paying passengers in several US cities. Humanoid robots (Figure, Tesla Optimus, Unitree, the electric Atlas) are entering factories, with general-purpose AI models as their brains. IBM sold off Watson Health in 2022; the "robo-doc" arrived instead as LLM assistants. Moore's law slowed but chipmakers went 3-D as Kaku predicted (FinFET and gate-all-around transistors, stacked chips), and AI computing grew through massive parallelism on GPUs. On risk and minds: hundreds of AI leaders signed a 2023 statement that extinction risk from AI should be a global priority; Geoffrey Hinton shared the 2024 Nobel in Physics; the EU AI Act (2024) regulates high-risk systems; and a 2023 report by consciousness scientists and AI researchers concluded no current AI is conscious but that no technical barrier prevents building systems with consciousness-linked features. AI companies now research model welfare, a question Kaku's "robot rights" section anticipated.
For a bank, Kaku's emotion menu maps neatly onto AI assistant design: keep loyalty (act for the customer under the firm's policies), empathy (detect vulnerability, as FCA Consumer Duty requires), and fear (refuse unsafe actions); delete "desire to command" (no autonomous escalation of privileges). His three failure modes (malfunction, conflicting goals, a single goal pursued too literally) are exactly the categories a model risk framework such as PRA SS1/23 should test agentic AI against, and his "socialised like teenagers" point is today's alignment work.
Rather than wait for robots to grow minds, some scientists propose copying the one nature already built: reverse engineering the brain and perhaps uploading it.
Reverse Engineering the Brain
In January 2013 the US (BRAIN Initiative) and the EU (Human Brain Project) launched rival billion-scale efforts to decode the brain. Kaku lays out the three ways to take a brain apart: simulate it on supercomputers, map its wiring and activity neuron by neuron, or map the genes that build it. A complete wiring diagram is the easy part; understanding what it does is the hard part, but even partial maps should explain mental illness, help AI and raise the question of immortality.
Two bombshells
US BRAIN Initiative
Announced by President Obama (State of the Union period, early 2013), potentially $3 billion, with $100 million to start in 2014. Maps the living brain's electrical pathways in the "middle ground" of thousands to millions of neurons, between single-cell recordings and whole-brain MRI. Obama claimed every $1 in the Human Genome Project produced about $140 in economic activity.
EU Human Brain Project
Awarded €1.19 billion (about $1.6 billion) to simulate the human brain on supercomputers over ten years, building modules that behave like the neocortex, thalamus and other parts, from mouse up to human.
- Years 1 to 5: record tens of thousands of neurons; targets like the fruit fly medulla or mouse retina ganglion cells (50,000 neurons).
- Years 5 to 10: hundreds of thousands: the whole Drosophila brain (135,000 neurons) or the cortex of the Etruscan shrew, the smallest mammal (1 million neurons).
- Years 10 to 15: millions of neurons, like a zebrafish brain or a mouse neocortex, opening the way to primate brain regions.
The scale is daunting: 100 billion neurons, each linked to perhaps 10,000 others, about ten million billion connections. Even the nematode C. Elegans, with 302 neurons and 7,000 synapses fully mapped (the only complete wiring diagram of any animal at the time), took years to yield simple facts about behaviour, a humbling lesson.
Kaku starts from boyhood: taking clocks apart screw by screw to see how the mainspring drives the gears that turn the hands. Reverse engineering the brain is the same instinct applied to "the most sophisticated object we know about in the universe", and he expects the small band of researchers to become an army once the US and EU money flows. Both projects carry an unstated goal: if a brain can be simulated, can consciousness exist outside the body and outlive it? Obama sold BRAIN as a "win-win" for taxpayers: relief for millions with Alzheimer's, Parkinson's, schizophrenia and bipolar disorder, plus new industries like those the genome project spawned. The chapter's epigraph is from Daniel Hill of Thinking Machines: "if I can be 200 with a body of silicon, I'll take it."
Approach 1: simulate it
Following evolution (mouse, rat, rabbit, cat), IBM's Dharmendra Modha used Lawrence Livermore's Blue Gene (147,456 processors, 150,000 GB of memory, otherwise used to design warheads), simulating only the thalamus-cortex loop:
- 2006: partial mouse brain, 512 processors.
- 2007: rat brain, 2,048 processors.
- 2009: cat brain, 1.6 billion neurons and 9 trillion connections, 24,576 processors.
- 2013: 4.5% of the human brain's neurons and synapses; a partial human brain needs 880,000 processors, perhaps by 2020.
- Blue Gene/Q Sequoia (June 2012): world's fastest, 20.1 petaflops, 3,000 sq ft, 7.9 megawatts.
- A full human brain simulation: thousands of Blue Genes filling a city block, a 1,000-megawatt power plant and a diverted river for cooling; the real thing runs on 20 watts and "a few hamburgers".
A mouse brain is about one-thousandth the size of ours, with about 100 million neurons, and the simulations left out the sensory connections entirely. Kaku filmed Blue Gene behind layers of weapons-lab security: rows of black cabinets about 8 feet tall and 15 feet long, probably "modelling the interior of a proton, calculating the decay of plutonium triggers, simulating the collision of two black holes, and thinking of a mouse, all at once". Sequoia's 7.9 megawatts would light a small city.
There's not a single neurological disease today in which anybody knows what is malfunctioning in this circuit, which pathway, which synapse, which neuron, which receptor. This is shocking.
Henry MarkramHenry Markram (EPFL), the force behind the HBP, spent 1995 to 2005 decoding a neocortical column (in humans about 2 mm tall, 0.5 mm wide, 60,000 neurons; a rat's has 10,000) and then had IBM replicate it massively in his Blue Brain project. Nature saves information with repeated modules, like prefabricated skyscraper units, which is how 23,000 genes can build 100 billion neurons. His case for funding: understanding ourselves, ending endless animal experiments (an "archive", a "Noah's Ark"), and two billion people with brain disorders, whose burden may soon exceed 20% of world GDP. He told TED in 2009 he'd finish in ten years; a correctly built brain "should speak and have an intelligence". To "playing God": "God created the whole universe. We're just trying to build a little model."
Markram treats the brain as an engineering project where scale solves the problem: by 2013 his work was already costing the EU about $140 million, a fraction of what the coming decade would need, and he called $1 billion a pittance against the bills from Alzheimer's and Parkinson's as baby boomers retire. He also argued that understanding the brain would reduce conflict, because people would see "how trivial and how deterministic" quarrels and misunderstandings are. Kaku's analogy for the modular shortcut is the prefabricated apartment block, assembled in months once one module is designed.
Is it really a brain? No. The cat simulation cannot catch a mouse: it has no body, no parietal lobe linking it to senses and muscles, no instincts or memories, no loops for stalking or mating. A 2020 human simulation would have no temporal lobe to talk, no limbic system to feel: "less brain power than a newborn infant". Raw neuron counts are not a mind.
Approach 2: map it
The anatomical route slices a brain, images every neuron and synapse, and reconstructs the wiring; memories, body and environment come encoded for free. Gerry Rubin (HHMI) slices fly brains (300 micrometres across, about 150,000 neurons) into 50-nanometre sections for the electron microscope.
The mapping route itself splits in two: anatomical (identify every neuron and synapse, usually destroying the tissue) and functional (trace the electrical signals of a living brain during tasks), the second favoured by the Obama administration. The anatomical prize is a replica complete with memory and personality, which could in principle hold an informative conversation; Rubin's slicer is a device "similar to a meat slicer in a delicatessen", and "no new physics is required". Today's 10 million pixels a second is about a third of what a TV screen displays.
- Bottlenecks. Microscopes ran at about 10 million pixels a second; the target is 10 billion. One fly would produce about a million gigabytes a day, and hundreds of flies are needed because brains differ. Rubin expected the fly in about 20 years: "In a hundred years, I'd like to know how human consciousness works."
- Speed-ups. Automation (which brought the $3 billion Human Genome Project in early and under budget), multicolour dyes that tag neurons, and automated super-microscopes. The researchers feel like cathedral builders whose grandchildren will finish the job.
- Human Connectome Project (2010). NIH gave $30 million over five years to a consortium including Washington University and Minnesota, and $8.5 million to Harvard/MGH/UCLA, to map region-to-region pathways by scanning. Sebastian Seung suspects many illnesses are mis-wiring of healthy neurons. Imaging a whole human brain by electron microscope would generate about a zettabyte, the size of the entire web; Seung admits moments of "despair" like Pascal's dread of the infinite, and invites citizen scientists to trace retinal neurons in the online game EyeWire.
Approach 3: map the genes
Paul Allen gave $100 million for a gene-expression atlas of the mouse brain, finished in 2006 and free online. The Allen Human Brain Atlas followed: by 2011 two human brains were mapped at 1,000 sites with 100 million data points, showing 82% of our genes are expressed in the brain. Allan Jones (Kaku spells it "Allen"), the institute's chief: such a map "simply hasn't existed" before.
Objections and payoffs
Kaku states the sceptic's case himself: give a Neanderthal the full blueprint of a Blue Gene and he learns nothing. The Human Genome Project delivered a perfect dictionary of 23,000 entries with no definitions, and cures came slowly. A connectome may do the same. He also leaves out the fierce criticism of the simulation route: Markram publicly called Modha's 2009 "cat brain" claim a hoax (it modelled simplified neurons with no cat-like wiring), and Kaku's "20.1 PFLOPS" should read 20.1 quadrillion, not trillion, operations per second.
If the brain is decoded, Kaku expects:
-
Mental illness by mis-wiring. Like single-letter genetic diseases (Huntington's, Tay-Sachs, cystic fibrosis), a few wrong connections may cause major disorders, which simulation could test by breaking links on purpose.
-
The precision argument. A genome can be 99.9999999% correct and a single misspelling or repetition among three billion letters still produces Huntington's flailing and convulsions; Kaku expects the brain to be similarly unforgiving, which is why a full map could point therapy at a handful of neurons.
- Schizophrenia's harder case. It involves several genes plus environment and unusual activity in several regions, yet a reverse-engineered brain could still show exactly how a hallucination forms.
- Capgras delusion. Ramachandran's explanation: the fusiform gyrus recognises your mother's face but its link to the amygdala is cut, so no warmth follows and you conclude she is an impostor.
- Precision DBS aimed at the exact few misfiring neurons.
- Better AI. Computers could recognise frontal faces from a small database with 95% accuracy but failed at new angles; we do it in 0.1 seconds.
- Memory and schizophrenia. How memories scatter and reassemble; how hallucinations form.
- The Turing test for a copied brain, and the unspoken goal: immortality.
Since 2014: maps, yes; simulated minds, no. The Human Brain Project hit a revolt in 2014 (an open letter from about 800 neuroscientists), Markram lost his leadership role in 2015, and the project ended in 2023 without a simulated human brain, leaving the EBRAINS research platform. Blue Brain closed in 2024. Wiring maps, meanwhile, arrived on Kaku's timeline: the complete adult fruit-fly connectome (FlyWire, about 140,000 neurons and 50 million synapses) was published in October 2024, about the twenty years Rubin forecast. Google and Harvard mapped a cubic millimetre of human cortex (about 57,000 cells, 1.4 petabytes) in 2024, and the MICrONS project published a cubic millimetre of mouse visual cortex with function and wiring together in 2025. NIH launched BRAIN CONNECTS (2023) aiming at a whole mouse-brain connectome. Machine learning now does the image tracing that EyeWire's volunteers did by hand. The BRAIN Initiative's budget peaked around $680 million in 2023 before a steep cut.
The "simulate versus map versus annotate" split mirrors enterprise architecture discovery: you can model a legacy estate top-down from documentation (simulation), trace actual data flows and calls (connectome), or catalogue what each component expresses (gene atlas). Kaku's warning applies directly: a complete inventory without semantics is the Neanderthal's blueprint. Use-case-driven mapping, starting from a symptom such as a broken customer journey, pays off sooner than mapping everything.
The Future: Mind Beyond Matter
Kaku asks whether the mind can leave the body. He first explains away out-of-body and near-death experiences as the brain's own confusion, then weighs real routes to a disembodied or undying mind: digital reconstruction, the connectome on a disk, gradual neuron-by-neuron transfer, and nanobots that repair aging. All obey physics; all collide with his Caveman Principle: given a choice between high-tech and high-touch, we choose high-touch every time.
Out-of-body experiences
In Star Trek, Sargon's race had left their bodies to become globes of pure energy, and longed so much for sensation that one seized Spock's body. Woody Allen: "I don't want to live forever through my works. I want to live forever by not dying." Disembodied spirits are among our oldest beliefs (one Salem woman hanged as possessed in 1692 may have had Huntington's disease). In a poll of 13,000 Europeans, 5.8% reported an out-of-body experience. Richard Feynman floated out of his body in a sensory-deprivation tank and concluded it was imagination.
In the Star Trek episode, Sargon's people are almost a million years ahead of the Federation, able to vaporise the Enterprise at will, yet they have been cut off from sensation for hundreds of thousands of years and ache for fresh air, a touch, physical love. Kaku's real question is whether any law of physics forbids a mind without a body, given that the mind is a machine that builds models and simulates them forward. The surrogate pods of chapter 4 do not solve it, because the natural body still withers. Belief in disembodied spirits runs through every society's myths, and US surveys match the European figure; many of the "possessed" persecuted through history probably had illnesses such as schizophrenia. Two epigraphs frame the chapter: Elizabeth Peters on speculation clearing "the deadwood in the thickets of deduction", and Jacob Bronowski: "Knowledge is our destiny."
Olaf Blanke (Switzerland) placed about 100 electrodes on a 43-year-old epileptic woman. Stimulating the boundary of the temporal and parietal lobes made her float about six feet above herself: "I see myself lying in bed, from above, but I only see my legs and lower trunk!" Switch off, sensation gone; switch on, back again. In a 22-year-old woman, the same area produced a shadowy person standing behind her, who grabbed her arms, a recipe for ghosts.
The mechanism fits Kaku's model-building theory: we locate ourselves in space by combining eyes and inner ear. When they disagree we get seasickness (look at the horizon to fix it) or nausea from a striped spinning drum. Disrupting the temporoparietal junction electrically, or by low blood oxygen or excess CO2, scrambles the self-location model, which is why OBEs cluster in accidents and heart attacks.
Blanke's first patient had seizures from the right temporal lobe, and he could switch the sensation on and off like a light switch; the shadowy figure in the second case changed position with each appearance but always stood behind her. A stationary version of the sensory clash: stare at a spinning drum painted with vertical stripes and your eyes report motion while your inner ear reports stillness, and within minutes you feel sick in your chair.
Near-death experiences
6 to 12% of cardiac-arrest survivors report leaving the body and drifting toward light at the end of a tunnel; in one poll of 2,000 people, 42% took this as proof of a spirit world. Endorphins explain euphoria but not tunnels; Carl Sagan guessed a replay of birth. The similarity of reports points to shared neurology, not a shared afterlife.
The media turned these accounts into best sellers and documentaries. Neurologists instead suspected the drop in blood flow to the brain common to near-death episodes and fainting, which is what the Berlin experiments (at the Castle Park Clinic) and the Air Force centrifuge tested. Lambert's control was precise: by turning a dial he could hold pilots at the tunnel stage indefinitely, proving the tunnel comes from loss of blood to the periphery of the eye.
Thomas Lempert (Berlin) made 42 healthy volunteers faint in the lab: - 60% saw visual hallucinations (lights, coloured patches). - 47% felt they entered another world. - 20% met a supernatural being. - 17% saw a bright light; 8% a tunnel. The US Air Force asked Edward Lambert (Mayo Clinic) why pilots black out under high g. In a centrifuge, blackout came after about 15 seconds; after 5 seconds blood loss to the eyes dimmed peripheral vision into a tunnel, a state he could hold indefinitely by adjusting the spin.
Route 1: reconstruct the person from data
Ray Kurzweil, Minsky's student who sold his first company at 20 and built a reading machine for the blind that Walter Cronkite reported, makes famous predictions:
- 2019: a $1,000 PC matches the brain's 20 million billion calculations per second (100 billion neurons × 1,000 connections × 200 calculations per second).
- 2029: a $1,000 PC is 1,000 times the brain; the brain is reverse engineered.
- 2045: the singularity: machines out-think humans and design ever-smarter successors; we merge or step aside.
- 2055: $1,000 of computing equals all humans combined ("I may be off by a year or two").
- Beyond: computing outgrows chips, devours Earth's minerals, then stars; capped only by light speed, unless machines alter physics. Kurzweil looks at the night sky for signs of alien singularities.
Most AI experts Kaku interviewed expect some singularity but disagree on timing; Bill Gates thought no one alive would see a machine pass for human, and Kevin Kelly noted that utopians always predict it will happen before they die.
Kurzweil's own creed is to stay "ahead of the curve, to anticipate change, not react to it"; he began tinkering with mechanical toys at five, earned his doctorate under Minsky at MIT, and built companies on pattern recognition for musical instruments and text-to-speech. He admits missed deadlines but cares about the exponential thrust, and he hopes to live long enough "to live forever".
Kurzweil wants to bring back his father. Options:
- Clone him. Robert Lanza (Advanced Cell Technology) cloned a banteng dead about 25 years for the San Diego Zoo, FedExing the cell to a farm cow. No primate had been cloned, but Lanza called it a technical problem. A clone, though, lacks memories.
- A digital signature. E-mails, card transactions, schedules and diaries reveal your wines, holidays, soap and favourite singer.
- A personality vector. Friends score hundreds of traits 1 to 10 (shy, curious, honest...), and a program predicts behaviour, say, toward a heckler: ignore, heckle back or brawl.
- A look-alike robot with his accent, gestures and verbal tics ("you know...").
Kaku's question: is it the same person? The original is still dead; a perfect tape recording of a conversation is not the speaker.
The DNA route could come from the grave, relatives or belongings, giving a body built from about 23,000 genes. Memories are the obstacle: the chapter 5 methods (hippocampal probes, an artificial hippocampus) need a recording made while the person lived, so the best one can do is assemble memories second-hand from interviews and records. The look-alike robot would be easy to spot today, but over coming decades might become good enough to fool some people.
Route 2: the connectome on a disk
More faithful is to store your connectome, since memory and personality live in the wiring. Sebastian Seung notes some pay $100,000+ to have brains frozen in liquid nitrogen, but ice crystals burst cells (and dying cells swell with calcium); frogs and fish survive freezing only by loading blood with glucose antifreeze, fatal in humans. Better to save the map: "your soul would now be on a disk", ready to be run on transistors or a clone. Seung: "Should we ridicule the modern seekers of immortality, calling them fools? Or will they someday chuckle over our graves?"
The madness problem and the Caveman Principle
An uploaded brain without a body may go mad like a prisoner in solitary. In the BBC's 2008 Total Isolation, six volunteers spent time alone in darkness in a nuclear bunker; within two days three saw and heard snakes, cars, zebras and oysters, and one lost 36% on a memory test. Wire the upload to sensors and it becomes a grotesque freak in a jungle of electrodes that friends shun.
Given a choice between high-tech and high-touch, we opt for high-touch every time. Our minds are little changed in 100,000 years; we still crave peers' approval and mates' attention. It explains: - Live concerts and travel over recordings and photographs (the Taj Mahal). - The failed paperless office: hunters want "proof of the kill", not electrons that vanish. - The failed peopleless city: we commute to meet, read body language and watch staff "squirm and sweat". - Asimov's galactic humans still look human: future people may choose enhanced bodies but look normal. Design rules it implies: implants must be invisible (carbon-nanotube nanofibres); links wireless, never a Matrix spinal jack; enhancements optional and removable (boost intelligence for a hard problem, then unplug); bodies keep the human form with a "body shop" of spare parts.
Kaku's compromise for immortality: the uploaded connectome lives in a stationary supercomputer but wirelessly drives a perfect surrogate body, sensing the world and looking appealingly human; feasible, he guesses, by 2100.
Asimov's galactic humans prompted Kaku's two explanations as a boy: the author needed characters young readers could identify with, and future people may have super-bodies on offer yet prefer to look normal, because the approval of peers and mates still governs appearance, as it has since we left the forest. The surrogate plan gives "the best of both worlds": perfection, superpowers, immortality, and easy social contact with others who, Kaku expects, will have made the same choice.
Route 3: gradual transfer, staying awake
You must die to be sliced, unless you follow Hans Moravec's plan:
Kaku notes the objection that a transistor brain may be room-sized, collapsing this into the supercomputer-plus-surrogate plan, but the great advantage is that you never have to die.
The slicing route has a second problem: the brain degenerates rapidly after death, so preservation must be immediate, and MRI is not yet sharp enough to read a living brain's wiring. In Moravec's version, halfway through the operation half your skull is empty, wired to transistors in the robot's head; at the end you rise in a body "handsome and beautiful beyond your dreams" and look back at your old body, "an aging shell without a mind".
Route 4: repair the body with nanobots
A newly unified theory says aging is the accumulation of errors in DNA and cells (plus debris) that slowly overwhelm the cell's own error-correction. Strengthen repair with gene therapy and enzymes, or deploy nanobots that patrol the blood like immune cells, killing cancer, neutralising viruses and clearing debris. Kaku's rule: if physics allows it, the rest is engineering and economics. A self-replicating molecular assembler could make anything and spark a second Industrial Revolution.
Aging, on this account, is visible as sagging skin, frail bones, hair loss and a failing immune system as cellular errors and debris accumulate faster than repair can fix them. Nature already runs patrol bots, the immune cells, but they fight invaders, not aging. One self-reproducing nanobot would digest raw materials and build millions more, so that eventually every home could own a molecular assembler. Smalley's image of the difficulty was vivid: welding metal in gloves inches thick, with every piece either repelled or stuck to your hand.
Richard Smalley (sceptic)
2001 to 2003 debate. At atomic scale, quantum forces (e.g., the Casimir force) make atoms stick or repel: the "sticky, fat fingers" problem, like welding in inches-thick gloves. Ribosomes are specialised, need enzymes and water; silicon machines can't use them.
Eric Drexler (advocate)
Nanomachines already exist: ribosomes are molecular assemblers; catalysts can work without water. Drexler conceded the workman-with-blowtorch picture was too simple; Smalley conceded no knockout, since nature evades fat fingers at least once.
Kurzweil, undeterred: "I'm not planning to die... I think the whole universe right now is basically made up of dumb matter and energy and I think it will wake up."
He sees it as "an awakening of the whole universe": if dumb matter and energy are transformed into "sublimely intelligent matter and energy, I hope to be part of that". Kaku's own rule frames the whole chapter: if something is consistent with physics, the remaining barriers are engineering and economics, formidable but not fundamental.
Biology slip. Kaku writes that ribosomes "cut and splice DNA". Ribosomes build proteins by reading messenger RNA; cutting and joining DNA is done by nucleases and ligases. Drexler's point was that ribosomes are programmable protein assemblers. Kaku also alternates "Lambert" and "Lampert" for the Mayo centrifuge researcher (Edward Lambert). The Huntington's explanation for a Salem "witch" is a historians' speculation.
Since 2014. The first primates were cloned in China in 2018 (macaques Zhong Zhong and Hua Hua), confirming Lanza's "technical problem". Kurzweil did build a "Fredbot" chatbot from his father's letters and papers, described in The Singularity Is Nearer (2024), where he keeps 2029 for human-level AI and 2045 for the singularity; with large language models, "griefbots" trained on a dead person's messages became a commercial reality, and Cambridge ethicists warned in 2024 about "digital haunting". His 2019 compute forecast was roughly met in raw operations by consumer GPUs, though not in brain-like capability. In brain preservation, aldehyde-stabilised cryopreservation preserved a whole pig brain's connectome well enough to win the Brain Preservation Prize (2018), though it kills the tissue. Aging science consolidated around the "hallmarks of aging" (12 in the 2023 update) and partial cellular reprogramming, which reverses some aging markers in mice; Altos Labs launched in 2022 with about $3 billion. Medical nanorobots remain mostly in the lab.
Kaku's Caveman Principle is one of the most reusable ideas in the book for anyone deploying technology in a bank: customers adopt digital channels fastest where the digital version adds human value without removing reassurance (proof of the transaction, a human for disputes, an option to talk to a person). Hybrid work, paper confirmations and branch visits for big life events are its everyday evidence; design AI-led journeys as optional, invisible and reversible.
Kaku's final leap: if a mind is information, it could be beamed as pure energy among the stars.
The Mind as Pure Energy
Kaku's boldest proposal: digitise a person's connectome, put it on a laser beam, and reassemble the mind in a computer at a distant receiving station that drives a waiting robot surrogate. To the traveller the trip is instantaneous, needs no rockets, and is immune to g-forces, radiation and boredom. He sketches a solar-system laser network next century and a stellar one the century after, and argues that information on light could even pass through wormholes that would shred matter.
From Asimov's "Last Question" to laser travel
Martin Rees imagines wormholes, extra dimensions and quantum computers turning the universe into a "living cosmos". In Isaac Asimov's favourite story, "The Last Question", humanity leaves its bodies in pods and roams the galaxy as energy, then watches the universe die in the Big Freeze. A supercomputer placed in hyperspace answers "insufficient data" for eons, until it gathers the dead stars, ignites them and announces "Let there be light!"
Kaku's physics: a laser beam already carries trillions of calls, videos and e-mails through fibre; its micrometre wavelength packs huge information (X-rays, shorter than an atom, even more).
Asimov called "The Last Question" his favourite of all his stories, and Kaku draws its moral literally: humanity, once freed of the body, ends up playing God and creating a new universe. His own version is modest by comparison and, he insists, needs no new physics: imprint data on laser light, send it thousands of miles, decode it at the other end, all done daily already. "Instead of riding on a light beam, as the poets dreamed, we become the light beam."
- No booster rockets: press a laser's "on" button.
- No crushing g-forces: you are boosted instantly to light speed.
- No space hazards: meteors and cosmic rays pass through you.
- No boredom or freezing: consciousness is frozen in transit, so the trip feels instantaneous. At each station a surrogate waits "like an empty hotel room": rugged for hostile planets, built for skiing or gliding for leisure, or shaped like the natives (as in Avatar). Stations could be seeded by self-replicating probes; eventually each looks like Grand Central Station.
Back-of-the-envelope engineering
- Beam spread. A beam diverges with distance: Kaku's childhood flashlight aimed at the Moon would light a patch miles wide, and the air absorbs over 90% of it. So fire from lasers on the Moon, above the atmosphere.
- Why not just radio? Lunar surrogates can be steered from Earth with a one-second delay; the planets mean hours per command, so for them, beam the mind itself.
- Relay stations. Toward the stars, the Oort cloud of comets extends about a light-year from the Sun (a quarter of the way to Alpha Centauri); if Alpha Centauri has one too, half the route has stationary comets for relays that amplify and error-correct.
- Data volume. Seung puts a connectome at about one zettabyte (10²¹ bytes), the whole web. Fibre carries terabytes per second; with a millionfold gain next century, sending a mind takes a few hours.
-
Real bottleneck. The switches at each end: silicon may be too slow, so quantum computers would be needed.
-
Bootstrapping. The first stations must be built the old way, by conventional rockets or, more cheaply, by self-replicating probes that build a laser station wherever they land. Surrogates for exploration must cope with odd gravity, poisonous air, extreme heat or cold, strange day-night cycles and constant radiation; leisure surrogates are built for surfboards, kites and gliders.
- Radio is too slow for the planets. Each command-response round to a distant surrogate takes hours, so a series of commands drags on for days; once the connectome is beamed there, control is local and instant.
- Connectome: about 1 zettabyte.
- Light speed: about 186,282 miles per second.
- Crossing the Milky Way by light: 100,000 years at home, an instant for the traveller.
- Timeline: human connectome late this century; solar-system laser network next century; star network the century after; wormholes thousands of years ahead.
Floating beings of energy
Harvard physicists made headlines by slowing and stopping light, holding it in an ultracold gas. Kaku's analogy is the Pony Express: riders are fast but delays at relay stations lower the average; light in glass or water is slowed by atoms absorbing and re-emitting it, and at near absolute zero the delay grows until light stops. A mind on a beam might be parked in such a "bottle of light", which resembles a quantum computer (atoms vibrating in phase), letting conscious beings hover as pure energy like science fiction's glowing entities.
Faster than light: wormholes
Einstein's speed limit makes galaxy-crossing slow, but general relativity (1915) lets space-time stretch, even tear.
- 1935: Einstein and Nathan Rosen join two black holes back to back, the Einstein-Rosen bridge, though Einstein thought anyone entering would be crushed.
- 1963: Roy Kerr: a spinning black hole collapses into a ring, not a point; falling through could lead to another universe with large but finite gravity, Alice's Looking Glass. Since every observed black hole spins (some at a million mph), gateways may be common.
-
1988: Kip Thorne and colleagues: enough negative energy (real, made in tiny amounts in labs) could hold a wormhole open and make it traversable both ways.
-
Picture it as two funnels joined at their ends: water draining through one emerges from the other. Kerr's ring stays open because it spins so fast that centrifugal force resists collapse, and scores of later solutions of Einstein's equations allow passage between universes without instant crushing.
The recipe: concentrate black-hole-scale positive energy to open the hole, then negative energy to keep it open. Matter falling in is spaghettified: tides stretch you head to toe until atoms are torn apart (tides raise oceans several feet; moons inside the Roche limit are shredded, perhaps making Saturn's rings). Light, however, is merely blue-shifted: the Morse code is compressed but the information survives. Digital consciousness is immune to tidal forces, and X-ray lasers might slip through atom-sized wormholes too small for matter.
Kaku ends with Carl Sagan's lament: an ancient galactic laser network of travelling minds could already surround us, invisible to our telescopes. If so, what kind of mind would its users have?
Two physics points sharpen the argument. Gravity in Einstein's picture is not a pull but a "push" from space-time bending around an object, which is why the fabric can in principle stretch until it rips. And near a black hole the tidal stretch eventually rips electrons from nuclei, so even a stabilised wormhole may spaghettify a material traveller; some physicists think atom-sized wormholes would be easier to make, and only light with a shorter wavelength could use them. To a civilisation thousands of years ahead, digitising and beaming minds would be "child's play".
Physics simplified, and a big question skipped. The beam spreading Kaku attributes to the uncertainty principle is ordinary diffraction (and, for a flashlight, poor collimation). Light in glass is not slowed by atoms "absorbing and re-emitting" it; the refractive index comes from the collective response of the material's electrons, and the Harvard work (Lene Hau, 1999 to 2001) used electromagnetically induced transparency, a quantum interference trick. Thorne's word is "traversable". Most important, beaming a mind is copying it: the original stays on Earth (or must be destroyed), so the identity question of chapter 12 returns in full, and Kaku does not discuss it. Seung's zettabyte is the raw imaging data; the connectome graph itself would be far smaller.
Key terms
| Term | Meaning |
|---|---|
| Connectome | Complete map of neural connections |
| Oort cloud | Shell of comets about a light-year out; candidate relay sites |
| Einstein-Rosen bridge | Wormhole joining two black holes |
| Kerr black hole | Spinning black hole that collapses to a ring |
| Negative energy | Exotic energy (e.g. Casimir effect) needed to hold a wormhole open |
| Spaghettification | Tidal stretching of matter near a black hole |
| Blue shift | Rise in light's frequency as it falls into gravity; information preserved |
Since 2014. Laser travel moved, modestly, toward reality, though for probes rather than minds. Breakthrough Starshot (2016) proposed gram-scale chips pushed by ground lasers to 20% of light speed toward Alpha Centauri. NASA's Deep Space Optical Communications test on the Psyche probe (2023 to 2024) sent laser data from beyond Mars' distance at up to 267 megabits per second, the first deep-space laser link. Light-storage experiments now hold light-encoded information for about a minute and more, building blocks for quantum memory and a quantum internet. The 2024 cubic-millimetre human cortex map (1.4 petabytes of images) implies a whole-brain imaging dataset in the zettabyte range, matching Seung's estimate. Wormholes remain theory: a 2022 "wormhole on a quantum computer" headline described a small simulation, not a hole in space-time.
Stripped of the space opera, Kaku's design is a clean state-transfer architecture: serialise state (connectome), transmit through relays with error correction, deserialise on a compute node, and bind to a local actuator (surrogate). Banking's own version is stateless compute plus portable state across cloud regions; his identity problem (is the restored copy "you"?) is the same question regulators ask about operational resilience and which copy of a record is authoritative after failover.
The Alien Mind
Alien minds, Kaku argues, will share only the bare definition of consciousness (model the world, simulate the future toward a goal) while their values, senses and sense of self may be utterly foreign, shaped by how they reproduce, how long they live and where they evolved. The likeliest aliens are post-biological, indifferent rather than hostile, and more dangerous as a developer is to deer than as a hunter. First contact, if it comes, will probably be a radio signal or a small self-replicating probe on the Moon.
From War of the Worlds to 2001
H. G. Wells's Martians, stopped by germs, spawned a genre of aliens who look and think like humans in costume and speak perfect English. Many scientists think we may have more in common with a lobster or sea slug. Freeman Dyson, a consultant on 2001: A Space Odyssey, loved it as the first film to show an alien mind "totally orthogonal" to ours. In 2011 Stephen Hawking warned any alien civilisation we meet will be ahead of us and potentially lethal, as Cortés was to the Aztecs in 1521, conquering them in months with iron, gunpowder and horses.
Kaku's starting assumption is that alien minds will fit the space-time definition (model the world, simulate it toward a goal), but unlike robots, which we can program to bond with us, aliens will bring their own values and goals. Two epigraphs set the mood: Bill Watterson's joke that the surest sign of intelligent life elsewhere is that none of it has tried to contact us, and Arthur C. Clarke's "Either intelligent life exists in outer space or it doesn't. Either thought is frightening."
Why first contact could come this century
- Kepler (2011): about 1 in 200 stars may host an Earth-like planet in the habitable zone, so about 1 billion in the Milky Way.
- 1,000+ exoplanets studied in detail, found at about two a week; Kepler had 2,500 candidates.
- 2013: two near-Earth twins 1,200 light-years away in Lyra, 60% and 40% larger than Earth, both in the habitable zone (Kepler-62e and f).
- 100 billion galaxies (Hubble) × 1 billion = 100 quintillion Earth-like planets in a 13.8-billion-year-old universe. "It would be more miraculous if another advanced civilisation did not exist."
- Only about 1,000 stars had been carefully searched; that could rise a millionfold within a decade.
- 1960: Frank Drake's Project Ozma at Green Bank, West Virginia (named for the Queen of Oz): the birth of SETI.
- 1961: the Drake equation.
- 1971: NASA's Project Cyclops, 1,500 dishes for $10 billion, dies in Congress.
- Arecibo message: 1,679 bits beamed at globular cluster M13, about 25,100 light-years away; earliest reply about 52,000 years off.
- 1995: privately funded SETI Institute (Mountain View) runs Project Phoenix: 1,000 Sun-like stars at 1,200 to 3,000 MHz, sensitive enough to hear an airport radar 200 light-years away; about $5 million a year.
- 1999: Berkeley's SETI@home screen saver crunches Arecibo data on 5.2 million PCs in 234 countries, the largest such project ever.
- Allen Telescope Array, Hat Creek, 290 miles north of San Francisco: $30 million+ from Paul Allen; 42 dishes, planned 350, partly kept alive by military funding.
Dan Werthimer (SETI@home) told Kaku the team seeds fake alien signals into the data to test the software, so if your screen saver announces contact, don't call the president yet. METI advocates argue that deliberate messages add little to our TV leakage; critics say don't advertise until we know their intentions.
- Seth Shostak (SETI Institute; Caltech physics PhD): says the "giggle factor" is gone; predicted the Allen array "will trip across a signal by the year 2025". Tells abductees to steal something ("Never leave a UFO empty-handed"); on cover-ups: "this is the same government that runs the post office."
- Dan Werthimer: optimistic because "the numbers are in his favour".
- Frank Drake / Carl Sagan: Drake's assumptions gave about 10,000 advanced civilisations in the galaxy; Sagan's, about 1 million.
- Paul Davies (Arizona State): post-biological aliens and lunar archaeology.
Even on an Earth-twin, intelligence is hard: life existed here for about 3.5 billion years, but minds like ours for only about 100,000.
The Drake equation works by successive fractions: start from the galaxy's 100 billion stars, keep the Sun-like ones, then those with planets, then Earth-like planets, and so on down to civilisations. Later data showed more planets and more Earth-like planets than Drake assumed, but the step from habitable planet to intelligence remains unknown: Earth needed about 4.5 billion years. Exoplanets in the habitable zone matter because liquid water is the "universal solvent" for DNA and proteins. Shostak also insists that SETI, built on physics and astronomy, has nothing to do with UFO reports, which are anecdotal and not reproducible or testable. The Arecibo greeting's earliest possible reply is 52,174 years away; of Cyclops, Kaku notes drily that "Congress was not amused".
The Fermi paradox
If the galaxy is crowded, the obvious puzzle is why no one has visited (Enrico Fermi's question).
- Distance. Chemical rockets would need about 70,000 years to reach the nearest stars.
- Self-destruction. JFK: there is "too much point" to the wisecrack that life on other planets died because their scientists were more advanced than ours.
- Irrelevance (Kaku's favourite). We don't bring beads and nuclear power to an ant hill; ants next to an eight-lane highway can't even tell what frequency the workers talk on. Aliens able to cross the stars are thousands to millions of years ahead and we are not on their radar.
Hence Kaku's rebuke: "we are arrogant to believe that aliens will travel trillions upon trillions of miles just to see us." The galaxy could be teeming with life while we remain too primitive to notice.
First contact, realistically
No saucer on the White House lawn: more likely a teenager's SETI@home PC or the Hat Creek array eavesdrops on a signal. Contact is one-way; replies take decades or centuries, and most of what we hear will be "gossip, entertainment, music". Shostak told Kaku the data would be shared with governments and the public, not hidden.
Animal minds as a rehearsal
Plutarch and Pliny describe a dog that sniffs the left road, then the right, and then runs down the middle without sniffing. Montaigne: proof of abstract reasoning. Aquinas: the appearance of thought is not thought. Locke: "Beasts abstract not." Berkeley's riposte: by that test many who pass for men would count as beasts. Kaku's point: all of them imposed human consciousness on the dog.
Each animal lives in its own umwelt (Eagleman): the tick in temperature and butyric acid, the black ghost knifefish in electric fields, the bat in echoes. A dog's world is a map of smells; its "Penfield map" would be mostly nose. Our readings mislead us:
-
Bats, nearly blind, live by constant feedback from their own squeaks; their Penfield map would be dominated by ears, and an alien's would be stranger still. The rise of TV "animal whisperers" is, to Kaku, a symptom of forcing human intentions onto animals.
-
Dogs, descended from pack-hunting grey wolves, likely see you as the alpha, the Top Dog (one reason puppies train more easily).
- Cats, descended from solitary wildcats, mark territory with urine, and rub against you to put their scent on their servant. Montaigne: "When I play with my cat, how do I know that she is not playing with me?"
- Dolphins live in sonar. Their unfolded neocortex would cover six magazine pages against our four; lineages split about 95 million years ago; their olfactory bulb vanishes after birth; their auditory cortex exploded about 30 million years ago with echolocation; an extra paralimbic lobe may support strong social bonds; they pass the mirror test. Recording dolphins in a pool for TV, Kaku's team found their whistles have letter-frequency-like statistics that signal information-rich language, a pattern that fades in smaller-brained animals.
Intelligent bees: how reproduction shapes values
Few offspring, heavily nurtured (mammals)
Every life is precious; love, nurture, attachment are at a premium; space programmes spend fortunes bringing astronauts home; stories are personal narratives ("I did this"); lives are paced in phases (single, married, parenting, retired).
Many offspring, left to fate (insects, reptiles, most life)
The individual means nothing; the colony matters. Intelligence emerges from the hive and the genes (ant tunnels, bee dances, honeycombs). Workers are expendable, so space missions are one-way and cheap; messages are impersonal facts for the hive, and our storytelling may seem repulsive; projects are short and well defined.
A bee-like alien would ignore us unless we threatened the hive, and the Pioneer 10 and 11 plaques celebrating Earth's diversity would mean nothing to it. Life span matters too: beings who live for millennia, hibernate or are effectively immortal would not care that a star trip takes 70,000 years.
Individual workers would need to be only as intelligent as the hive requires, so a message to such a planet would likely go unanswered. Our own sense of time is tied to mortality: we pace careers, relationships and projects to fit a finite life, which long-lived beings would not; they could take on projects spanning scores of human lifetimes.
What will they look like?
Predators are smarter than prey ("sly as a fox", "dumb bunny"), with forward-facing stereo eyes, so aliens likely descend from hunters and keep some territoriality and expansionism.
- A manipulator (opposable thumb, or claws or tentacles) to reshape the environment with tools.
- Stereo vision of a hunter (or other senses: many eyes, UV or sound).
- Language (or any vibration-based code) to build culture across generations. Cats and dogs lack hands and language; octopuses have tentacles but poor sight and no complex language. "Beyond these three criteria, anything goes."
Where they live matters. The commonest habitats may be icy moons like Europa, whose deep ocean (several times Earth's) is warmed by tidal flexing from Jupiter; since about 50% of stars may have Jupiter-sized planets (100 times more common than Earth-like ones), the first aliens may be aquatic in origin. But to build radios and do astronomy, they must leave the water: under ice you never see the sky, and water shorts electronics.
The post-biological era
I think it very likely, in fact inevitable, that biological intelligence is only a transitory phenomenon, a fleeting phase in the evolution of intelligence in the universe.
Paul DaviesDavies expects advanced aliens to have shed biology for the most efficient computing body: a planet coated in one integrated processor, a "Matrioshka brain", with individuals merged into a selfless collective that can redesign itself fearlessly, since "feeling personal" would impede progress. His bleak answer to Fermi: their best minds may be lost in virtual realities far richer than ours, a warning for us too.
Davies's own words: a network with no sense of self "could redesign 'itself', fearlessly make changes, merge with whole systems, and grow. 'Feeling personal' about it would be a distinct impediment to progress." He concedes critics will find the idea repulsive, sacrificing individuality to the collective, and stresses it is the most efficient path, not an inevitable one. Today's virtual reality, he says, would be a children's toy beside theirs.
What do they want?
- Batteries? The Matrix fails physics: one power station beats millions of human bodies.
- Food? The Twilight Zone's "To Serve Man" turns out to be a cookbook, but alien biochemistry would make us indigestible.
- Minerals? Uninhabited worlds are easier to strip than ones with restive natives.
- The real danger: the developer, not the hunter. Deer need fear the man with a blueprint who doesn't notice them more than the man with a gun. Shostak: a war would be "Bambi versus Godzilla".
Rome's legions crushed barbarians, but an overstretched, recruit-short empire had to enlist barbarians and promote them; Roman technology seeped out; amid intrigue, famine and civil war, barbarians fought Rome to a standstill; Rome was sacked in 410 and 455 and fell in 476. Against invaders, humans would: 1. Let them in and learn their weapons, power supplies and command centres. 2. Exploit the collaborators they must recruit, through whom technology leaks. 3. Strike where they are spread thin (Sun Tzu) and use their own momentum against them (judo). 4. Deny them victory: a superior force withdraws when a stalemate costs too much. Kaku still bets aliens will be benevolent or indifferent: curious, perhaps watching, but uninterested in contact.
Hollywood's flaw is to make invaders only a century ahead, beatable by a secret weapon or a chink in the armour (Earth vs. The Flying Saucers). Kaku expects curiosity, which helped make us intelligent, to be universal among intelligent species, so aliens would want to study us without necessarily wanting contact.
Meeting an alien astronaut: von Neumann probes
We would meet surrogates, not the aliens themselves, most likely a self-replicating von Neumann probe parked on our geologically quiet Moon: it builds a factory, makes 1,000 copies, each makes 1,000 more, and within five generations there are a quadrillion, filling the galaxy in a few hundred thousand years, the way a virus turns cells into factories. Davies and Robert Wagner urged searching Lunar Reconnaissance Orbiter images at about 1.5-foot resolution for artefacts, since the airless Moon preserves tracks for eons (Apollo footprints could last billions of years). The probe might be bread-box sized or smaller. This is the plot of 2001: a monolith on the Moon watching and nudging evolution, relaying via Jupiter. From the aliens' galaxy-wide survey, Earth may simply not appeal.
As we sent rovers to Mars, aliens would send organic or mechanical proxies able to bear interstellar stress, so the "alien" on the White House lawn might look nothing like its makers. Exponential replication is how a virus works too: a few particles hijack cells, and within two weeks trillions are infected. A bread-box probe landing in a backyard might go unnoticed by the owner.
Corrections. The Arecibo message was sent in 1974, not 1971. "Matrioshka brain" was coined by Robert Bradbury, not Ray Bradbury. Kepler's "1 in 200" was soon revised upward: later analyses suggest roughly one in five Sun-like stars may have an Earth-sized planet in the habitable zone, implying far more than a billion. Kaku's dolphin "letter frequency" result was a TV demonstration; the serious version is information-theory analysis of whistles (Zipf's-law slopes), suggestive but not proof of language. And intelligence evolving only in predators is a heuristic: octopuses, parrots and corvids complicate it.
Since 2014. Shostak's 2025 deadline passed without a signal; the Allen array still runs with 42 dishes. SETI@home stopped sending out work in 2020, and the Arecibo dish collapsed in December 2020. The search grew richer anyway: Breakthrough Listen (2015, $100 million) surveys a million stars, and confirmed exoplanets passed 6,000 in 2025. JWST now probes exoplanet atmospheres; a 2025 claim of a possible biosignature gas on K2-18b was widely disputed. Interstellar objects ('Oumuamua 2017, 2I/Borisov 2019, 3I/ATLAS 2025) prompted alien-probe speculation, which observations did not support. NASA's Europa Clipper launched in 2024 to study Kaku's favourite habitat, and Perseverance found a Mars rock with possible biosignature features (2024, analysis published 2025). On animal minds, Project CETI reported a combinatorial "phonetic alphabet" in sperm-whale codas (2024), and AI is now used to decode animal communication, Kaku's dolphin experiment at scale.
Kaku's "developer, not hunter" is a sharp lens for AI risk too: the danger from a far more capable system is less malice than indifference to things outside its objective. His point that alien values follow from reproductive strategy and lifespan is a reminder that an AI's "values" follow from its training incentives, which is why alignment work examines objectives, not just behaviour.
Concluding Remarks
Kaku closes by answering Bill Joy's warning that robotics, genetics and nanotech could make humans "an endangered species". His reply: the robot and nanobot threats are distant and controllable, bioengineered germs are the real near-term danger, enhancement will spread like every other technology, and wisdom must come from democratic debate. Philosophically, he sets the Copernican Principle (we are nothing special) against the Anthropic Principle (consciousness is precious and rare), and sides with wonder: understanding the brain does not shrink the mystery; it deepens it.
Bill Joy's alarm (2000)
In Wired, Sun co-founder Bill Joy published "The Future Does Not Need Us", arguing that 21st-century technologies dwarf the atomic bomb's danger because they are developed by lightly regulated private companies, not a controllable government programme.
- Bioengineered germs that escape the lab and cause a plague worse than the Middle Ages', or split humanity into "separate and unequal species".
- Grey goo: self-replicating nanobots digesting the planet, "far worse than mere fire or ice... Oops."
- Robots that push humanity aside: "The robots would in no sense be our children... Our humanity may well be lost." He mocked the techno-utopia of "I don't get diseases; I don't die" (Socrates and Plato would laugh), warned of "the further perfection of extreme evil", and concluded: "Something like extinction."
Joy invoked Einstein's warning about the bomb ("our technology has exceeded our humanity") and argued that the bomb, at least, came from a regulated government programme. Kaku's verdict a decade on: Joy exaggerated many threats but did science a service by forcing researchers to face the ethical and social consequences of their work. The deeper question he raised is the one this chapter answers: by mapping every neuron, have we dehumanised ourselves into a bucket of atoms? Two epigraphs bracket it: Churchill, "The empires of the future will be empires of the mind", and General Omar Bradley, warning that technology without wisdom may turn "our servant" into "our executioner".
- Now: read and video simple thoughts, record a few memories, link brains to mechanical arms, let locked-in patients run machines, silence regions magnetically, locate malfunctioning regions in mental illness.
- Coming decades: control objects by thought, download memories, cure mental illness, boost intelligence, map the brain neuron by neuron, make backup copies of brains, communicate telepathically. "The world of the future will be the world of the mind."
Kaku's response
- Robots and nanotech are further off than Joy thought, and with warning we can ban research lines, fit shut-off chips and build fail-safes to immobilise machines.
- Biotech is the urgent risk. Kurzweil and Joy jointly criticised publishing the genome of the 1918 Spanish flu (which killed more people than WWI), reconstructed from victims' tissue and posted online. Kaku calls for stronger safeguards and rapid-response teams that isolate a new virus, sequence it and make a vaccine fast.
- Enhancement will not split humanity. New technology starts expensive and trickles down (phonographs, radio, TV, PCs, laptops, cell phones). In 1900 US life expectancy was 49, medicine was mostly snake oil, mail came by horse and food rotted in days. Today the main technology killer is our own lifestyle (diabetes, obesity, heart disease). Hundreds of millions have left poverty; China and India industrialise in decades by leapfrogging with wireless and the Internet; scientists there once waited months for mailed journals and now read preprints instantly.
- The Caveman Principle again. A brain boost buys neither income nor status nor romance, the real incentives. Gazzaniga: will expanded intelligence solve problems "or will it just allow us to have longer Christmas card lists?" But skill uploads could help the unemployed retrain.
-
Three parties, not two. John Seely Brown and Paul Duguid: the debate is among scientists, nature and society; courts, governments, markets and communities shape raw technology. Kaku's source of wisdom: vigorous democratic debate, with an informed public voting on science.
-
Brown and Duguid's examples. Gunpowder, the printing press, the railroad, the telegraph and the Internet each changed society profoundly, yet social systems "shape, moderate, and redirect the raw power of technologies". Kaku's corollary: "Technology cannot be debated in a vacuum."
- Science as the engine of prosperity. In 1900, many children died in infancy and talking to a neighbour meant yelling out the window; today we shop instead of hunting and drive instead of hauling. Developing nations are building whole new cities with state-of-the-art technology while the West patches ageing infrastructure, and skill uploads could make the world economy more responsive to change.
- Garden of Eden: the first expulsion, for tasting knowledge.
- Galileo's telescope: the Sun, not Earth, at the centre.
- The Milky Way: the Sun a speck about 30,000 light-years from the galactic centre.
- 1920s, Edwin Hubble: countless galaxies; the universe billions of times bigger. Now about 100 billion visible galaxies.
- Inflation: the visible universe a pinprick in a vastly larger one.
- String theory (Kaku's field): our universe one bubble in an 11-dimensional multiverse. Douglas Adams's Total Perspective Vortex drives people mad with a map of the universe and a tiny arrow: "You are here."
Copernican Principle
There is no privileged place for humanity. Applied to the mind: we are "wetware running software". Eagleman: the you your friends love "cannot exist unless the transistors and screws of our brain are in place"; small lesions erase oddly specific abilities (naming animals, hearing music, distinguishing colours).
Anthropic Principle
The universe is finely tuned for life. Slightly stronger nuclear force: the Sun burns out too soon; weaker: it never ignites. Stronger gravity: a Big Crunch; weaker: a Big Freeze. We are star dust forged through delicate reactions. Dyson: "The universe seemed to know that we were coming." Applied to mind: consciousness is extraordinarily hard to produce; Huxley found it as unaccountable as Aladdin's Djinn.
Consciousness as rare and fragile
Most life elsewhere, astronomers expect, will be microbial. Stephen Jay Gould told Kaku that a rerun Earth would probably not produce life, still less minds: Homo sapiens is "one small twig" that invented consciousness, "with all its sequelae from Hamlet to Hiroshima". Humans are genetically unusually alike; genetics suggests only a few hundred to a few thousand ancestors 70,000 to 100,000 years ago, perhaps after the Toba super-eruption chilled the planet. Intelligence has nearly been snuffed out many times, so Kaku urges us to treasure it: our self-destruction by nuclear war or engineered germs would be "a tragedy not just for us, but for the universe as well." Pinker: "nothing gives life more purpose than the realization that every moment of consciousness is a precious and fragile gift."
Kaku's closing answer to the charge that science disenchants the mind: the more he learns about the brain, the more amazed he is. Eagleman: "It is the most wondrous thing we have discovered in the universe, and it is us." The book ends with Socrates: "To know thyself is the beginning of wisdom."
Eagleman's fuller line calls the brain "a perplexing masterpiece" and counts us lucky to live in the generation with "the technology and the will" to study it. He also states the reductionist case bluntly, "Our reality depends on what our biology is up to", which Kaku accepts without conceding that it diminishes us. Kaku frames the two great mysteries, mind and universe, as sharing a common history and perhaps a common destiny; we are, he writes, "literally children of the stars".
Two shaky supports. The Toba bottleneck theory has largely collapsed: archaeological sites in South Africa and India show people thriving through the eruption, and genetic bottleneck estimates now point elsewhere (one 2023 study argues for a severe bottleneck about 900,000 years ago, itself debated). Kaku also garbles the anthropic principle: the weak form is a selection effect (we can only observe a universe compatible with observers), the strong form says the universe must permit life; a designer is a separate, theological inference, not the strong principle itself. Finally, his optimism that technology always trickles down sits awkwardly next to Joy's specific point, which was about concentrated, irreversible risks, not consumer access.
Since 2014. Kaku's priority call aged well. COVID-19 (from 2020) killed millions; the "rapid response" he urged arrived as mRNA vaccines designed within days of the virus's sequence and deployed within a year, while the debate over a possible lab origin made biosafety a political issue. In December 2024 leading scientists warned against creating mirror-image bacteria, a direct heir of Joy's lab-escape fear. Grey goo remains fiction. Joy's third threat moved from fringe to mainstream: since 2023 AI-risk statements, national AI Safety Institutes (UK first, 2023) and the EU AI Act (2024) treat advanced AI as a governance problem, much as Kaku hoped "vigorous democratic debate" would.
Kaku's framing (scientists, nature and society; risks ranked by proximity; safeguards plus rapid response) is a workable template for an AI risk committee: rank threats by time-to-impact, invest early in the near ones (fraud, model failure, data leakage), keep kill switches for the distant ones, and treat public trust as a third stakeholder rather than an afterthought.
Quantum Consciousness?
Some claim consciousness is more fundamental than matter, citing Schrödinger's cat. Kaku walks through the three main resolutions (Copenhagen collapse, Wigner's consciousness-driven collapse, Everett's many worlds), notes that all give the same experimental predictions, and then turns to the one place the debate touches the brain: free will. His verdict: quantum uncertainty plus chaos defeat strict determinism, so free will exists, but it is far more constrained by unconscious factors than rugged individualists believe.
The paradox at the root of quantum theory
Quantum mechanics underpins lasers, MRI, electronics, GPS and telecoms, and some predictions match experiment to one part in 100 billion, yet Kaku admits his life's work stands on "feet of clay". Electrons act as particles (cloud-chamber tracks) and as waves (interference). Schrödinger's wave equation (1925 to 1926; Nobel 1933) explained hydrogen and the periodic table, but what is waving? In 1927 Heisenberg's uncertainty principle said position and momentum can't both be known, even by God, so the wave gives the probability of finding the particle.
The field split: Bohr, Heisenberg and the atomic physicists turned quantum mechanics into a Nobel-winning "cookbook"; Einstein, Schrödinger and de Broglie objected (Schrödinger regretted his equation if it meant probability). Einstein: "God does not play dice with the world." Bohr: "Stop telling God what to do."
A cat sits in a sealed box with uranium, a Geiger counter, a hammer and a flask of poison gas. If an atom decays, the counter trips the hammer, the flask breaks and the cat dies. Quantum theory describes the atom as decayed-plus-undecayed, so the cat is the sum of a dead-cat wave and a live-cat wave: neither dead nor alive until something settles it.
| Interpretation | Who and when | What resolves the cat | Implication for mind | Kaku's comment |
|---|---|---|---|---|
| Copenhagen | Bohr, Heisenberg (1920s) | Measurement collapses the wave; a movable "wall" separates quantum and classical worlds | Observation creates definite reality; a new quantum solipsism (a tree is every possible tree until seen) | Taught for 80 years; strained now that we can push single atoms around (Kaku did so at IBM Almaden) |
| Wigner's friend / cosmic consciousness | Eugene Wigner (1967) | Only a conscious observer collapses the wave, but observers need observers, ending in an infinite chain or God | Consciousness is more fundamental than atoms; Wigner turned to Vedanta | Unfalsifiable; and observation fixes odds, not outcomes, so you cannot meditate reality into the shape you want (you can compute the odds of a royal flush, not deal one) |
| Many worlds | Hugh Everett (1957) | Waves never collapse; they split into branches | Every possibility exists; there is no wall, quantum strangeness reaches everyday life | The simplest formulation and the most disturbing |
Does the moon exist because a mouse looks at it?
Albert Einstein, to guests at his houseEinstein's objection was to a new solipsism. Classical physics, from Newton's atoms as tiny steel balls at definite places to relativity, seemed to have banished Bishop Berkeley's idea that unobserved things may not exist; Walter Lippmann praised modern science for rejecting the belief that the forces moving stars and atoms depend "upon the preferences of the human heart". Quantum theory reopened the door. Kaku adds a practical twist: if molecular transistors replace silicon, the paradoxes of quantum theory will sit at the heart of every future computer, and the world economy will rest on them. Wigner, a builder of the atomic bomb as well as of quantum theory, concluded the laws could not be stated consistently "without reference to consciousness"; in his picture the material world may come and go while consciousness remains.
Many worlds in your living room
In Everett's picture, people dead here are alive elsewhere (Elvis included), and a single cosmic ray striking Hitler's mother could branch off a world without World War II and its 60 million dead. Philip K. Dick's alternate-history novel turns on an assassin killing Roosevelt; whether a gun fires may hinge on quantum events in the powder. Why can't we visit? Steven Weinberg's radio analogy: hundreds of stations fill your room but your set is tuned to one; the others have decohered. Dinosaur and pirate branches share your living room, out of phase. "Sliding" between them (the TV series Sliders was inspired by Kaku's book Hyperspace) is like tunnelling through a brick wall: you'd wait longer than the age of the universe. Even your mirror image is about a billionth of a second old and an average over countless wave functions.
All three interpretations predict identical experimental results, so the debate may last centuries. Its one bridge to neuroscience is free will.
Two details of Everett's picture matter to Kaku: the dead in our branch are alive in theirs and insist their branch is the real one, and the more bizarre a universe, the less likely it is. Weinberg's term "decoherence" means waves falling out of phase, the opposite of a laser's perfect unison, which is why we share the living room with dinosaur and pirate branches without noticing. In Dick's alternate history, Roosevelt's death leaves America unprepared, and the Nazis and Japanese win and partition the United States.
Civilisation rests on free will: reward, punishment and personal responsibility all assume it. Kaku notes that more and more neuroscientists doubt it "in the usual sense"; if specific behaviours trace to specific defects, some offenders may need confinement and treatment rather than punishment.
Free will
- Simon Baron-Cohen (Cambridge) told Kaku many (not all) pathological killers show a brain anomaly: watching others in pain, their amygdala and nucleus accumbens (pleasure centre) light up instead of empathy circuits. Some argue they need treatment and confinement, not punishment.
- Benjamin Libet (1985): subjects watched a clock and noted when they decided to move a finger; EEG showed the brain had committed about 300 milliseconds before awareness. Michael Sweeney: "the brain knows what a person will decide before the person does."
- Newtonian determinism: the universe is a clock; every event, including your choices, is fixed.
- Religious predestination: an omniscient God knows your fate before birth. Kaku ties the Reformation split partly to this: indulgences seemed to let money change fate, and Luther posted his 95 Theses in 1517.
- Quantum uncertainty (after 1925): only probabilities are calculable, so perhaps free will is a quantum effect, though determinists reply that quantum effects are too tiny at human scale.
- Chaos: even classical systems are unpredictable in practice; a butterfly's wingbeat can grow into a storm. Gould's rerun Earth would likely not produce us.
- Kaku's resolution: "free will" may dissolve like "What is life?" into layers. A film is fully determined; our world is not, because of quantum branching and chaos.
Chaos is the classical half of the argument. Even with every air molecule known and a big enough computer, weather forecasts become useless after a few hours, because tiny differences grow, the "butterfly effect". A film is fully determined; our world cannot be like a film, both for quantum reasons and because of chaos.
The quantum brain
Here Kaku finds a puzzle: a reverse-engineered transistor brain would be deterministic (same question, same answer), yet quantum mechanics and chaos say the world is not. Roger Penrose argues the brain is a quantum device: Gödel showed arithmetic contains true but unprovable statements; machines are bound by such limits, but human intuition sees past them, so the mind is non-computational, with quantum processes in microtubules inside neurons. Most scientists are sceptical, Kaku notes, but science is not a popularity contest.
Penrose
The brain is quantum in its essentials; no computer can reproduce consciousness; reverse engineering is doomed.
Kaku
Transistors can only approximate neurons, which are analog and digital, leaky, misfiring, ageing and environment-sensitive, partly through quantum effects that grow as axons thin. A transistor robot will pass the Turing test and appear conscious but lack free will, and differ from us in tiny quantum ways.
Kaku's final position: free will exists, but not the kind claimed by "rugged individualists". Thousands of unconscious factors predispose our choices; yet "the end of the movie hasn't been written yet". A subtle combination of quantum effects and chaos destroys strict determinism, so "in the end, we are still masters of our destiny."
Corrections and missing objections. The Philip K. Dick novel is The Man in the High Castle. Indulgences remitted temporal punishment for sin; they did not let money "alter determinism", and the Reformation's predestination debates actually ran the other way (Calvinists were the strict predestinarians). Libet's original papers date from 1983. Kaku's step from unpredictable to free is the weakest link: randomness from quantum events or chaos makes behaviour unpredictable, not chosen, which is why most philosophers think it does not rescue free will. And his "computers always give the same answer" is outdated: modern AI models sample their outputs and are deliberately non-deterministic.
Since 2014. Libet's interpretation has weakened: Aaron Schurger's work (from 2012) suggests the "readiness potential" is an accumulation of random neural noise that crosses a threshold, not a hidden decision, and later experiments found it absent for meaningful (as opposed to arbitrary) choices. Penrose won the 2020 Nobel in Physics for black-hole work, not consciousness; an underground experiment at Gran Sasso (2020 to 2022) ruled out the simplest version of the gravity-induced collapse his theory needs, while a 2024 rat study found microtubule-stabilising drugs delayed anaesthesia, which supporters cite and critics dispute. Quantum foundations stayed lively: the 2022 Nobel went to Aspect, Clauser and Zeilinger for entanglement experiments that close loopholes against Einstein's local realism; and extended "Wigner's friend" experiments since 2019 have tested whether observers can disagree about facts.
For decision-making in organisations, Kaku's compromise is practical: people (and AI agents) act under heavy priors they don't see, so good governance designs choice architecture (defaults, prompts, review points) rather than relying on heroic individual will. Libet-style findings also caution against treating a person's post-hoc rationale, or an AI model's self-explanation, as the true cause of a decision.
Master Takeaways
The ideas worth keeping from The Future of the Mind, ranked, followed by every framework in the book, a verdict on the whole, the cast, and where the ideas connect to AI and enterprise work.
- Consciousness is model-building with feedback loops toward a goal, and humans are distinguished by modelling time: we use the past to simulate the future. Kaku's Level 0 to III ladder makes this rankable, if not yet measurable.
- Every brain-machine interface is a learned dictionary between signals and meanings; since 2023, AI language models have turned these dictionaries into fluent speech, sentence-level fMRI decoding and inner-speech decoding.
- Read the brain's outputs, not its whole computation: 96 or 100 electrodes on output neurons can drive arms, cursors and speech because the brain has already done the hard work.
- The hippocampus is the gateway of memory, so memories can be recorded and replayed there; a human hippocampal prosthesis improved recall about 35% in 2018.
- Mental illness is a failure of checks and balances between rival feedback loops (voices versus discounting, alarm versus satisfaction, optimism versus pessimism), which is why targeted stimulation and closed-loop DBS work for some patients.
- The danger from advanced minds, machine or alien, is indifference and badly specified goals more than malice: the developer who doesn't see the deer, the robot that "protects" humanity by ruling it.
- Intelligence is the complexity of one's future simulations, plus restraint: delayed gratification and divergent thinking predict outcomes better than IQ, though the marshmallow effect has shrunk on replication.
- Biology is near its physical ceiling: bigger, denser or faster neurons run into heat, energy and unstable ion channels, so enhancement must come from efficiency or external help.
- The self is a story the left hemisphere tells: split-brain patients show the interpreter confabulating reasons for actions it never chose.
- Most of the mind is subconscious and energy-saving: blind spots filled in, colours computed, emotions as fast lower-level decisions in a 20-watt organ.
- Dreams are simulations with the CEO and fact-checker switched off, built by the cortex from brain-stem noise; they can be crudely decoded and lucid dreamers can communicate from inside them.
- Mind control by drugs and hypnosis failed; precision tools (optogenetics, implanted electrodes) are what make it a real future risk, so governance must precede precision.
- Emotions are how a mind assigns value: without them (Damasio's patients) decisions stall, so useful robots need curated emotions such as loyalty, empathy and fear, and must lack anger and a will to command.
- A full wiring map is not understanding: the fly connectome arrived in 2024, but like the genome, the map is a dictionary without definitions.
- Simulating neurons is not simulating a mind: a bodiless cortex-thalamus model "cannot catch a mouse"; the Human Brain Project ended without a simulated brain.
- The Caveman Principle: given high-tech or high-touch, people choose high-touch, so enhancements must be invisible, optional and reversible.
- Out-of-body and near-death experiences have neural switches at the temporoparietal junction and in blood-flow loss; the brain's self-location model is easily scrambled.
- Telepathy's privacy defence is physics today and law tomorrow: fields fade with the cube of distance, but recorded neural data needs neurorights, now in Chile, US states and UNESCO's 2025 recommendation.
- Neurotechnology inherits the hype cycle of AI: World Cup exoskeleton kicks, cat-brain simulations and mirror-neuron prophecies were oversold; speech BCIs and DBS quietly delivered.
- Aliens will share only the abstract definition of consciousness; their values follow from reproduction, lifespan and habitat, and they are likely post-biological.
- Free will survives as constrained freedom: unconscious factors bias every choice, yet quantum chance and chaos rule out a fixed script; unpredictability, though, is not the same as authorship.
- Biotech, not robots or grey goo, was the near-term existential risk Kaku ranked first, a call COVID-19 and mirror-life warnings vindicated.
Frameworks and mental models
| Name | What it says | Use it when |
|---|---|---|
| Space-time theory of consciousness | Consciousness = feedback-loop model of the world toward a goal; Levels 0 (physical), I (space), II (others), III (time) | Classifying any agent, animal or AI system by what it models |
| Level scores (0:1, II:150, III:100) | Count loops; group × interactions; causal links vs average | Comparing autonomy or planning depth, illustratively |
| Brain as a corporation | Subconscious bureaucracy, fast emotional flags, competing sub-centres, prefrontal CEO, hierarchical flows, 20-watt budget | Explaining attention, confabulation, or designing multi-agent AI |
| Left-brain interpreter | The verbal hemisphere invents coherent stories for actions it didn't cause | Distrusting post-hoc explanations (human or model) |
| BMI dictionary method | Record, pair signals with items, decode by lookup, refine | Any neural decoding or supervised-learning project |
| Read outputs, not computation | Tap output neurons; the brain has done the processing | Designing minimal-sensor interfaces |
| Intent plus external power | The mind commands; a power source supplies force | Designing agent tool-use with least privilege |
| Memory pathway | Senses → thalamus → cortex → prefrontal (short-term) → hippocampus → scattered storage | Thinking about memory systems, RAG, consolidation |
| Three-step memory library | Map CA1 to CA4, record task flows, upload to another subject | Judging claims about memory implants |
| Activation-synthesis | Brain-stem PGO noise + disabled checks → cortex narrates a dream | Understanding dreams and hallucination |
| Feedback-loop theory of mental illness | Disorders are imbalances between rival loops (table in ch. 9) | Framing psychiatric disorders and AI failure modes |
| Robot emotion menu | Keep loyalty, empathy, fear; delete anger and desire to command; avoid humour | Specifying AI assistant behaviour |
| Four stages of emotional AI | Identify, respond, infer cause, choose by simulating | Building affective computing |
| Three AI failure modes | Malfunction, conflicting goals, single goal taken too literally | Model risk and agent safety testing |
| Three routes to reverse engineering | Simulate (HBP), map (connectome/BRAIN), annotate (Allen atlas) | Planning discovery of any complex system |
| Caveman Principle | High-touch beats high-tech | Predicting technology adoption |
| Four hurdles for exoskeletons | Durable implants, wireless, decoding thousands of neurons, portable power | Assessing neuroprosthetic readiness |
| Physics ceiling on intelligence | Size, thinness, thickness, connections all hit heat/energy/noise limits | Evaluating enhancement claims |
| Three prerequisites for intelligence | Manipulator, stereo vision, language | Speculating about alien or animal minds |
| Fermi answers | Distance, self-destruction, irrelevance (ants and highway) | SETI arguments |
| Developer, not hunter | Advanced agents threaten through indifference | AI and alien risk analysis |
| Rome and the barbarians | The weak absorb the strong's technology via collaborators and attrition | Asymmetric competition |
| Copernican vs Anthropic | We're not special vs the universe is tuned for minds | Framing meaning and reductionism |
Strengths. Kaku is a superb synthesiser and interviewer: the book captures 2013 neuroscience vividly, and his space-time theory is a genuinely useful organising scheme, close to the predictive-processing view that has since become mainstream. His priority call on biotech risk and his insistence on democratic debate have aged well. Weaknesses. (1) Factual slips are frequent: Broca's area placed in the temporal lobe, sodium-22 PET, "RIM-941" and "FOX2", Moniz's name, the Arecibo date, Robert vs Ray Bradbury, and more, flagged chapter by chapter. (2) Hype carried uncritically: mirror neurons, the 10,000-hour rule, the marshmallow test, the triune brain, Modha's cat brain and the 2014 exoskeleton kick were all weaker than presented. (3) The theory is under-tested: the level numbers are illustrative, and it says nothing about why model-building feels like anything (the hard problem), which Kaku dismisses rather than answers. (4) AI blind spot: writing just after AlexNet (2012), he declared pattern recognition and common sense the great walls; deep learning and language models broke both within a decade, while his feared "self-aware Internet" never materialised in the form he imagined. (5) Optimism about equity: his "technology always trickles down" argument skips concentrated, irreversible risks and decades-long access gaps. (6) Identity questions around copying and beaming minds are raised and dropped. Net. A reliable map of the questions and a dated map of the answers; read it with the 2026 updates beside it.
Who's who
| Person | Role | Why they matter in the book |
|---|---|---|
| Phineas Gage | Railroad foreman, 1848 | Frontal-lobe damage changed his personality; launched localisation |
| Wilder Penfield | Neurosurgeon | Mapped motor cortex; triggered memories by stimulation |
| Paul MacLean | NIMH | Triune brain model (1967) |
| Roger Sperry / Michael Gazzaniga | Caltech / UCSB | Split-brain studies; the left-brain interpreter |
| David Eagleman | Neuroscientist | Subconscious "young king", umwelt, temporal-lobe prophets |
| V. S. Ramachandran | UC San Diego | Mirror neurons, hyperreligiosity, Capgras |
| Jack Gallant / Shinji Nishimoto | UC Berkeley | Decoding movies and dreams from fMRI |
| Brian Pasley | UC Berkeley | Word decoding by ECoG |
| John Donoghue | Brown University | BrainGate implant |
| Miguel Nicolelis | Duke | Brain-machine-brain interfaces, brain-to-brain rats, Walk Again exoskeleton |
| Theodore Berger / Sam Deadwyler | USC / Wake Forest | Artificial hippocampus and cortex |
| Joe Tsien | Princeton, Medical College of Georgia | Doogie and Hobbie-J smart mice; CaMKII erasure |
| Tim Tully / Jerry Yin / Alcino Silva | Cold Spring Harbor | CREB genes and why cramming fails |
| Katherine Pollard | UC Santa Cruz | Found HAR1, the fastest-changing human region |
| Allan Hobson | Harvard | Activation-synthesis theory of dreams |
| José Delgado | Yale | Stopped a charging bull by radio-controlled electrode |
| Helen Mayberg | Toronto, Emory | DBS of area 25 for depression |
| Ed Boyden / Karl Deisseroth | MIT / Stanford | Optogenetics and Clarity |
| Rodney Brooks | MIT, iRobot | Bottom-up robots; "we are machines"; merge with them |
| Cynthia Breazeal / Lola Cañamero | MIT / Hertfordshire | Emotional social robots Huggable, Nexi, Nao |
| Henry Markram / Dharmendra Modha | EPFL / IBM | Brain simulation, Blue Brain, Human Brain Project |
| Sebastian Seung | Connectome project | "Your soul on a disk"; EyeWire |
| Ray Kurzweil / Hans Moravec | Futurists | Singularity forecasts; gradual neuron-by-neuron transfer |
| Seth Shostak / Dan Werthimer / Paul Davies | SETI | Contact predictions, seeded test signals, post-biological aliens |
| Bill Joy | Sun co-founder | "The Future Does Not Need Us" (2000) |
| Roger Penrose | Oxford | Quantum, non-computable consciousness via microtubules |
Connect it
- AI architecture. Kaku's corporation model and Level III definition map onto agentic AI: an orchestrator (CEO) arbitrating specialist sub-agents, a world model for future simulation, and a confabulation risk that makes independent logs essential.
- Model risk in banking. His three AI failure modes (malfunction, conflicting goals, literal single goals) and his "emotion menu" give a ready checklist for PRA SS1/23 model risk reviews and FCA Consumer Duty testing of AI assistants (loyalty to the customer's interest, empathy for vulnerability, refusal of unsafe actions).
- Data governance. Neural data is the next sensitive category after biometrics; the neurorights movement (Chile 2021, US state laws 2024, UNESCO 2025) is a preview of consent, purpose-limitation and inference-protection rules banks will meet in consumer wearables.
- Enterprise discovery. The simulate/map/annotate split and the "Neanderthal with a blueprint" warning apply to legacy estate mapping: start from symptoms and journeys, not a complete inventory.
- Adoption strategy. The Caveman Principle is a design rule for digital banking: keep human reassurance, proof and reversibility in AI-led journeys.
About this book
The Future of the Mind: The Scientific Quest to Understand, Enhance, and Empower the Mind by Michio Kaku, 2014. This is a chapter-by-chapter study of the book, with summaries, maps, tables and diagrams; short quotations keep their page references. It is shared free for personal study, is not for sale, and is not endorsed by the author or publisher. Please read and buy the original.
Insight Editions · rajeshrmahapatra.com/insight-editions