Two hands, one breakfast
Anand is 34 and works as an accountant in Chennai. At 22 he had surgery for epilepsy that no medicine could control. His seizures started on one side of his brain and spread to the other within seconds, and he fell without warning several times a week. The surgeon cut the corpus callosum, the thick band of fibres that connects the two halves of the brain. The falls stopped.
Twelve years later, Anand's life looks ordinary. He commutes, files tax returns for small businesses, plays carrom with his nephew and argues about cricket. His wife, Divya, says that he is the same person she married, and he says the same about himself. If you met him at a wedding you would notice nothing.
Anand, Divya and the neuropsychologist we will meet are invented. I will return to them throughout, changing one thing at a time, because a single person makes it easier to see what each experiment can and cannot tell us. The studies and numbers around them are real and cited.
In a laboratory in Bengaluru, a neuropsychologist, Dr Nandini Rao, asks Anand to fix his eyes on a dot in the middle of a screen. For a tenth of a second, a word appears to the left of the dot. Because of the way the eyes connect to the brain, a word on the left is seen first by the right hemisphere. Nandini asks him what he saw.
"Nothing," Anand says. "Was there something?"
Then she asks him to reach under a screen with his left hand and pick out the object the word named. His left hand, controlled mainly by the right hemisphere, finds a key among a dozen objects and holds it up. Anand looks at the key in his hand with mild surprise.
This is the scene that made split-brain research famous. One half of the brain seems to know something the other half can talk about, and cannot. For fifty years it has been read as evidence that cutting the corpus callosum creates two minds in one skull.
But Nandini runs a second experiment, and it is less famous. She flashes a dot somewhere on the left side of the screen and asks Anand to point to where it appeared, sometimes with his left hand and sometimes with his right. If the two hemispheres were fully separate minds, his right hand, controlled by the left hemisphere, should be guessing. In some patients it is not. Both hands point accurately.
The question is not only academic. How we count minds affects how we interpret consent, responsibility and testimony, and it shapes how we think about any system built from parts that communicate through narrow channels. If one skull can hold two subjects, then the assumption that a body is a person needs qualification. If it holds one subject with divided access, then unity of the subject survives damage better than the classic picture suggested.
That second experiment is the reason for this paper. The question I take from the series list is: does commissurotomy create two independent streams of phenomenal consciousness, or two unintegrated access channels for one subject? A stream of phenomenal consciousness is the flow of experience itself, what it is like to be someone. An access channel is a route by which information reaches report and the control of action. The two readings predict different things, and the evidence of the last decade has started to separate them.
My answer, stated early so the rest of the paper can test it, is that the current evidence fits one subject with two partly separate access channels better than two complete streams. But the fit rests on few patients and contested interpretations, and there is a limit to what behaviour can show. I will set out the classic findings, the newer challenge, the reply from the classic camp, and four candidate models. Then I run a small instrument that compares those models on patterns of data, read the result against the five layers of the Recursive Tattva Model, and set out the strongest case against my conclusion.
The classic split
The story begins with a patient known as W.J. He was a Second World War paratrooper who began having seizures after a head injury during the war. By the early 1960s they were severe and frequent, and medicine had failed. In 1962, the neurosurgeons Philip Vogel and Joseph Bogen in Los Angeles cut his corpus callosum and other connections between the hemispheres. The seizures were controlled.
Before and after the operation, a graduate student named Michael Gazzaniga, working in Roger Sperry's laboratory at Caltech, tested him. An account in PNAS describes the episode in detail, including film of the tests. W.J. could name images flashed in the right half of his visual field, which go to the left, speaking hemisphere. For images flashed on the left he reported seeing nothing. Yet his left hand, controlled by the right hemisphere, could point to a picture of what had been shown.
In another test he was asked to arrange coloured blocks to match a pattern. His right hand fumbled. His left hand did the task with ease and even tried to step in and help. The first results were published in PNAS in 1962 by Gazzaniga, Bogen and Sperry. Gazzaniga later recalled the moment when he realised that W.J. could no longer describe stimuli shown to his right hemisphere as the one that launched his career.
Why does a word on the left go to the right hemisphere? Each eye sends information from the left half of the visual field to the right hemisphere, and from the right half to the left hemisphere. In an intact brain, the corpus callosum then shares each half with the other side within a fraction of a second. After the operation, that sharing is largely lost. If the eyes stay fixed and the image is brief, each hemisphere receives only its own half of the world.
The flash has to be brief for a reason. If an image stays on the screen, the patient moves his eyes towards it, and within a fraction of a second both hemispheres have seen it. This is why so much of the split shows only in the laboratory. In daily life, people look around freely, and the two hemispheres see most things in turn.
Predict before reading on
?> Before the next paragraph, make a guess. A split-brain patient sees the word "TAXABLE" so that "TAX" falls only in the left half of his visual field and "ABLE" only in the right. He is asked to say what he saw, and then to write it with his left hand. What does he say, and what does he write?
The Stanford Encyclopedia of Philosophy uses exactly this example. The spoken answer, controlled mainly by the left hemisphere, is "ABLE". The left hand, controlled mainly by the right hemisphere, writes "TAX". Neither hemisphere reports the whole word, and the patient does not report a conflict. The entry concludes that visual information in these patients does not seem to be fully access unified.
Two spheres of awareness
Sperry drew the strong conclusion. In a 1968 article in American Psychologist, drawing on studies of a series of patients operated on by Vogel, he described an apparent doubling in most realms of conscious awareness. Instead of a single stream of consciousness, he wrote, these patients behave in many ways as if they have two independent streams, one in each hemisphere, each cut off from the experiences of the other. Each hemisphere seemed to have its own sensations, perceptions, concepts, impulses to act and memories.
He went further. He favoured the view that in the minor hemisphere we deal with a second conscious entity that runs in parallel with the dominant stream in the major hemisphere. This became the classic two-streams model. It was the default reading in textbooks for decades.
Sperry was also careful about why the duality did not show in daily life. The two mental spheres, he noted, have only one body. They are dragged to the same places, meet the same people and see the same things, so they share a great deal of nearly identical experience. Outside the laboratory, where both halves of the visual field see the same scene and the eyes move freely, the split rarely shows.
The interpreter
The most vivid finding came later, from Gazzaniga's work with other patients. In a review of fifty years of split-brain research, Lukas Volz and Gazzaniga describe a classic experiment with a patient called P.S. A chicken claw was shown to his speaking left hemisphere, and a snow scene to his right hemisphere. He was then asked to choose matching pictures from an array.
His right hand chose a chicken, which matched the claw. His left hand chose a snow shovel, which matched the snow scene that only the right hemisphere had seen. Asked why, he replied: "Oh, that's simple. The chicken claw goes with the chicken, and you need a shovel to clean out the chicken shed."
The speaking hemisphere did not say "I don't know". It produced a confident explanation for an action whose real cause it could not see. Gazzaniga called this system the left-hemisphere interpreter. Volz and Gazzaniga describe it as getting the gist, seeking patterns and weaving information into a coherent story, and suggest that it may contribute to our feeling of being a single, unified self.
The interpreter in intact brains
It would be a mistake to treat the interpreter as a peculiarity of surgical patients. In a 2005 study in Science, Petter Johansson, Lars Hall and colleagues asked healthy volunteers to choose which of two photographs of faces they found more attractive. On some trials, the experimenter covertly swapped the cards and handed over the face the participant had rejected. Most of the swaps went unnoticed, and participants often went on to explain why they had preferred a face they had in fact rejected.
The researchers called this choice blindness. Its relevance here is that confident explanation of one's own choices does not require access to their causes, even with an intact corpus callosum. The split-brain interpreter may be an ordinary system revealed by an unusual disconnection, rather than a new system created by surgery.
That observation cuts in two directions. It weakens the evidential value of any patient's report about the unity of their own mind. But it also weakens the inference from the interpreter to two minds, because ordinary single-minded people confabulate too.
The interpreter matters for this paper in two ways. It shows that a patient's own report of feeling unified cannot settle the question, because the reporting system is known to fill gaps with stories. And it raises a sharper question. If the left hemisphere's account of itself can be wrong about why the hand moved, can it be wrong about how many experiencers there are?
The 2017 challenge
For decades, the two-streams reading rested on a particular pattern. Information shown to one side could be reported only by the hand or voice controlled by the opposite hemisphere. If the left hemisphere could not answer about the left visual field, the inference seemed natural. The information was not available to it, so it belonged to a different mind.
In 2017, Yair Pinto and colleagues in Amsterdam tested that pattern directly. In a paper in Brain, they studied two patients whose corpus callosum had been completely cut. They used a set of tasks in which stimuli could appear anywhere in the visual field, and the patients could respond with the left hand, the right hand or speech.
What Pinto found
The results came in two parts, and the contrast between them is the heart of this paper. First, the patients could not compare two stimuli placed on opposite sides of the midline. Asked whether two shapes, one in each half-field, were the same or different, they performed at chance. That part matched the classic picture.
Second, the patients could locate, orient and identify stimuli anywhere in the visual field, well above chance, whatever response they used. A dot on the left could be pointed to with the right hand. A shape on the right could be named or indicated with the left hand. Confidence ratings showed the same pattern on trials where patients felt sure. The authors concluded that severing the cortical connections between the hemispheres splits visual perception, but does not create two independent conscious perceivers in one brain.
| Task | Classic reading predicts | Pinto 2017 found |
|---|---|---|
| Compare two shapes across the midline | Chance | ✗ Chance |
| Locate a dot, hand on the same side | Accurate | ✓ Accurate |
| Locate a dot, hand on the other side | Chance | ✓ Accurate |
| Name or identify, any response | Only one side | ✓ Above chance |
Return to Anand. In the classic version of Nandini's experiment, he says "nothing" about the word on the left and then finds the key with his left hand. In the Pinto version, she shows a dot on the left and asks him to point with his right hand. If he points accurately, the left hemisphere has some access to where the dot was, even though it may not be able to describe what the dot looked like or compare it with something on the right.
That thought experiment cuts in a direction worth noticing. It weakens the simplest argument from Pinto's data to unified experience. Accurate pointing proves unified access to location. It does not prove unified experience of what is at that location.
The reply from the classic camp
Volz and Gazzaniga responded in the same journal with a broad review of fifty years of work. They agreed that patients report a single consciousness and that verbal intelligence is unaffected by the surgery. But they argued that the Pinto results could reflect cross-cueing.
Cross-cueing is the idea that one hemisphere signals to the other through the body. A hemisphere might move the eyes towards a stimulus, tense a muscle, or start a movement that the other hemisphere notices. Volz and Gazzaniga also raised the possibility of transfer through the superior colliculus, a midbrain structure involved in orienting, or through the anterior commissure, a smaller bundle of fibres that is sometimes left intact. They noted that some patients show alien hand signs or intermanual conflict, in which one hand acts against the other.
Pinto's counter-reply
Pinto, Victor Lamme and Edward de Haan replied in Brain. Their argument rested on timing and precision. The difference in response time between congruent and incongruent conditions was less than 200 milliseconds. Overt cueing, in which one hemisphere would need to notice a bodily signal and then act on it, should take longer.
They also reported that pointing errors were small, between one and three degrees, whichever hand was used. Cueing through eye movements or posture would be expected to give coarse information, not precise coordinates. And they found no interaction between response type and visual field, which a cueing account would tend to predict.
Why confidence ratings matter
The confidence ratings in the 2017 study deserve separate attention, because they bear on a common objection. A sceptic might argue that accurate opposite-hand responses reflect blindsight-like guessing: information that guides action without reaching awareness at all. Patients with damage to the primary visual cortex sometimes point accurately to targets they insist they cannot see, so accuracy alone does not establish conscious access.
Pinto's team asked patients to rate their confidence and reported that the pattern held on high-confidence trials. If the opposite-hand responses were unconscious guesses, one would expect confidence to be low on those trials, or accuracy to collapse when only confident trials were analysed. Neither happened, which makes the guessing explanation harder to sustain. It does not rule it out completely, since confidence itself is a report, and reports are produced by the same access systems whose unity is in question.
The reply conceded one point. Subcortical transfer of some information remained possible. They added a methodological complaint: cross-cueing is not clearly defined, which makes it an easy post hoc explanation, but also one that is difficult to falsify. This is the right complaint, and it applies to both sides. A theory that can explain any result after the fact explains nothing, and a theory that cannot say what it would take to be wrong is not yet a theory.
What each side conceded
It is worth listing what the two sides agreed on, because the agreement is larger than the dispute. Both accepted that the patients cannot compare stimuli across the midline. Both accepted that the patients report a single consciousness and that their intelligence is preserved. Both accepted that some information may travel through subcortical routes. The disagreement concerned only whether the opposite-hand accuracy reflected one subject's access or coordination between two.
That narrowing is a form of progress. A question that began as "one mind or two?" became a question about the speed and content of residual transfer, which is a question that laboratories can answer.
Where the field landed
In 2020, a group that included both camps published a collective review in Neuropsychology Review. Edward de Haan and colleagues found that integration of perceptual information breaks down broadly after the operation, but that action control remains largely unified. Their consensus was that the evidence was insufficient to decide between the classic models and that intermediate conceptualisations were needed.
That is a striking position for a field after sixty years. It does not say that the two-streams model is false. It says the data are not yet good enough to choose, and that the truth may sit between the familiar options.
Four candidate models
The question can now be put more exactly. Four models have been proposed, and they make different predictions about the same patterns of data. I will set them out plainly before testing them.
Model one: two streams
On the two-streams model, each hemisphere carries its own stream of experience after the operation. This is Sperry's reading, in which the minor hemisphere is a second conscious entity. The left and right streams are each unified internally but not with each other.
A modern refinement comes from Elizabeth Schechter. In her 2018 book with Oxford University Press, she argues that a split-brain patient is two conscious psychological beings who together make themselves into one person. The two streams share a body, a life and a name, and they cooperate so closely that the result functions as a single person.
The model predicts the classic pattern. Information shown to one hemisphere should be available only to that hemisphere's responses. On this view, accurate pointing with the opposite hand needs a special explanation, such as cueing.
Model two: the switch
Tim Bayne offered a different answer in a 2008 paper in the Journal of Philosophy. On his switch model, the patient has a fully unified consciousness at all times. But the contents of that consciousness switch between hemispheres, depending on where attention is directed. At one moment the stream is fed by the left hemisphere, and at the next by the right.
The switch model keeps one stream. It explains why the patient never reports two conflicting experiences at once: at any moment only one hemisphere is supplying the contents. According to the Stanford Encyclopedia of Philosophy, the most serious challenge to the switch model is the rare evidence of interhemispheric conflict, in which the two hands seem to pursue different goals at the same time.
Model three: partial unity
On the partial-unity model, the two hemispheres share some experiences but not others. Imagine three experiences, A, B and C. A and B are unified, and B and C are unified, but A and C are not. The relation is not transitive. This is hard to imagine from the inside, which is one of the main objections to it.
Thomas Nagel, writing in 1971, took a related view. The Stanford Encyclopedia summarises his position: there is no whole number of centres of consciousness in these patients. There is too much unity, in daily life and even in laboratory behaviour, to say two, and too much separation in the contrived laboratory tasks to say one.
What partial unity would look like in data
Partial unity is often dismissed because it is hard to imagine, but imaginability is a poor guide to truth in this area. A more useful question is what it would predict. If experiences A and C are each unified with B but not with each other, then tasks requiring integration of A with B, or B with C, should succeed, while tasks requiring integration of A with C should fail.
Applied to the split brain, the shared element B might be information that travels through intact subcortical routes, such as coarse location or emotional tone. Content specific to each hemisphere would then be unified with that shared element but not across hemispheres. Behaviourally, this predicts something close to the one-subject pattern for location and a split pattern for detailed identity, which is why the next generation of studies needs identity tasks as well as location tasks.
The difficulty is that partial unity and model four can make the same behavioural predictions. The difference between them concerns whether there is one subject who has all the experiences or a non-transitive web of experiences without a single owner. That difference is again located at the level the experiments do not reach.
Model four: one subject, two access channels
The fourth model is the one Pinto's data suggest, and the one I will defend with caveats. There is one subject of experience. But the information that reaches that subject comes through two channels that are no longer fully integrated. Each channel can deliver some kinds of content to the subject and to action control, such as where something is. But the two channels cannot be combined for operations that need both at once, such as comparing a shape on the left with one on the right.
A useful way to see the difference is to ask what is split. On model one, the subject is split. On model two, nothing is split at any instant, but the source switches. On model three, the unity relation itself is partial. On model four, the subject and its control of action are whole, and what is split is the integration of perceptual content.
The access and phenomenal distinction
Models one and four turn on a distinction that is easy to blur. Access consciousness, in the usage of many philosophers, is information that is available for report, reasoning and the control of action. Phenomenal consciousness is what it is like to have the experience. The Stanford Encyclopedia frames the split-brain question partly in these terms, and concludes that visual information in these patients is not fully access unified.
The difficulty is that every experiment measures access. A patient points, names, presses a button or rates confidence. Each response is an act of access. So the data can show directly that access is split in some ways and unified in others. Any claim about phenomenal consciousness is an inference from those patterns.
That is why I phrased my answer carefully. The evidence fits one subject with two access channels better than two complete streams. Whether the access channels correspond to one phenomenal field or two is a further question, and behaviour reaches it only through assumptions.
This thought experiment is the main limit on the whole field. It is not a reason to stop. It is a reason to be clear about what each experiment can decide. Some distinctions can be decided by behaviour, such as whether information is available to the opposite hand. Others cannot, such as whether two coordinated streams differ from one stream fed by two channels.
Running the comparison
I built a small instrument to make the first kind of distinction concrete. It is a toy: it does not fit real patient data, and the numbers it uses for each model are my stipulations, chosen to express each model's qualitative prediction. Its purpose is to show which patterns of data would separate the models and how many trials that would take.
How it works
The instrument considers four kinds of trial. In localise-congruent trials, the patient locates a stimulus using the hand controlled by the hemisphere that saw it. In localise-incongruent trials, the patient uses the other hand. In compare-across trials, the patient judges whether two stimuli on opposite sides of the midline are the same. In compare-within trials, both stimuli are on one side.
Each model predicts an accuracy for each kind of trial. Chance for localising is 25 per cent, because there are four positions. Chance for comparing is 50 per cent. All four models predict about 90 per cent for congruent localising and within-field comparison, and chance for comparison across the midline. They differ in incongruent localising and in the response time gap between incongruent and congruent trials.
The two-streams model predicts chance for incongruent localising, with no time gap. The one-subject model predicts about 88 per cent, with a small gap. The switch model predicts an intermediate value of about 58 per cent, because on some trials attention is on the wrong hemisphere, with a moderate gap. The cross-cueing account predicts about 70 per cent with a large gap of about 450 milliseconds, because cueing takes time.
Given a set of trial counts and a measured time gap, the instrument computes how likely the data are under each model and converts this into a share of belief across the four, starting from equal shares.
The positive case
First I fed it a summary that follows the qualitative pattern Pinto reported. The exact counts are invented, because I do not have the trial-level data: 54 of 60 correct for congruent localising, 53 of 60 for incongruent, 31 of 60 for comparison across the midline, and a time gap of 150 milliseconds.
The instrument gave the one-subject model a share of 1.00 and the other three models 0.00 to two decimal places. The reason is simple. Incongruent localising at 88 per cent is far above what the two-streams or switch models allow, and a gap of 150 milliseconds is far below what cueing predicts.
The negative control
Next I gave it the classic pattern. It had 56 of 60 correct for congruent localising and 16 of 60 for incongruent, which is near the 25 per cent chance level. It had 29 of 60 across the midline, and no time gap. This is what a patient who fits Sperry's reading would show.
The instrument gave the two-streams model 1.00 and the others 0.00. That is the expected result, and it matters for the reading of the positive case. The instrument is not biased towards the one-subject model. It reads the classic pattern as classic.
The ambiguous case
The interesting case sits between. I gave it 40 of 60 correct for incongruent localising, about 67 per cent, which lies between the switch and cueing predictions. Without response times, the instrument split its belief: 0.31 for the switch model and 0.69 for cueing. Adding a measured time gap of 180 milliseconds moved the switch model to 1.00.
Accuracy alone cannot separate switching from cueing; timing can. This was the point of Pinto's reply, and the instrument makes it explicit. If a study reports only accuracy, a cueing explanation can always be fitted after the fact. If it reports timing too, cueing has to pay for its extra steps.
How many trials it takes
I also asked how often the instrument recovers the right model when data are generated from each model in turn, over 200 simulated patients. With response times and 20 trials per condition, it picked the true model 96 to 100 per cent of the time. Without response times, recovery at 20 trials fell to 54 per cent for the switch model and 65 per cent for cueing. It rose to 85 and 81 per cent at 60 trials, and to 96 and 94 per cent at 200 trials.
Switch, no timing (solid)Cueing, no timing (dashed)One subject, no timing (dotted)Worst model, with timing (long dash)
The lesson for design is concrete. A study of two patients with dozens of trials per condition can separate the two-streams model from the one-subject model easily, because their predictions are far apart. Separating switching from cueing needs either timing data or hundreds of trials per condition.
What the instrument cannot do
The instrument compares models that differ in behaviour. It cannot compare the two Anands of the earlier thought experiment, because they behave identically. Any model of two coordinated streams that mimics the one-subject pattern exactly would receive the same share. That is not a flaw in this instrument. It is the limit described above, now visible in the arithmetic.
It also assumes that each model makes one fixed prediction. Real models have free parameters. A two-streams model with fast subcortical coordination could be tuned to predict high incongruent accuracy with a short delay. Each tuning makes the model more flexible and less testable, which is the same complaint Pinto made about cross-cueing.
A design for the next study
The instrument suggests what a decisive behavioural study would look like. It would test every available patient with confirmed complete section, not one or two. It would use the four conditions above, with at least 60 trials each, and record response times to the millisecond. It would control eye movements with an eye tracker and discard trials in which the eyes moved before the response.
It would also add one condition that the 2017 study did not stress. In a richer-content condition, the patient would identify a word or face in one half-field by pointing with the opposite hand to a matching item. Location is the content most likely to leak through subcortical routes. Identity of a face or word is much less likely to. If opposite-hand accuracy stayed high for identity, the cueing explanation would have to carry far more than it has so far.
Finally, it would state the predictions of each model in advance, with numbers. The two-streams camp would say how fast and how accurate subcortical coordination could be. The one-subject camp would say what opposite-hand accuracy it expects for each kind of content. Then the data would pick, rather than being read afterwards by whichever camp found them more congenial.
A study designed this way would not answer the phenomenal question. But it would settle which access model is right, which is the part that behaviour can settle. It would also make the remaining question sharper, because it would leave only models that predict the same behaviour.
Reading the result in five layers
The Recursive Tattva Model, or RTM, is the framework this series uses to organise claims about consciousness. It distinguishes five layers. L1, spanda, is the dynamic activity of a system. L2, prakāśa, is manifestation: the fact that anything appears at all. L3, vimarśa, is reflexive grasp: the capacity to take what appears as one's own and act on it. L4, pratyabhijñā, is recognition: the sense of being the same self over time. L5, cit, is consciousness as ground. RTM is a hypothesis, and I use it here to separate questions that the split-brain debate tends to run together.
The operation acts directly on L1. It removes most of the fibres that carry activity between the hemispheres. The Stanford Encyclopedia puts the corpus callosum at about 200 million fibres. After surgery, the dynamics of each hemisphere continue, but the large-scale coupling between them is largely gone.
The question of how many streams there are is a question about L2. Is there one field of appearing, or two? The experiments do not reach L2 directly. What they reach is L3: whether what appears can be grasped as one's own and used to point, name or compare. On the RTM reading, Pinto's data show that L3 is partly split. Location can be grasped and acted on by either hand, but a comparison across the midline cannot be grasped at all.
The everyday reports of patients concern L4. Anand says he is the same person he was before surgery, and his wife agrees. Volz and Gazzaniga report that patients describe a single consciousness. On the layered reading, the sense of one self is a fact about recognition, not proof about manifestation. The interpreter story shows how recognition can be maintained by a narrating system even when it misreports the cause of an action.
This gives a cleaner statement of my conclusion. The data support a partial split at L3, with unified action control and divided perceptual integration. They support a unified L4, at least as reported by the speaking hemisphere. They are silent about whether L2 is one or two. And they say nothing about L5, which RTM treats as ground rather than as something a laboratory task could count.
The layers as a prediction
The layered reading does more than relabel. It predicts that the split should be greatest for operations that require integration of content, and smallest for operations that require only access to a location or a command. That is the pattern Pinto found and the 2020 consensus described: integration breaks down broadly, action control stays largely unified.
It also predicts something not yet tested well. If L4 recognition depends partly on the left-hemisphere narrator, then a task that asks the right hemisphere to recognise itself, for instance by choosing between photographs of the patient's own past, should show recognition that the speaking hemisphere cannot report. That would not show two selves. It would show recognition distributed across channels, like perception.
A contemplative experiment
The contemplative thought experiment does not prove anything about patients. It shows that L2 and L3 can be distinguished from the inside. The field that appears and the grasp that makes parts of it one's own are not the same thing, even in an ordinary brain.
A rival camp cuts against the layering
This objection cuts against RTM, and it should be taken seriously. It explains the data with fewer posits. Its weakness is that it explains the question away rather than answering it. If there is something it is like to be Anand, which he insists there is, then the count of experiencers is a real question even if behaviour cannot settle it. RTM treats that as an open empirical and philosophical question, not as a confusion.
| Analogy | What it maps to | Status |
|---|---|---|
| One house, two windows | One subject, two channels | ✓ holds One occupant, two partial views. |
| Walking between windows | Carrying content across | ✗ breaks Patients may not be able to carry it. |
| Firm with two departments | Unified action, split files | ✓ holds The firm acts as one. |
| A chief executive who reads | A subject above the channels | ✗ breaks There is no extra reader. |
| River around an island | Channels rejoin in one subject | ~ strained Assumes one source. |
What the question means for artificial systems
The split-brain debate has an obvious relevance to engineered systems, which are often built from modules that share information through narrow interfaces. A large software system may have a language component and a vision component connected by a limited channel, with a single controller choosing actions. Structurally, that resembles model four more than model one.
I take no position on whether any such system has experience, and nothing in this paper bears on machine L5. The point is narrower and methodological. If behavioural evidence cannot settle how many streams a human split-brain patient has, it will not settle the corresponding question for a modular machine either. Counting experiencers is not something that output alone can do, whether the system is biological or artificial.
What the split-brain work does offer is a template for asking better questions. It distinguishes integration of content from unity of control, and it shows that the two can come apart. Any claim that an engineered system is unified, or divided, should say which of these it means, and should specify what result would count against it.
Two cases and a control
Worked case one: Anand
Nandini runs a full battery on Anand over three sessions, following the four-condition design described above. She uses 60 trials per condition, four target positions, and records response times. His results are invented for this case but follow the pattern reported in the 2017 study.
For congruent localising, he scores 55 of 60. For incongruent localising, pointing with the right hand to targets on the left and the reverse, he scores 52 of 60. For comparison across the midline, he scores 32 of 60, close to the 50 per cent chance level. For comparison within one field, he scores 54 of 60. The response time gap between incongruent and congruent trials is 140 milliseconds.
Run through the instrument, these numbers give almost all the belief to the one-subject model. The incongruent score of 87 per cent is far from the 25 per cent chance level predicted by two streams. The time gap is too short for the cueing model's 450 milliseconds. The switch model predicts 58 per cent for incongruent localising, and Anand is well above that.
What can Nandini conclude? She can say that location information on either side is available to either hand, quickly and accurately. She can say that Anand cannot compare content across the midline. She cannot say, from these data, whether he has one field of experience or two closely coordinated fields. If she writes the report carefully, she will say that his results fit one subject with two partly separate access channels.
What Nandini writes
The report Nandini sends to Anand's neurologist is short and careful, and its wording matters. It states that perceptual integration across the midline is absent, which is the established effect of the operation. It states that location information from either half-field is available to either hand, with a response-time difference too small for overt cueing. And it states that these results are consistent with a single subject with partly separate access channels, while noting that behavioural testing cannot determine the number of streams of experience.
That last sentence may seem like hedging, but it is the most informative line in the report. It tells the reader exactly which question has been answered and which has not. A report that said simply "unified consciousness" or "two minds" would claim more than the data can bear, and would be repeated by readers who never saw the qualifications.
| Condition | Anand | Two streams predicts | One subject predicts |
|---|---|---|---|
| Same-side hand | 92% | 92% | 90% |
| Other-side hand | 87% | 25% | 88% |
| Compare across | 53% | 50% | 50% |
| Compare within | 90% | 90% | 90% |
| Time gap | 140 ms | 0 ms | 60 ms |
Worked case two: Anand with an intact commissure
Change one fact. Suppose Anand's scan showed that his anterior commissure, a small bundle connecting the temporal lobes, was spared. Volz and Gazzaniga raise this route as a possible explanation for transfer.
Now the same results have a second explanation. Information may cross through the commissure, so the two hemispheres are not fully disconnected. The one-subject reading becomes harder to separate from a two-streams reading with residual transfer. The instrument cannot help here, because it compares models of a fully disconnected brain. The anatomy has to be established before the behaviour can be read.
This is why the 2017 patients matter. Pinto's study concerned two patients with complete callosal transection. The case against cueing rests on the timing and precision data. The case against residual cortical transfer rests on the anatomy. Each patient's surgical record and imaging have to carry part of the argument.
Anand at home
The laboratory results matter less in Anand's daily life than one might expect. At the carrom board, both halves of his visual field see the board as his eyes move around it, and both hands work together. When he reads a tax return, his eyes scan line by line, so each word falls on both sides in turn. Divya has noticed only small oddities. Occasionally, his left hand picks up something he did not intend to take, and he puts it down with a laugh.
This is the pattern Sperry described in 1968, and it fits all four models. Free viewing removes the conditions that reveal the split. So daily life cannot be used as evidence for one model over another, any more than the patient's report can. The only evidence that discriminates comes from controlled tasks with brief images and fixed eyes.
The negative control
A negative control is a case where the answer is known in advance. The obvious one is a healthy volunteer with an intact corpus callosum. Her scores should be high for all four conditions, including comparison across the midline. If the task showed her at chance for comparison across the midline, the task would be broken, not her brain.
The instrument's negative control, described earlier, is complementary. When given the classic two-streams pattern, it returns the two-streams model. Together these two controls show that neither the task nor the model comparison is rigged towards the conclusion I favour.
A counterexample
Intermanual conflict is the most serious counterexample to a unified account. In some patients, one hand acts against the other: one hand buttons a shirt and the other unbuttons it, or one hand reaches for an object the other puts back. Volz and Gazzaniga note alien hand signs in some patients. The Stanford Encyclopedia describes rare interhemispheric conflict as the most serious challenge to the switch model.
If action control were fully unified, such conflict should not occur. Its existence, even if rare, suggests that at least some of the time two control systems pursue different goals. The one-subject model has to treat these episodes as failures of a usually unified system. The two-streams model treats them as moments when two agents disagree. On the data available, the rarity of conflict favours unity. The existence of conflict shows that the unity is not complete.
The failure boundary
The argument has a boundary beyond which it fails. It applies to patients with complete callosotomy, tested years after surgery, with good evidence that the anterior commissure is cut or not used. It does not straightforwardly apply to partial callosotomy, which is more common in modern practice. It may not apply soon after surgery, when the brain has not adapted. It may not apply to patients whose surgery was in childhood, whose brains may have reorganised differently.
It also depends on the stimulus. Location is the content most likely to travel through subcortical routes linked to orienting and eye movements. Shape, colour and word identity may not. A conclusion drawn from pointing to dots should not be extended without new data to naming objects or reading words. Pinto's study included identity tasks, which is why it carries weight. The boundary is that each new kind of content needs its own test.
| Condition | Does the argument hold? |
|---|---|
| Complete section, years after surgery | ✓ Holds |
| Anterior commissure spared | ~ Strained: residual transfer |
| Partial callosotomy | ✗ Does not apply |
| Soon after surgery | ~ Strained: before adaptation |
| Surgery in childhood | ~ Strained: reorganisation |
| Location only, not identity | ~ Strained: subcortical route |
The strongest case against
The strongest case against my conclusion has four parts. I will state each as fairly as I can.
First, the evidence for unified access comes mainly from two patients in one study. Split-brain patients with complete callosotomy are now rare, because the operation is used less often and usually partially. Two patients cannot represent the variety seen in sixty years of work, in which many patients showed the classic pattern clearly. If the Amsterdam patients are unusual, perhaps because of some residual connection or long adaptation, the general conclusion does not follow.
Second, the cross-cueing explanation is not refuted. Pinto and colleagues showed that overt cueing, through eye movements or posture, would be too slow. But some cueing could be fast and subcortical. A signal through the superior colliculus that tells the other hemisphere where to point could take tens of milliseconds, not hundreds. That is exactly the kind of transfer that would let two streams act as though they share location without sharing experience.
Third, the 2020 consensus review did not endorse the one-subject model. It concluded that the evidence was insufficient and that intermediate views were needed. My favoured model is one such intermediate view, but it is not the only one, and the authors who know the data best declined to choose.
Fourth, the access and phenomenal distinction cuts against any conclusion about streams of experience. If every experiment measures access, then even perfect data would leave the phenomenal question open. A defender of two streams can accept every result in this paper and say that two streams coordinate well. My conclusion survives only because I stated it as a claim about access channels and a single subject of action, not as proof of one phenomenal field.
The joke has a point. In daily life, the split rarely matters, because people adapt. Sperry noted that both hemispheres share one body and the same surroundings. A couple who learn to put important information where both hemispheres see it are solving, in practice, the problem the laboratory poses in theory.
This case is not invented from nothing. Volz and Gazzaniga describe intermanual conflict and alien hand signs. The literature I have read does not describe a formal case in which a patient's two hemispheres disagreed about medical consent. I raise it because the question of how many subjects there are is not only a matter for philosophers.
Why the classic camp may be right after all
There is a version of the two-streams model that I find harder to dismiss than its critics suggest. Suppose each hemisphere does sustain its own stream, and suppose the brain has evolved, over a lifetime of operating as one organism, strong habits of cooperation through every remaining route. After surgery, those habits continue. The right hemisphere, noticing an orienting movement, prepares the left hand; the left hemisphere, noticing the same movement, prepares the right hand. Over years, this coordination becomes fast and precise.
On this account, the Amsterdam patients would be exactly the patients in whom coordination had become most refined. Their accurate opposite-hand responses would be evidence of skilled cooperation between two streams, not of one subject. The account is consistent with Sperry's remark that the two spheres share one body and one environment, and with Schechter's view that two psychological beings can make themselves into one person.
What this account needs, to be more than a rescue, is a specific claim about speed and route. If it says that coordination occurs through the superior colliculus within 100 milliseconds, it can be tested by disrupting that route or by measuring the timing more finely. If it says only that coordination is fast enough, it is the undefined explanation that Pinto criticised.
Where the objection still bites
The strongest single objection is the second one: fast subcortical transfer. It is the one that would let a two-streams model reproduce the one-subject pattern. The instrument shows that it would need a version of cueing fast enough to look like unity in the timing data. That version is possible. It has not been shown. The burden falls on whoever proposes it to say what result would count against it.
What would change my mind
Four kinds of result would move me towards two streams. First, a group of patients with confirmed complete callosotomy who show chance-level accuracy with the opposite hand on Pinto's tasks, using the same timing and trial numbers. That would show that the 2017 result does not generalise.
Second, evidence that a subcortical route carries location information fast enough to explain the timing data, with a lesion or stimulation result showing that disrupting the route abolishes accurate opposite-hand responses. That would turn cross-cueing from a post hoc idea into a testable mechanism.
Third, simultaneous conflicting reports. If each hemisphere could give a confident, contradictory report about the same moment, with the left hand rating high confidence in one answer and the voice in another, that would be hard for any one-subject model.
Fourth, neural evidence of two independent integrated states. If measures of integration, such as those discussed in the first paper of this group, showed two separately integrated systems with no coupling, during tasks where behaviour looked unified, I would take the two-streams reading more seriously.
A fifth result would move me in a different direction, towards abandoning the question rather than answering it. If careful studies showed that every pattern of access could be produced equally well by one-subject and two-stream models with reasonable parameters, I would conclude that the count of streams is not an empirical question at all. That would be a substantive finding, and it would support the illusionist's reply more than I currently think the evidence does.
Three kinds of result would move me further towards one subject. A larger group replicating Pinto's pattern in patients with confirmed complete section. Evidence that accurate opposite-hand responses survive when eye movements are prevented. And tasks with richer content, such as words and faces, showing the same pattern as location.
How I would weigh a replication
Not every replication would count equally. A study that reproduced Pinto's pattern in patients whose anterior commissure was intact would add little, because residual transfer would remain the simpler explanation. A study in patients with confirmed complete section, using eye tracking and identity tasks, would add a great deal, especially if it was pre-registered with predictions from both camps.
Equally, a failure to replicate would need careful reading. If new patients showed the classic pattern, the question would become whether the Amsterdam patients were unusual or whether the methods differed. Variation between patients is itself informative. It would suggest that the outcome of the operation depends on individual anatomy and adaptation, which would favour intermediate models over any single universal answer.
What I still don't know
- Anatomy of the 2017 patients in full
- Trial-level timing distributions
- Replication in more patients
- Identity tasks with eye tracking
- Speed limit of subcortical transfer
- One phenomenal field or two coordinated fields
I do not know how many patients with complete callosotomy are alive and able to be tested. I do not know whether the Amsterdam patients' surgery included the anterior commissure, beyond what the published reports state. I do not know whether the timing gap of under 200 milliseconds rules out every form of subcortical signalling. And I do not know any method that could answer the phenomenal question directly.
I am also uncertain how much weight to place on my own instrument. Its model values are stipulations, and a different choice of numbers would change the share of belief in the ambiguous cases, though not in the clear positive and negative ones. The qualitative lesson, that timing separates switching from cueing, survives any reasonable choice of values. The specific percentages do not.
[AUTHOR NOTE: if you have a view on whether the phenomenal question is decidable in principle, or a contemplative observation about the unity of your own visual field, a short personal paragraph here would help readers judge how much weight to place on the RTM reading.]
Claim Ledger
| ID | Claim | Status | Evidence |
|---|---|---|---|
| MF43-C1 | Complete callosotomy prevents comparison of stimuli across the midline. | established | Sperry 1968; Pinto 2017; de Haan 2020 |
| MF43-C2 | After complete callosotomy, some patients localise stimuli in either half-field with either hand. | likely | Pinto 2017 (two patients) |
| MF43-C3 | Action control remains largely unified after the operation, while perceptual integration breaks down. | likely | de Haan 2020 consensus |
| MF43-C4 | Overt cross-cueing is too slow to explain the 2017 response-time pattern. | likely | Pinto, Lamme and de Haan 2017 reply |
| MF43-C5 | Fast subcortical transfer is not ruled out as an explanation. | established | Both 2017 papers |
| MF43-C6 | The speaking hemisphere confabulates explanations for actions it did not initiate. | established | Volz and Gazzaniga 2017 |
| MF43-C7 | Opposite-hand accuracy separates the two-streams model from the one-subject model with tens of trials. | likely | Toy instrument |
| MF43-C8 | Response timing is needed to separate switching from cueing at realistic trial numbers. | likely | Toy instrument |
| MF43-C9 | Behavioural data cannot decide between one field fed by two channels and two coordinated fields. | speculative | Argument from access measures |
Dated predictions
- MF43-P1 (by 2030): a peer-reviewed study will test at least three further patients with complete callosotomy on opposite-hand localisation and report accuracy well above chance in at least two of them.
- MF43-P2 (by 2032): a study will report that accurate opposite-hand localisation is reduced when a subcortical orienting route is disrupted or eye movements are controlled, or will report that it is not, and either result will be cited as decisive against one side of the cross-cueing dispute.
Established (solid)Likely (striped)Speculative (hatched)
Glossary
- Corpus callosum: the largest bundle of fibres connecting the two hemispheres of the brain.
- Callosotomy: surgery that cuts the corpus callosum; commissurotomy also cuts other connecting bundles.
- Anterior commissure: a smaller bundle connecting the two temporal lobes, sometimes left intact.
- Superior colliculus: a midbrain structure that helps direct the eyes and attention.
- Half-field: the half of the visual field on one side of the point of fixation.
- Cross-cueing: one hemisphere signalling to the other through the body, for example by eye movements.
- Interpreter: Gazzaniga's term for the left-hemisphere system that builds explanations.
- Access consciousness: information available for report, reasoning and control of action.
- Phenomenal consciousness: what it is like to have an experience.
- Switch model: one stream whose contents shift between hemispheres with attention.
- Partial unity: a model in which experiences can be unified in a non-transitive way.
- spanda (स्पन्द): vibration or pulsation; RTM layer L1, the system's dynamic activity.
- prakāśa (प्रकाश): light or manifestation; RTM layer L2, the fact of appearing.
- vimarśa (विमर्श): reflexive awareness; RTM layer L3, grasping what appears as one's own.
- pratyabhijñā (प्रत्यभिज्ञा): recognition; RTM layer L4, the self known as continuous.
- cit (चित्): consciousness as ground; RTM layer L5.
One subject, two doors
Divya's question about Anand was simple: is he still one person? The evidence in this paper gives a careful answer. In every sense that the law, his family and his own reports use, yes. In the laboratory, his brain no longer combines what each half of his visual field sees, and the speaking half cannot describe what the other half saw. But both hands can act on where things are, quickly and precisely, which fits one subject reaching the world through two partly separate doors.
The classic reading was not wrong about the facts it had. Information shown to one side often could not be reported by the other, and the interpreter's stories showed how readily the speaking hemisphere fills gaps. What changed was a test that let either hand answer about either side, and a result that did not fit two fully separate minds.
The dispute that followed was useful. It forced both sides to say what cueing would look like in the data, and the answer was timing. My instrument shows the same thing in a toy: accuracy separates two streams from one subject, and timing separates switching from cueing. A study with enough patients, enough trials and careful timing could settle the behavioural part of the question.
It could not settle the rest. Behaviour measures access, and any model of two well-coordinated streams can match any pattern of access. On the RTM reading, the operation splits L3 partly and leaves L4 whole as reported, while L2 remains out of reach. That is not a defeat. It is a precise statement of what we know and what we do not.
The broader lesson concerns method. The split-brain debate shows how a field can make progress on a question that it cannot fully answer. It cannot count experiencers directly. But it can separate models that differ in behaviour, specify which differences remain, and push the undecidable part into a smaller and more precise corner. That is a respectable form of progress, and it is the form most questions about consciousness are likely to take.
There is also a practical recommendation. Clinicians and researchers who describe these patients should avoid the phrase "two minds" in summaries for the public unless they mean the strong two-streams claim and can defend it. The safer and more accurate description is that the operation divides perceptual integration and leaves the person, as an agent and as a self in daily life, largely whole.
The patients deserve the last word. W.J. went home a soldier with his seizures controlled. Many patients since have lived ordinary lives. Whatever the count of streams, the operation did not produce two strangers sharing a body. It produced people who had to learn, with their families, where to put things so that both halves could see.
Sources
Classic studies and history
- Gazzaniga MS, Bogen JE, Sperry RW (1962). Some functional effects of sectioning the cerebral commissures in man. PNAS. https://pubmed.ncbi.nlm.nih.gov/13946939/
- Sperry RW (1968). Hemisphere deconnection and unity in conscious awareness. American Psychologist. https://pubmed.ncbi.nlm.nih.gov/5682831/
- PNAS. Feature on W.J. and the 1962 block design film. https://www.pnas.org/doi/10.1073/pnas.1422335112
- Wolman D (2012). A tale of two halves. Nature. https://www.nature.com/articles/483260a
- Gazzaniga MS. The split-brain: rooting consciousness in biology. PNAS. https://pmc.ncbi.nlm.nih.gov/articles/PMC4280607/
- Gazzaniga MS (2011). Neuroscience and justice. Edge Master Class. https://www.edge.org/conversation/michael_gazzaniga-neuroscience-and-justice-edge-master-class-2011
The 2017 dispute
- Pinto Y, Neville DA, Otten M, et al. (2017). Split brain: divided perception but undivided consciousness. Brain. https://pubmed.ncbi.nlm.nih.gov/28122878/
- Volz LJ, Gazzaniga MS (2017). Interaction in isolation: 50 years of insights from split-brain research. Brain. https://academic.oup.com/brain/article/140/7/2051/3892700
- Pinto Y, Lamme VAF, de Haan EHF (2017). Cross-cueing cannot explain unified control in split-brain patients. Brain. https://academic.oup.com/brain/article/140/11/e68/4157882
- de Haan EHF, Corballis PM, Hillyard SA, Marzi CA, Seth A, Lamme VAF, Volz L, Fabri M, Schechter E, Bayne T, Corballis M, Pinto Y (2020). Split-brain: what we know now and why this is important for understanding consciousness. Neuropsychology Review. https://pubmed.ncbi.nlm.nih.gov/32399946/
Philosophy
- Nagel T (1971). Brain bisection and the unity of consciousness. Synthese. http://link.springer.com/article/10.1007/BF00413435
- Bayne T (2008). The unity of consciousness and the split-brain syndrome. Journal of Philosophy. https://philarchive.org/rec/BAYTUO-2
- Schechter E (2018). Self-Consciousness and "Split" Brains: The Minds' I. Oxford University Press. https://academic.oup.com/book/10111
- Brook A, Raymont P. The unity of consciousness. Stanford Encyclopedia of Philosophy. https://plato.stanford.edu/entries/consciousness-unity/
Cite this work
Mahapatra, R. R. One person, two windows: does cutting the corpus callosum make two minds, or one mind with two doors? Mind Frontier, paper 43. https://rajeshrmahapatra.com/papers/mind-frontier/U0852-split-brain-two-streams-or-one-subject. DOI: to be assigned.
@misc{mahapatra_mf43,
author = {Mahapatra, Rajesh Ranjan},
title = {One person, two windows: does cutting the corpus callosum make two minds, or one mind with two doors?},
howpublished = {Mind Frontier, paper 43},
url = {https://rajeshrmahapatra.com/papers/mind-frontier/U0852-split-brain-two-streams-or-one-subject},
note = {DOI: to be assigned}
}