Binocular rivalry without button presses: how much prefrontal activity was only the report?
When perceptual switches are read from the eyes instead of the hands, most of the frontal fMRI signal tied to those switches disappears, yet the prefrontal cortex still carries the content of what is seen.
TLDR
- Claim: in the key human study, the switch-related differential activity in the middle frontal gyrus vanished once observers stopped pressing buttons, rather than merely shrinking. Occipital and parietal activity tied to switches remained.
- Mechanism: fMRI contrasts measure the average change in a region at the moment of a switch. Reporting adds a large frontal transient at that moment. Content coding by mixed populations of neurons can change without changing the average at all.
- Implication: most switch-locked prefrontal BOLD activity in classic rivalry studies belonged to noticing and reporting switches. Monkey recordings show that prefrontal neurons still encode the seen percept without any report, so the region did not fall silent.
- Boundary: a vanished contrast is a statement about one signal at one threshold. It cannot show that the prefrontal cortex plays no part in experience, and eye-movement read-outs agree with reports only about four-fifths of the time.
- Decision: treat report-free rivalry results as strong evidence against frontal switch transients as the cause of perception. Do not treat them as evidence about prefrontal content coding until a study measures both in the same people.
The frontal blob that would not hold still
Kavya Raman is a 31-year-old doctoral student at a university imaging centre in Mysuru. Her thesis is about binocular rivalry, and her supervisor, Professor Iqbal Siddiqui, has a favourite slide. It shows a bright orange patch in the right frontal lobe that lights up every time a volunteer’s perception flips. He calls it “the place where the switch is decided”. Kavya and Professor Siddiqui are invented for this paper. The studies she reads, and the numbers in them, are real.
Kavya has run a pilot with twelve volunteers. Each lay in the scanner wearing glasses that showed a red grating drifting left to one eye and a green grating drifting right to the other. Nobody sees both at once. Perception flips back and forth between red and green every second or two, while the screen stays exactly the same. The volunteers held down a button for red and another for green. Her frontal patch looked just like her supervisor’s slide.
Then she read a 2014 paper that tracked the same flips with no buttons at all, using the eyes alone. In that study the frontal difference disappeared, and her thesis committee meets in six weeks. She must preregister her main experiment before then, and she must decide what claim it tests. How much of the prefrontal activity tied to perceptual switches is the switch, and how much is the report of it?
That is the question of this paper, put in Kavya’s terms. The standard version asks how much prefrontal BOLD activation disappears when switches are tracked by optokinetic nystagmus and pupil size instead of button presses. My answer, which the rest of the paper earns, has two halves. Most of the switch-locked frontal BOLD contrast disappears, and in the main human study the largest frontal cluster went from present to absent. But the prefrontal cortex does not go quiet. In monkeys, its neurons still carry the content of what is seen when nothing is reported.
I will return to Kavya in every part, because her stakes are concrete. If she preregisters “prefrontal cortex drives the switch”, she is testing a claim the evidence has largely retired. If she preregisters “prefrontal cortex does nothing”, she is testing a claim the evidence already contradicts. The useful claim lies between, and finding it is what this paper is for.
Reading perception from the eyes
Binocular rivalry happens when each eye receives a different image at the same place. The brain does not blend them into a muddle. It shows you one, then the other, in turns that last a second or a few seconds. The input is constant and the experience changes, which is why consciousness researchers love it. Any brain signal that follows the experience, and not the input, is a candidate correlate of what is seen.
True story · Wheatstone’s mirrors
In 1838 Charles Wheatstone published a long paper on binocular vision in the Philosophical Transactions of the Royal Society. In it he presented the mirror stereoscope, an instrument that sent a separate picture to each eye. With matched pictures it produced depth. With different pictures it produced something stranger, which he also described: the two did not fuse, and one would give way to the other. The instrument built to show how the eyes cooperate also showed how they compete. For most of the time since, the only way to know which picture an observer saw was to ask.
Modern experiments ask with buttons. Hold the left key while you see red, the right key while you see green, and release both when the picture is mixed.
Buttons create a problem that took decades to take seriously. Every switch now comes with a decision, a motor command and a moment of introspection. If a frontal area lights up at the switch, it might be causing the switch. It might also be noticing the switch, deciding what it was and moving the finger. Naotsugu Tsuchiya and colleagues set out this concern in a 2015 review in Trends in Cognitive Sciences, arguing that “the requirement of reports has biased the search for the neural correlates of consciousness”.
Two reflexes that report for you
The fix is to find a signal that follows the percept without the observer doing anything. Two objective measures have proved useful. The first is optokinetic nystagmus, usually shortened to OKN. When you watch a pattern drifting across your view, your eyes follow it slowly, then flick back, then follow again. The slow phase goes in the direction of the motion you perceive. If the red grating drifts left and the green drifts right, the direction of the slow phase tells you which one is dominant. Randolph Fox and colleagues showed this in a 1975 Vision Research paper, titled “Optokinetic nystagmus as an objective indicator of binocular rivalry”.
The second is pupil size. The pupil constricts more for a brighter percept, so if the two images differ in brightness, the pupil tracks which one is seen. Wolfgang Einhäuser and colleagues found in a 2008 PNAS study that “pupil diameter increased just before the reported perceptual switch”. Marnix Naber, Stefan Frässle and Einhäuser then used both reflexes together in a 2011 PLoS ONE paper. They found that the reflexes “reflect the percept rather than the physical stimulus”, and that they respond to transitions with shorter latencies than the subjective report.
Analogy · the speedometer cable
Old cars measured speed with a cable spun by the gearbox. The dial never looked at the road. It read a mechanical consequence of the car’s motion, and that was good enough to drive by. OKN does the same for rivalry: it reads a mechanical consequence of the percept, the drift of the eyes, without asking the driver anything. Where it breaks: a speedometer cable is tied to the wheels by gears and cannot drift. The eyes can follow a mixed percept, lag a switch or miss a brief one, so the read-out is a good estimate and not a gear-locked copy.
That last point matters, and it is easy to skim past. An objective read-out is only as good as its agreement with the thing it stands for. If OKN agreed with perception perfectly, a no-report experiment would be a clean subtraction: same perception, minus the report. In practice the agreement is partial, and part of the disagreement comes from the report itself, which lags the eyes. I will put numbers on this in Part two.
What a no-report contrast actually measures
Here is the logic of a no-report fMRI study in its simplest form. Scan people during rivalry while they report with buttons, and find regions whose activity changes at each switch. Then scan them during rivalry without reports, find the switches from OKN or the pupil, and run the same analysis. Regions whose switch-related activity survives are candidates for perception, while regions whose activity disappears were tied to reporting.
Two things hide inside that logic, and the first is what “switch-related activity” means. An fMRI contrast of this kind asks whether a region’s average signal goes up or down around the moment of a switch. It does not ask whether the region represents which image is seen. A region can carry the content of a percept perfectly while its average signal stays flat across switches. I will show why in Part three.
The second hidden thing is the comparison. Most rivalry studies compare genuine rivalry with replay, in which the two images are physically swapped on the screen to mimic the observer’s switches. Replay removes the internal competition and keeps the visual change. A region more active at rivalry switches than at replay switches is doing something specific to the internal event. That is the contrast Erik Lumer, Karl Friston and Geraint Rees used in their 1998 Science paper, which found frontoparietal activity “specifically associated with perceptual alternation only during rivalry”. That paper shaped two decades of thinking, and it is also the paper behind slides like Professor Siddiqui’s.
Now apply this to Kavya. Her pilot used buttons, so her frontal patch is the 1998 result again: a frontoparietal transient at rivalry switches that exceeds replay. The no-report question is whether that patch survives when the buttons go. To answer it she needs the eyes to tell her when each switch happened, and she needs to know how often the eyes are wrong.
Why one slide carries so much weight
The answer matters beyond Kavya’s committee because the main theories of consciousness disagree precisely about the front of the brain. Global workspace theories expect prefrontal participation whenever a content becomes conscious, while posterior theories expect the prefrontal contribution to be optional, a matter of using experience rather than having it. A switch-locked frontal transient in rivalry was one of the most frequently cited pieces of evidence for the first family. If that transient was mostly generated by the act of reporting, the evidence changes sides, or at least stops counting for either side. The rivalry case is therefore a test of method as much as a test of anatomy, because it asks whether an experimental subtraction can separate experience from the machinery we use to describe experience.
What vanished and what stayed
The study Kavya read is by Stefan Frässle, Jens Sommer, Andreas Jansen, Marnix Naber and Wolfgang Einhäuser, published in the Journal of Neuroscience in 2014. Its title states its conclusion: “Binocular rivalry: frontal activity relates to introspection and action but not to perception”. It is the closest thing the field has to a direct answer to the question of this paper, so it repays a slow reading.
The design
Twenty observers took part, nine men and eleven women aged 19 to 30. There were two kinds of rivalry. In the OKN version, a red grating drifting left went to one eye and a green grating drifting right went to the other, each moving at 22.3 degrees per second. In the pupil version, the gratings were stationary and tilted 60 degrees in opposite directions, and the pupil tracked which was seen. Each kind of rivalry was crossed with two report conditions, active report and passive viewing, and with a replay control. That made eight conditions, each run four times.
In active report, observers held one of two buttons for as long as each percept lasted. In passive viewing, they simply watched, and in both conditions the eyes were recorded throughout. The design lets the same objective measure stand in for perception in both conditions, so the only planned difference between them is the act of reporting.
How well the eyes agreed with the hands
During rivalry with active report, OKN predicted the button press 82 ± 2 per cent of the time, and pupil size did so 59 ± 1 per cent of the time. During replay, where the switches were physical, agreement rose to 88 ± 2 per cent for OKN and 83 ± 1 per cent for the pupil. All four values were well above the 50 per cent expected by chance. The objective transitions came before the manual report, by 0.52 ± 0.07 seconds for OKN and 0.42 ± 0.12 seconds for the pupil.
Read those numbers with care. Agreement of 82 per cent does not mean the eyes were wrong 18 per cent of the time. Part of the gap is the half-second lag of the hand behind the eyes, which counts as disagreement in every switch. Part is mixed percepts, which the buttons and the eyes handle differently. Naber and colleagues had already found that the reflexes “show an abundance of short dominance durations” that observers fail to report. On this reading, some of the disagreement is the report being wrong, and the eyes being right.
The pupil is a weaker instrument for rivalry, at 59 per cent. That is above chance, but a read-out that agrees with report only three times in five is a blunt tool for timing switches. Any study that relies on the pupil alone will find smaller switch-related effects everywhere, because its switch times are noisier. I will come back to this when I run the instrument.
The result
With active report, the objective measures and the buttons told the same story. Occipital, parietal and frontal areas showed activity tied to rivalry switches, replicating the classic findings. Then came the passive condition. In the paper’s words, “differential neural activity in frontal areas was absent, whereas activation in occipital and parietal regions persisted”. The most extended clusters that were greater with report than without were in the bilateral middle frontal gyrus, along with bilateral supplementary motor areas, bilateral insula, middle occipital cortex and the left superior and inferior parietal lobules.
The key sentence is this one: “neural activation in the middle frontal gyrus was not only diminished but completely vanished during passive-viewing trials”. The authors also report that the absence of frontal differential activity held for both kinds of rivalry. In the passive comparison of rivalry against replay, one frontal region did appear, the right inferior orbitofrontal lobe. The paper does not list it among the regions that differed between report and no report.
So the first answer to Kavya’s question is direct. For the middle frontal gyrus, the region most often cited as the frontal switch signal, the answer to “how much disappears” is all of it, at the paper’s statistical threshold. Supplementary motor and insular activity also depended on report. Occipital and parietal switch-related activity survived.
Making switches unreportable
A year later, Jan Brascamp, Randolph Blake and Tomas Knapen attacked the same question from another side, in a 2015 Nature Neuroscience paper. Instead of removing the report, they made the switches themselves hard to notice. Each eye saw a field of moving dots with the same statistics but different individual dots. In one condition the two eyes’ dots had different colours, so a switch looked like a colour change. In the other they had the same colour, so a switch changed which eye’s dots were seen, with no visible sign that anything had happened.
The switches were real in both conditions, as the authors showed with brief motion probes. In a dual task, six observers detected same-colour switches about eight times less often than different-colour switches. Their switch sensitivity in the same-colour condition was not distinguishable from chance. In the scanner, frontoparietal responses to the invisible switches were minimised, by about as much as their reportability. The authors concluded that these responses “reflect the switches’ consequences rather than their cause”.
The two studies make a strong and complementary pair. Frässle removed the report and kept the switches noticeable, whereas Brascamp kept the task and made the switches unnoticeable. Both found that frontal switch-related activity tracks whether the switch is noticed and reported. Earlier work by Knapen and colleagues in 2011 had already shown that right frontoparietal activity is larger for longer transitions, and that replay matched rivalry when replayed transitions had their real durations. And Jiaxiang Zou, Sheng He and Peng Zhang found in a 2016 PNAS study that rivalry can occur between invisible patterns, with “minimal engagement of frontoparietal cortex”.
For Kavya, the practical consequence is that her pilot’s frontal patch has a straightforward explanation that needs no causal role in perception. It appears when her volunteers notice and report switches. A preregistration that predicts it will survive without report is predicting against two direct tests.
What this is for: seeing which of four legs each study covers. What it is not: a quality score; a study can be excellent and still answer a different question.
With JavaScript off, all six notes:
- Lumer 1998: Classic fMRI with report, comparing rivalry with replay. Frontoparietal switch activity found. Report and noticing both present.
- Frässle 2014: Removed report with OKN and pupil read-outs, kept a replay control. The middle frontal effect vanished. Switches were still noticeable.
- Brascamp 2015: Made switches unnoticeable with matched dot fields and checked they were real with probes. Frontoparietal responses shrank in step.
- Zou 2016: Rivalry from patterns observers could not see, so the competition ran without anything noticeable. A different question from frontal BOLD.
- Kapoor 2022: Macaque prefrontal units without report, with physical alternation. Percept decoded on single trials. Two animals, no fMRI comparison.
- Needed study: Report, noticing and physical control removed in turn, in humans, with both the transient and content decoding measured. Not yet run.
The audit above scores each study on four things a no-report rivalry study can remove or measure. They are the report, the noticing of the switch, the replay control and recordings of content at the single-neuron level. No study covers all four, since the human fMRI studies removed the report and kept the replay control. The monkey studies recorded content and had no human-style report to remove. The study Kavya could run would combine them.
Below the threshold of a scanner
An fMRI voxel is a few millimetres across and holds hundreds of thousands of neurons. Its signal reflects their combined metabolic demand, smoothed over several seconds by the slow response of blood flow. That makes it good at detecting when a whole region works harder. It is poor at detecting when the same amount of work is shuffled among the region’s neurons. To see that, you need electrodes.
True story · the monkeys who pulled levers
In the 1990s, David Leopold and Nikos Logothetis trained monkeys to report rivalry by pulling levers, then recorded single neurons as they did so. Their 1996 Nature paper found that many cells, “particularly in V4”, followed the monkey’s reported percept rather than the unchanging stimulus. Two years later, Frank Tong and colleagues found in a 1998 Neuron study that the human face and place areas responded during rivalry as strongly as during physical alternation. These studies placed percept-following activity in the visual system, and both used reports, which is why the no-report question came later.
Content without report, in the prefrontal cortex
The decisive recordings for this paper come from the same Tübingen tradition. In a 2022 Nature Communications paper, Vishal Kapoor, Abhilash Dwarakanath and colleagues recorded from the lateral prefrontal cortex of two rhesus macaques with implanted 96-electrode arrays. They collected 987 units across six sessions. The monkeys saw upward and downward drifting gratings in rivalry and were never trained to report. Their percept was inferred from OKN, as in the human studies.
The result runs against a simple reading of the human fMRI work. Feature-selective prefrontal neurons were “modulated concomitantly with subjective perception” during both externally induced and spontaneous switches. Decoders trained on the population could tell which grating was perceived on single trials. Within a recording session, mean decoding accuracy was 86.35 per cent for spontaneous dominance during rivalry and 94.61 per cent for physical alternation. In control experiments, stimulus content could still be decoded when the eye movements used to infer perception were suppressed.
Across sessions, generalisation was weaker, at 57.70 per cent for rivalry dominance. That is above chance but far from the within-session figure. Chronic arrays can drift, and the same electrode may record different neurons on different days, so this is a limit of the method rather than evidence against content coding. It is still a number to keep in mind before claiming that the code is stable.
The authors’ own conclusion is careful: internally generated changes in conscious content “are reliably reflected within the activity of prefrontal populations in the absence of volitional reports or changes in sensory input”. The phrase “reliably reflected” does not say “caused”. It also does not say “not caused”; it says the information is there.
How both results can be true
So the human scanner says frontal switch activity vanishes without report, and the monkey electrodes say frontal neurons still track the percept without report. These look contradictory, but they are not contradictory once you see what each method measures.
This thought experiment is the paper’s minimal worked case, and it is small enough to do by hand. With equal groups, a perfect switch produces zero change in the total, so the switch-related BOLD effect is zero. Now add a brief burst at each switch, say 10 extra spikes per second for half a second in both groups, because something in the region registers that a change happened. The total rises from 25 to 45 for half a second, then returns to 25. The scanner sees that burst, smeared over several seconds by blood flow. Remove the burst, and the scanner sees nothing, while an electrode still sees the content flip.
That is my proposed reconciliation. The classic frontal switch signal is mostly a transient, a burst that marks the event of a change. Reporting enlarges it a great deal, because reporting requires noticing, deciding and acting at that moment. Content coding is a different kind of signal, because it reshuffles activity among neurons and can leave the average untouched. The human studies removed the transient, and the monkey studies found the content.
Analogy · crowd noise at a match
A microphone above a cricket ground measures how loud the crowd is. It cannot tell you which team the crowd is supporting. If half the stands cheer for each side, the score can change hands without the noise level changing, while a person walking among the seats would hear exactly who is winning. A BOLD signal is the microphone, and an electrode array is the person in the stands. Where it breaks: a crowd roars at every wicket, whichever side takes it. That roar is the transient, and it is real information. The analogy hides that the transient itself might do work, for instance by updating other regions.
The state of the region, as well as its content
A year later the same group added a second layer. In a 2023 Neuron paper, Dwarakanath and colleagues analysed local field potentials in the prefrontal cortex during no-report rivalry. They found that fluctuations between low-frequency activity, at 1 to 9 hertz, and beta activity, at about 20 to 40 hertz, “reflect competition between spontaneous updates and stability of conscious contents”. Low-frequency events came just before spontaneous transitions, and beta synchronised the neurons coding the seen content.
In the Max Planck Society’s press release on the study, Dwarakanath was cautious: “we cannot decode the content of consciousness from the waves”. He added that the data cannot show whether the wave patterns cause the switch or merely precede it. The senior author, Theofanis Panagiotaropoulos, called the results a “refinement of the global workspace theory”.
Taken together, the monkey work says three things about the prefrontal cortex during report-free rivalry. It holds content-specific populations that follow the percept, and it shows state fluctuations that come before switches, and none of this requires a report. What the human fMRI work removed was something else: a large, report-dependent transient in the middle frontal gyrus.
A caution is needed on the comparison itself. The monkey recordings were in ventrolateral prefrontal cortex, and the human cluster that vanished was in the middle frontal gyrus. These are neighbouring parts of the frontal lobe, and the mapping between macaque and human frontal areas is not exact. The reconciliation I propose is about kinds of signal. It does not require the two studies to have measured the same patch of cortex, and they almost certainly did not.
For Kavya, this changes what a good thesis claim looks like. “Frontal cortex shows a switch transient without report” is the claim the human evidence weighs against. “Frontal cortex holds a percept code without report” is the claim the monkey evidence supports, and fMRI contrasts of the classic kind cannot test it. She needs a different analysis, one that asks whether the pattern across voxels predicts the percept, instead of whether the average jumps at each switch.
What each theory needed to see
The no-report rivalry results are often presented as a win for one side of a long argument. On one side are theories that tie consciousness to the front of the brain. On the other are theories that place it in posterior sensory cortex. The argument is real, but the results do less damage to either side than the headlines suggest. Each theory needed something specific from these experiments, and it helps to ask what that was.
Global workspace
Global workspace theory holds that a content becomes conscious when it is broadcast across a network of frontal and parietal areas, so that many systems can use it at once. Stanislas Dehaene and colleagues set out a testable version in a 2006 Trends in Cognitive Sciences paper, describing conscious access as a sudden, self-sustaining “ignition” of that network. A prefrontal switch transient fits the theory naturally, since ignition at every switch should look like a burst in frontal cortex.
So the vanished transient is a problem for the theory, but a smaller one than it seems. A workspace theorist can say that ignition during rivalry is content-specific, sustained through each dominance period and carried by populations of neurons. That is what Kapoor and colleagues found. The report transient, on this reading, was the extra ignition needed to turn a percept into an action. Panagiotaropoulos, a workspace sympathiser, reads the 2023 state-switching results as a refinement of the theory rather than a refutation.
Posterior theories
Victor Lamme’s recurrent processing theory, set out in a 2006 paper in the same journal, places consciousness in recurrent loops within sensory cortex. Integrated information theory also points to a posterior hot zone. Both predicted that frontal activity during rivalry would turn out to be about report. The human fMRI results are exactly what they predicted, and Lamme was a co-author of the 2015 no-report review.
But the posterior theories also needed something they did not get. They needed the prefrontal cortex to have no content-specific role without report. The macaque decoding results do not show that frontal content coding is necessary for consciousness. They do show that the frontal cortex knows what is seen, without report, on single trials. A posterior theorist must say this is a copy sent forward for possible later use. That is a reasonable thing to say, but it is a claim, not a finding.
Higher-order theories
Higher-order theories hold that a state is conscious when the brain represents itself as being in it. Hakwan Lau and David Rosenthal reviewed the evidence in a 2011 paper, pointing to prefrontal areas as the likely seat of those higher-order representations. For these theories, a report-dependent frontal transient is ambiguous. It could be the higher-order state, which would make removing the report a way of removing consciousness itself. Or it could be the motor and decision machinery downstream of it.
Matthias Michel and Jorge Morales pressed this kind of point in a paper in Mind and Language, arguing that the evidence against prefrontal involvement is weaker than often claimed. A higher-order representation need not be large, and it need not change the average activity of a region. A few neurons tagging the current percept as “mine, now” would be invisible to the contrasts Frässle used.
Predictive processing
Predictive theories treat rivalry as the brain alternating between two hypotheses, neither of which can explain away all the input. Veith Weilnhammer and colleagues tested a version of this in a 2021 Current Biology paper. They found that inferior frontal cortex was active in detecting perceptual conflict, and that disrupting it with transcranial magnetic stimulation “slowed down the updating of conscious experience”. They concluded that prefrontal cortex is “causally involved in determining the contents of conscious experience”.
That result matters because it uses causal disruption, not correlation, and because it points to a frontal region that no-report studies did not clear. The right inferior orbitofrontal activity that remained in Frässle’s passive condition sits close to this territory, though not in exactly the same place. A frontal role in conflict detection would produce a small transient at switches that survives without report. On the evidence so far, that transient is either small or absent in the middle frontal gyrus, and present in inferior frontal regions under some conditions.
The no-cognition objection
Ned Block raised a deeper worry in a 2019 paper titled “What is wrong with the no-report paradigm and how to fix it”. Even without report, observers “might engage in post-perceptual cognitive processing”. They may still notice switches, think about them and remember them. Removing the button removes one consequence of cognition, without removing cognition itself. Block argued for a “no-cognition” paradigm, in which observers have no reason to think about the rivalry at all.
This objection cuts in an interesting direction. If passive viewers still noticed switches, the residual cognition should have produced some frontal activity, and in the middle frontal gyrus it did not. So either passive viewers did not think much about the switches, or the noticing they did was too small to see. Brascamp’s unnoticeable switches come closest to Block’s fix, and they gave the same answer.
Where the RTM layers sit
The RTM hypothesis is a working map of five layers, treated in this series as a hypothesis under test. It is not a finding, and here is how the rivalry evidence maps onto it. L1, spanda, is the alternation itself: the competition and its rhythm, located mainly in visual cortex, with prefrontal state fluctuations as a candidate pacemaker. L2, prakāśa, is the manifest content: which grating is seen, carried by percept-following populations in visual, temporal and prefrontal cortex.
L3, vimarśa, is reflexive awareness: in rivalry, the registering that “it changed”. The report-dependent frontal transient is the best candidate for a measurable L3 event. L4, pratyabhijñā, is recognition, the step from “it changed” to “that is red, press the left key”. L5, cit, is outside the reach of any experiment in this paper.
The RTM finding of this paper follows from that mapping. No-report designs work as an L3 subtraction. They remove the prompted, reportable kind of reflexive noticing and leave L1 and L2 largely intact. But Block’s point suggests that L3 has two parts. One is prompted by the task and disappears without report. The other is unprompted monitoring, which may continue whenever anything changes. The evidence suggests that vimarśa in rivalry is mostly called up by the task, not running constantly. That is a claim about L3, and the decisive study in the section on changing my mind is designed to test it.
For Kavya, the theories give her a way to make her preregistration useful whatever the outcome. If she predicts that a frontal percept code survives without report while the switch transient vanishes, a positive result helps the workspace and higher-order camps. A null result on the code helps the posterior camp. Either way the experiment tells the field something it does not yet know.
One further change to Kavya’s situation sharpens the point. Suppose her supervisor insists on keeping the buttons, because the committee distrusts eye-movement read-outs. Every theory above then predicts the same thing for her pilot, a frontal transient at the switches, and her experiment can no longer separate them. The decision about the buttons is therefore a theoretical decision disguised as a practical one. Removing them is what gives the four families different predictions, and adding a silent-noting condition is what gives the higher-order and workspace families different predictions from each other.
One cognition at a time
Binocular rivalry is a modern laboratory effect, but the claim behind it is old. The mind, it says, does not hold two incompatible things at once. Two Indian traditions made versions of that claim central to their theories of mind, and each made a further distinction that bears on the no-report question. I use them here as structured hypotheses about how cognition is organised, not as evidence about brains.
The Nyāya mark of the mind
The Nyāya Sūtra, the founding text of the Nyāya school of logic, lists the objects of valid knowledge in its first chapter. Its sixteenth sūtra gives the sign by which the mind, manas, is known. In the Sanskrit text on the GRETIL edition with Vātsyāyana’s commentary, it reads: yugapajjñānānutpattir manaso liṅgam. My rendering is: “the non-arising of simultaneous cognitions is the mark of the mind”.
Text · Nyāya Sūtra 1.1.16, with Vātsyāyana
युगपज्ज्ञानानुत्पत्तिर्मनसो लिङ्गम्
The commentary explains the argument. Even when the senses are in contact with their objects at the same time, smell with odour and the rest, “cognitions do not arise simultaneously” (yugapajjñānāni notpadyante). From this, the commentator infers an extra factor, which must join each sense in turn for a cognition to arise. That factor is manas. The translation here is mine and should be checked against Ganganatha Jha’s English version, which Wisdom Library hosts; I have not verified Jha’s wording for this sūtra, so treat it as to verify.
This is a bottleneck theory. Information is available in parallel at the senses, but cognition is serial, because one inner instrument must attend to each channel. Rivalry is a vivid case: two images are present at the retinas, and only one becomes a cognition at a time. The Nyāya account also predicts something the no-report work bears out. If manas is the bottleneck, the bottleneck is upstream of any report. The alternation should continue whether or not anyone speaks of it, and in the eyes of Frässle’s passive observers it did.
Patañjali on the mind and its object
The second tradition makes a finer and more useful cut. Yoga Sūtra 4.20 reads ekasamaye cobhayānavadhāraṇam. In the translation of Hariharānanda Āraṇya given on Wisdom Library’s study page for the sūtra: “Besides, both (the mind and its objects) cannot be cognised simultaneously”. The same page summarises the Bhāsvatī commentary with an example: hearing a sound and “I am aware that I am hearing a sound” do not take place at the same moment.
Text · Yoga Sūtra 4.20, with the Bhāṣya
एकसमये चोभयानवधारणम्
The Bhāṣya attributed to Vyāsa, in the GRETIL text, begins: na caikasmin kṣaṇe svapararūpāvadhāraṇaṃ yuktaṃ. My rendering is: “and it is not reasonable that in a single moment there should be a fixing of both its own form and another’s”. The argument is aimed at Buddhist momentariness, and the sūtra uses it to show that the mind is not self-luminous. I borrow only the distinction, between cognising an object and cognising the cognising.
That distinction is the no-report problem in miniature, since seeing red is a cognition of an object. Noticing “I now see red, so I should press the left key” is a cognition of the cognition. Patañjali’s claim is that these do not happen in the same moment. The modern data fit a version of this. The report lagged the eyes by about half a second in Frässle’s study, and the frontal transient tied to reporting appeared only when the second kind of cognition was required.
Both traditions also insist on something that a neuroscientist might pass over. The bottleneck is not the self. For Nyāya, manas is an instrument, distinct from the self that knows through it. For Yoga, the mind is an object seen by a witness, puruṣa, that does not itself fluctuate. In RTM terms, both put the serial alternation at L1 and L2, the reflexive cognition at L3, and something unmoving beyond them at L5. Neither tradition gives a reason to expect that removing the report would remove the experience. Both would expect it to remove the second cognition.
The traditions disagree with each other in ways that matter here. Nyāya treats the self as the real knower and manas as its tool. Patañjali’s Yoga treats the mind as unconscious matter that only appears aware because the witness illuminates it. Buddhist Abhidhamma, the target of the Bhāṣya’s argument, denies any lasting self and treats each moment of cognition as complete in itself. These are three different pictures of what the serial bottleneck serves. A rivalry experiment cannot decide between them, and I do not claim it can.
For Kavya, the practical yield is small but real. Her passive condition should specify what observers are told to do inside their heads. “Just watch” leaves room for silent noting, and some volunteers will do it. A short instruction to avoid labelling, plus a debriefing question afterwards, costs little. It would also let her test whether silent noting brings back part of the transient, which is a question the field has not asked.
A toy scanner with three signals
The reconciliation in Part three is an argument about kinds of signal. Arguments like that are easy to state and easy to get wrong, so I built a small simulation to check it. The instrument is a toy, and it can show only what follows from its assumptions. Its job is to make those assumptions explicit and to find where the argument stops working.
What the model contains
The model simulates twenty observers, each doing four runs of two minutes of rivalry. Dominance periods are drawn from a gamma distribution with a mean of 1.6 seconds, close to the durations Frässle reported. A simulated prefrontal region has three components. A report transient fires about half a second after each switch, only when the observer is reporting, with a size that varies between observers. An update transient fires at every switch, report or not, and is small by default, at 0.15 of the report transient. A content code of 80 units, half preferring each percept, follows the current percept with noise.
The region’s total activity is passed through a standard haemodynamic response and sampled every two seconds, with noise, to give a BOLD signal. Switch times are then estimated by a simulated OKN read-out. It sees each true switch with a random timing error, a standard deviation of 0.35 seconds by default, and it misses most dominance periods shorter than 0.4 seconds. The switch-locked BOLD effect is estimated by regression, exactly as a no-report fMRI study would. Separately, a simple decoder is trained to read the current percept from the 80 units.
The core of the model (Python, about 140 lines in full)
# one observer, one run, 50 ms steps
neur = update * box(true_switches, 0.6 s) # every switch
if reporting:
neur += gain * box(true_switches + 0.5 s, 0.8 s) # report burst
rates = 1 + 0.5 * outer(percept, pref * sel) + noise # content units
neur += 0.05 * (rates.mean(axis=1) - 1) # their average
bold = convolve(neur, hrf)[::TR] + noise
okn_switches = true_switches + normal(0.1, jitter) # read-out
beta = regress(bold, convolve(box(okn_switches), hrf)) # switch-locked
decoder: nearest-centroid on rates, labels from the OKN read-out,
scored against the OKN labels and against the true perceptThe output
I ran five conditions with random seed 38, and the first is active report. The second is passive viewing, identical except that the report transient is switched off. The third is a negative control that uses the passive data with switch times shuffled at random. The fourth is a passive condition with a large update transient, 0.8 instead of 0.15. The fifth is passive viewing with a much noisier read-out, tuned to give pupil-grade agreement of about 60 per cent.
seed 38, 20 simulated observers x 4 runs x 120 s, 80 content units
condition agree beta t dec(OKN) dec(true)
active report 0.824 1.4296 12.84 0.813 0.952
passive, no report 0.829 0.1063 5.46 0.823 0.955
shuffled timings (control) 0.828 0.0117 0.97 0.820 0.957
passive, large update 0.829 0.6709 26.40 0.823 0.955
passive, pupil-grade labels 0.601 0.0729 4.27 0.597 0.937
share of switch-locked beta that disappears without report: 92.6%“Agree” is the share of time the OKN read-out matches the true percept. Beta is the switch-locked BOLD effect, and t is its group t-value across the twenty simulated observers. “Dec(OKN)” is decoding accuracy scored against the read-out’s labels, which is all a real experiment can do. “Dec(true)” is scored against the true percept, which only a simulation can know.
The default read-out agrees with the true percept 82 per cent of the time, close to Frässle’s OKN figure. I tuned the timing error to give that agreement deliberately. Without report, the switch-locked beta falls from 1.43 to 0.11. That is a 92.6 per cent reduction, and it is almost all the report transient. Meanwhile, decoding of the percept from the content units is unchanged: 0.95 against the true percept in both conditions. The region keeps its message while losing its burst.
Two features of this result deserve care. First, the passive beta is still detectable in the model, at t = 5.46, because the small update transient is still there. In Frässle’s real data the middle frontal effect was not detectable at all. That fits a real update transient that is smaller than my default, or real noise that is larger than mine. My noise level is a free choice, so the model’s t-values are not comparable to any real study. The ratio between active and passive is the meaningful number.
Second, decoding scored against the read-out can never do better than the read-out. With pupil-grade labels, decoding against the labels falls to 0.60, while decoding against the truth stays at 0.94. A real experiment with a weak read-out would conclude that the region carries little information, when in fact the labels are the weak link.
Negative control · shuffled switch times
The shuffled condition keeps everything from the passive run but places the switch events at random times. Its beta is 0.012, with t = 0.97, which is what a correct analysis should give when there is nothing to find. This sets the floor. The passive beta of 0.106 is about nine times that floor, so in the model it is a genuine residual signal and not a fitting artefact.
The counterexample
The large-update condition is the counterexample to the claim that no-report designs always strip frontal switch activity. With an update transient of 0.8, the passive beta is 0.67, nearly half the active-report value of the default model and more than six times the default passive value. A frontal region that genuinely marks every perceptual change, independent of report, would survive the subtraction. Such a region would not be shown by this survival to cause the switch. It would be shown to respond to it, which is Brascamp’s point turned around.
The failure boundaries
The first sweep, with seed 380, varies the update transient from 0 to 1.2. The share of the switch-locked beta that disappears without report falls steadily. It is 98.6 per cent at zero update, 91.0 per cent at 0.15, 77.1 per cent at 0.5 and 59.1 per cent at 1.2. So “how much disappears” is not a fact about report alone. It is a measure of the ratio between report-dependent and report-independent transients.
sweep (seed 380): update transient vs share that disappears
update 0.00 active 1.5569 passive 0.0216 disappears 98.6%
update 0.15 active 1.6872 passive 0.1518 disappears 91.0%
update 0.30 active 1.8176 passive 0.2819 disappears 84.5%
update 0.50 active 1.9914 passive 0.4554 disappears 77.1%
update 0.80 active 2.2520 passive 0.7157 disappears 68.2%
update 1.20 active 2.5996 passive 1.0628 disappears 59.1%What this is for: seeing how “how much disappears” depends on the report-independent transient. What it is not: an estimate of the real transient, which the model cannot supply.
With JavaScript off: update 0, active 1.557, passive 0.022, 98.6% disappears; update 0.15, active 1.687, passive 0.152, 91.0% disappears; update 0.3, active 1.818, passive 0.282, 84.5% disappears; update 0.5, active 1.991, passive 0.455, 77.1% disappears; update 0.8, active 2.252, passive 0.716, 68.2% disappears; update 1.2, active 2.600, passive 1.063, 59.1% disappears.
The second sweep, with seed 381, varies the timing error of the read-out from zero to 1.3 seconds. Agreement with the true percept falls from 0.94 to 0.59. The active-report beta falls by 37 per cent, from 1.38 to 0.87, because blurred switch times dilute the regressor. Decoding against the read-out tracks the read-out, falling from 0.93 to 0.59. Decoding against the truth barely moves, from 0.96 to 0.93.
sweep (seed 381): read-out jitter vs agreement, decoding and active beta
jitter 0.00 agree 0.936 dec vs OKN 0.930 dec vs truth 0.955 active beta 1.3838
jitter 0.20 agree 0.894 dec vs OKN 0.884 dec vs truth 0.953 active beta 1.3611
jitter 0.35 agree 0.830 dec vs OKN 0.822 dec vs truth 0.953 active beta 1.3314
jitter 0.60 agree 0.737 dec vs OKN 0.732 dec vs truth 0.953 active beta 1.2284
jitter 0.90 agree 0.661 dec vs OKN 0.656 dec vs truth 0.949 active beta 1.0652
jitter 1.30 agree 0.593 dec vs OKN 0.590 dec vs truth 0.930 active beta 0.8719What this is for: finding where a noisy read-out starts to hide real effects. What it is not: a measurement of any real eye tracker.
With JavaScript off: jitter 0 s, agreement 0.936, decoding vs labels 0.930, vs truth 0.955, active beta 1.384; jitter 0.2 s, agreement 0.894, decoding vs labels 0.884, vs truth 0.953, active beta 1.361; jitter 0.35 s, agreement 0.830, decoding vs labels 0.822, vs truth 0.953, active beta 1.331; jitter 0.6 s, agreement 0.737, decoding vs labels 0.732, vs truth 0.953, active beta 1.228; jitter 0.9 s, agreement 0.661, decoding vs labels 0.656, vs truth 0.949, active beta 1.065; jitter 1.3 s, agreement 0.593, decoding vs labels 0.590, vs truth 0.930, active beta 0.872.
This is the method’s failure boundary. Below about 0.7 agreement, a no-report study loses a quarter or more of any real switch effect to label noise, and it cannot measure content coding at all, because its decoder can be no better than its labels. A pupil-only study at 59 per cent agreement sits on the wrong side of that line. An OKN study at 82 per cent sits on the right side, though not by a wide margin.
What this is for: comparing the five conditions on all four measures. What it is not: evidence about brains; the model was built to embody the argument.
- Active report: beta 1.43, decoding 0.95 against truth. The classic result, frontal burst included.
- Passive: beta falls to 0.11, a 92.6 per cent loss; decoding unchanged. Burst gone, message kept.
- Shuffled timings: beta 0.012, t 0.97. The negative control; nothing to find, nothing found.
- Large update: passive beta 0.67. The counterexample; a region that marks every change survives.
- Pupil-grade labels: agreement 0.60 and decoding against labels 0.60, though the truth is still 0.94. The failure boundary.
For Kavya, the instrument gives two design rules. First, her passive condition needs an OKN read-out at the 80 per cent level or better, checked against report in a calibration block, or her null results will be uninterpretable. Second, if she wants to test content coding, she must analyse patterns across voxels against the read-out labels, and report the read-out’s agreement next to every decoding score.
The instrument also has limits that no parameter sweep can reveal. It assumes that the report transient, the update transient and the content code are independent and add linearly in the voxel, whereas real cortex may interact in ways that make the subtraction imperfect. It assumes a single canonical haemodynamic response, although frontal haemodynamics vary between people and regions. And it assumes balanced populations, the most favourable case for the reconciliation. Each assumption is stated so that a reader can replace it and rerun the code, which is the main reason for building the model at all.
The case against the subtraction
I have argued that most switch-locked frontal BOLD activity in rivalry belonged to noticing and reporting, and that prefrontal content coding survives without report. Here is the strongest case I can make against that reading, objection by objection.
The absence is only a threshold. “Completely vanished” describes a cluster that did not reach significance. Twenty observers give limited power, and a smaller frontal effect might have been there. My own model shows that a residual a tenth the size of the report effect can exist and still be missed with more noise. This objection is fair. It means the human data rule out a large report-independent frontal switch transient, not every transient. I accept it, and I have phrased the claims to match.
The read-out is not the percept. In the human study, OKN agreed with report 82 per cent of the time. If the passive switch times were less accurate than the active ones, the frontal effect could have been diluted out of sight. The objection weakens because the occipital and parietal effects survived with the same read-out. If label noise had erased frontal effects, it should have weakened posterior effects too. It did not erase them.
No-report is not no-cognition. This is Block’s objection, and I take it to be the deepest. Passive observers may still notice and think about switches. If so, the residual frontal activity should be larger than zero, which argues against a perfect subtraction but in favour of my conclusion, since even with residual cognition the middle frontal effect was not detectable.
Replay is not a neutral baseline. A physical swap on the screen differs from a spontaneous switch in timing, sharpness and predictability, so subtracting replay may remove some genuine perceptual signal along with the competition. If replay switches are more abrupt, they could even produce larger transients than rivalry, hiding a frontal contribution. This objection is reasonable, but it applies equally to the report and no-report conditions, which used the same replay control. It therefore cannot explain why the middle frontal difference appeared with report and disappeared without it.
The monkeys are not people. In the content-coding work, two macaques, one lab and one prefrontal region carry the content-coding claim. Macaque and human frontal anatomy differ, and no human study has decoded rivalry content from prefrontal cortex without report at the same resolution. This objection does not touch the fMRI finding. It does limit how far the reconciliation can be pushed, and it is why the claim in my ledger is marked likely rather than established.
Content coding is not consciousness. Hesse and Tsao found in a 2020 eLife study that face cells in inferotemporal cortex multiplex “the conscious percept and veridical physical stimulus” without report. Prefrontal neurons might carry the percept for the same reason posterior ones do: because they receive it. Decoding shows information, and information is not experience; I agree, and the paper claims only information.
Analogy · the court stenographer
In a courtroom, the stenographer’s keys move at every spoken word. A visitor who only watched the stenographer might think the keys were producing the testimony. Send the stenographer home and the testimony continues, though no record of it is made. Button presses in rivalry are the stenographer, and the frontal transient is the clatter of the keys. Where it breaks: a stenographer cannot change what a witness says. Reporting may change rivalry itself, as Part four showed, so the analogy understates how much the record can alter what is recorded.
A realistic worked case
Kavya’s pilot gives a realistic version of the problem. The numbers that follow are invented, for illustration, at a scale that matches small published studies. With twelve volunteers, her middle frontal switch beta is 0.38 with report and 0.05 without. The naive share that disappears is 1 − 0.05 / 0.38, about 87 per cent. Her OKN read-out agreed with report 79 per cent of the time in a calibration block.
Now add the uncertainty. If the passive beta has a standard error of 0.04, its 95 per cent interval runs from about −0.03 to 0.13. The share that disappears could then be anything from about 66 per cent to slightly over 100 per cent. Her pilot cannot tell a modest update transient from none. To separate a residual of a tenth of the report effect from zero, she needs a passive standard error near 0.015, roughly seven times her pilot’s twelve volunteers if the noise scales in the usual way. That is a sample size argument, and it belongs in her preregistration.
What would change my mind
I would abandon the claim that the classic frontal switch signal was mostly report if a well-powered no-report study found a middle frontal switch transient at least half the size of the report-condition effect. It would need an OKN read-out at 80 per cent agreement or better and the same participants in both conditions. Two such studies from different labs would settle it.
I would abandon the reconciliation, that content coding survives where transients vanish, if a human study with intracranial recordings found no percept information in lateral prefrontal cortex during no-report rivalry. It would have to have good coverage and an adequate read-out. A single null would not be enough, because coverage in human intracranial work is set by clinical need. A consistent pattern of nulls would be.
I would revise the RTM placement if silent noting, as in thought experiment 4, failed to bring back any frontal transient with identical OKN. That would suggest the transient belongs to the motor step, which is L4 and beyond, rather than to reflexive noticing at L3.
I would also reconsider the decision rule in the conclusion if objective read-outs were shown to change the rivalry they measure. One example would be if the drifting gratings needed for OKN produced systematically different dominance durations or frontal responses from static gratings. In that case the no-report condition would differ from the report condition in more than one respect, and the subtraction would need a separate correction.
What I don’t know
I do not know how large the report-independent update transient is in human frontal cortex, and the evidence permits anything from zero to a modest value. I do not know whether the inferior frontal conflict signal found by Weilnhammer and colleagues survives without report. Their design used reports. I have not been able to check how many human no-report rivalry fMRI studies exist beyond the ones cited here. There may be replications or failures I have missed.
I do not know whether the macaque prefrontal content code is used for anything during passive viewing. It might be a standby copy, ready for action if needed, or it might shape what is seen. Only causal methods could tell, and those have not been applied under no-report conditions. Finally, I do not know whether rivalry generalises. It is one kind of perceptual change, with slow alternations and strong competition. Claims about consciousness drawn from it may not carry over to brief, masked or unattended stimuli.
The decisive study, specified
The study that would answer the question of this paper, and test the RTM claim, has five features. It records the same human participants in active report, passive viewing and silent-noting conditions. It reads switches from OKN calibrated against report in every session. It measures both the switch-locked transient and pattern-based decoding of the percept, in middle and inferior frontal regions and in visual cortex. It includes physical replay as a control. And it uses a recording method with enough resolution to decode content: intracranial electrodes in patients already implanted for clinical reasons, or high-field fMRI with pattern analysis.
Claim Ledger
Each claim below carries a confidence level, the kind of evidence behind it and what would refute it. “Established” means multiple consistent sources or direct arithmetic. “Likely” means good evidence from a small number of studies. “Speculative” means my own inference or modelling.
| ID | Claim | Confidence | Evidence type | What would refute it |
|---|---|---|---|---|
| C1 | Without report, switch-related differential activity in the middle frontal gyrus fell below detection while occipital and parietal effects remained. | Likely | One direct study | A well-powered replication finding a middle frontal effect without report. |
| C2 | Frontoparietal switch-related BOLD scales with how noticeable switches are. | Likely | One direct study, converging work | Unnoticeable switches producing full-size frontoparietal responses. |
| C3 | OKN agrees with rivalry reports about four-fifths of the time and leads them by about half a second. | Established | Measured, several studies | Large samples finding agreement near chance. |
| C4 | Macaque lateral prefrontal populations encode the seen percept during rivalry without report. | Likely | One lab, two animals | Failure to decode in further animals with matched methods. |
| C5 | Switch-locked BOLD contrasts cannot detect a balanced content code. | Established | Arithmetic, modelling | None in principle; only an error in the arithmetic. |
| C6 | The classic frontal switch transient mainly reflected noticing and reporting, not the cause of switches. | Likely | Inference from C1, C2 | A causal study showing that suppressing the transient stops or delays switches without report. |
| C7 | Inferior frontal cortex keeps a report-independent role in resolving perceptual conflict. | Speculative | Extrapolation from TMS with reports | A no-report study finding no inferior frontal switch signal and no TMS effect on OKN-based rates. |
| C8 | Read-outs below about 0.7 agreement cannot support null conclusions about switch effects or content. | Speculative | Toy model | Empirical work showing small loss of effect at low agreement. |
| C9 | In RTM terms, reflexive noticing (L3) in rivalry is mostly elicited by the task. | Speculative | Interpretation | Silent noting producing no frontal transient, or passive viewing producing a full one. |
| C10 | The Nyāya and Yoga distinction between cognising an object and cognising the cognising matches the split between percept and report. | Speculative | Textual comparison | A reading of the commentaries showing the distinction means something else. |
Dated prediction. By 31 December 2030, at least one human study with intracranial or high-field recordings will report above-chance decoding of the rivalry percept from lateral prefrontal cortex during no-report viewing, in the same participants in whom the switch-locked middle frontal transient is not detectable without report. If such a study finds no prefrontal percept information despite good coverage and an OKN read-out of 80 per cent or better, the prediction fails.
Glossary
- Agreement
- The share of time an objective read-out such as OKN matches the percept as reported, or in the instrument, the true percept.
- Binocular rivalry
- Alternation in perception between two different images shown one to each eye, with no change in the images.
- BOLD
- Blood-oxygen-level-dependent signal: the measure fMRI records, which follows local blood flow and lags neural activity by several seconds.
- Cit (चित्)
- Sanskrit, awareness as such. Layer L5 of the RTM hypothesis.
- Content code
- A pattern of activity across neurons that carries which stimulus or percept is present, which can change without changing the average activity.
- Decoding
- Training a classifier to read a stimulus or percept from recorded activity and testing it on held-out data.
- Dominance period
- The time for which one image is seen before perception switches to the other.
- Liṅga (लिङ्ग)
- Sanskrit, a mark or sign from which something is inferred. In Nyāya Sūtra 1.1.16, the mark of the mind.
- Manas (मनस्)
- Sanskrit, the mind as an inner instrument. In Nyāya, the factor that joins each sense in turn, so that cognitions arise one at a time.
- Middle frontal gyrus
- A ridge of the lateral frontal lobe, part of the dorsolateral prefrontal cortex. The site of the largest report-dependent rivalry cluster in Frässle and colleagues’ study.
- No-report paradigm
- An experiment that infers what is perceived from an objective measure, so that no report is required in the test condition.
- OKN
- Optokinetic nystagmus: the reflexive slow-follow and fast-return movements the eyes make when viewing a drifting pattern. Its slow phase follows the perceived motion.
- Prakāśa (प्रकाश)
- Sanskrit, “light” or manifestation. Layer L2 of the RTM hypothesis, read here as the content that is seen.
- Pratyabhijñā (प्रत्यभिज्ञा)
- Sanskrit, “recognition”. Layer L4 of the RTM hypothesis, read here as categorising and reporting the percept.
- Puruṣa (पुरुष)
- In Sāṅkhya and Yoga, the witnessing consciousness, distinct from the mind and its fluctuations.
- Replay
- A control condition in which the images are physically swapped to mimic rivalry switches, so the visual change happens without internal competition.
- Spanda (स्पन्द)
- Sanskrit, “vibration” or pulsation. Layer L1 of the RTM hypothesis, read here as the alternation itself.
- Switch-locked contrast
- An fMRI analysis asking whether a region’s average signal changes around the moments of perceptual switches.
- Update transient
- In this paper, a brief burst of activity at every perceptual change that does not depend on report.
- Vimarśa (विमर्श)
- Sanskrit, reflexive awareness. Layer L3 of the RTM hypothesis, read here as noticing that a switch happened.
- Yugapad (युगपद्)
- Sanskrit, “at the same time”. Used in Nyāya Sūtra 1.1.16 for the simultaneous cognitions that do not arise.
A claim worth preregistering
Kavya meets Professor Siddiqui a week before the deadline. She shows him Frässle’s middle frontal result, Brascamp’s unnoticeable switches and the macaque decoding. He is unhappy for about ten minutes, then interested, and they agree on a preregistration with two claims. The first is that the middle frontal switch transient will not be detectable without report. The second is that a pattern decoder trained on lateral prefrontal voxels will predict the OKN-labelled percept above chance in the same passive runs. She adds a silent-noting condition and a calibration block for OKN, and she raises her sample to the size her pilot’s standard errors require. The orange patch stays on his slide, with a new caption.
The question was how much prefrontal BOLD activation disappears when switches are tracked without button presses. In the main human study, the switch-related differential activity in the middle frontal gyrus disappeared completely at the study’s threshold, along with report-dependent effects in motor and insular areas. Occipital and parietal effects survived. A second design, which made switches unnoticeable, found frontoparietal responses minimised in step with how noticeable the switches were. So the classic frontal switch signal was mostly the noticing and reporting of switches.
That answer has a firm edge. It is a statement about the average signal at the moment of a switch, and it says nothing about the content a region carries. Monkey recordings show that lateral prefrontal populations still encode the seen percept on single trials without any report. The instrument shows why both results can hold at once: a report-dependent burst and a balanced content code are different signals, and a switch-locked contrast sees only the first.
The instrument adds two practical limits. How much disappears depends on the ratio between report-dependent and report-independent transients, so a region that really marks every change would keep part of its signal. And below about 0.7 agreement between read-out and percept, a no-report study loses much of any real effect and cannot measure content at all. OKN clears that bar, while the pupil, on Frässle’s figures, does not.
For the RTM hypothesis, the result is a refinement of L3. No-report designs act as a subtraction of prompted reflexive noticing, leaving the alternation at L1 and the manifest content at L2. The evidence suggests that, in rivalry, vimarśa is mostly called up by the task. The Nyāya and Yoga texts draw the same line between cognising an object and cognising the cognising, and both would expect a no-report design to remove the second cognition and keep the first.
The wider lesson concerns the relation between an experimental subtraction and the thing it is meant to isolate. Removing a behaviour removes its neural preparation, but it also removes whatever cognition that behaviour recruited, including the prompted noticing that the task demanded. No single design isolates experience. A set of designs that remove different things in turn, and that measure averages and patterns separately, can come close enough to make the theories disagree in public.
For anyone reading a rivalry paper, the decision rule is simple. Ask whether the frontal effect was measured with and without report, with an objective read-out of known agreement. If it was measured only with report, read it as a measure of noticing until shown otherwise. If it vanished without report, read that as evidence about the transient, and look separately for evidence about content. Kavya’s preregistration tests both, which is why it can fail informatively either way.
Sources
Primary research
- Brascamp, J., Blake, R., & Knapen, T. (2015). Negligible fronto-parietal BOLD activity accompanying unreportable switches in bistable perception. Nature Neuroscience, 18(11), 1672–1678. doi:10.1038/nn.4130
- Dwarakanath, A., Kapoor, V., Werner, J., Safavi, S., Fedorov, L. A., Logothetis, N. K., & Panagiotaropoulos, T. I. (2023). Bistability of prefrontal states gates access to consciousness. Neuron, 111(10), 1666–1683. doi:10.1016/j.neuron.2023.02.027
- Einhäuser, W., Stout, J., Koch, C., & Carter, O. (2008). Pupil dilation reflects perceptual selection and predicts subsequent stability in perceptual rivalry. PNAS, 105(5), 1704–1709. doi:10.1073/pnas.0707727105
- Fox, R., Todd, S., & Bettinger, L. A. (1975). Optokinetic nystagmus as an objective indicator of binocular rivalry. Vision Research, 15(7), 849–853. doi:10.1016/0042-6989(75)90265-5
- Frässle, S., Sommer, J., Jansen, A., Naber, M., & Einhäuser, W. (2014). Binocular rivalry: frontal activity relates to introspection and action but not to perception. Journal of Neuroscience, 34(5), 1738–1747. doi:10.1523/jneurosci.4403-13.2014
- Hesse, J. K., & Tsao, D. Y. (2020). A new no-report paradigm reveals that face cells encode both consciously perceived and suppressed stimuli. eLife, 9, e58360. doi:10.7554/eLife.58360
- Kapoor, V., Dwarakanath, A., Safavi, S., Werner, J., Besserve, M., Panagiotaropoulos, T. I., & Logothetis, N. K. (2022). Decoding internally generated transitions of conscious contents in the prefrontal cortex without subjective reports. Nature Communications, 13, 1535. doi:10.1038/s41467-022-28897-2
- Knapen, T., Brascamp, J., Pearson, J., van Ee, R., & Blake, R. (2011). The role of frontal and parietal brain areas in bistable perception. Journal of Neuroscience, 31(28), 10293–10301. doi:10.1523/jneurosci.1727-11.2011
- Leopold, D. A., & Logothetis, N. K. (1996). Activity changes in early visual cortex reflect monkeys’ percepts during binocular rivalry. Nature, 379, 549–553. doi:10.1038/379549a0
- Lumer, E. D., Friston, K. J., & Rees, G. (1998). Neural correlates of perceptual rivalry in the human brain. Science, 280(5371), 1930–1934. doi:10.1126/science.280.5371.1930
- Naber, M., Frässle, S., & Einhäuser, W. (2011). Perceptual rivalry: reflexes reveal the gradual nature of visual awareness. PLoS ONE, 6(6), e20910. doi:10.1371/journal.pone.0020910
- Tong, F., Nakayama, K., Vaughan, J. T., & Kanwisher, N. (1998). Binocular rivalry and visual awareness in human extrastriate cortex. Neuron, 21(4), 753–759. doi:10.1016/s0896-6273(00)80592-9
- Weilnhammer, V., Fritsch, M., Chikermane, M., Eckert, A.-L., Kanthak, K., Stuke, H., Kaminski, J., & Sterzer, P. (2021). An active role of inferior frontal cortex in conscious experience. Current Biology, 31(13), 2868–2880. doi:10.1016/j.cub.2021.04.043
- Wheatstone, C. (1838). Contributions to the physiology of vision. Part the first. On some remarkable, and hitherto unobserved, phenomena of binocular vision. Philosophical Transactions of the Royal Society of London, 128, 371–394. doi:10.1098/rstl.1838.0019
- Zou, J., He, S., & Zhang, P. (2016). Binocular rivalry from invisible patterns. PNAS, 113(30), 8408–8413. doi:10.1073/pnas.1604816113
Primary texts and translations
- Gautama, Nyāya Sūtra, with Vātsyāyana’s Bhāṣya. Sanskrit e-text, GRETIL, Göttingen. gretil.sub.uni-goettingen.de
- Jha, G. (trans.). Nyāya Sūtras of Gautama, with the Bhāṣya of Vātsyāyana and the Vārtika of Uddyotakara. Hosted by Wisdom Library. wisdomlib.org (wording of 1.1.16 to verify)
- Patañjali, Yogasūtra, with the Bhāṣya. Sanskrit e-text, GRETIL, Göttingen. gretil.sub.uni-goettingen.de
- Āraṇya, H. (trans.). Yoga Sūtra 4.20, as given on Wisdom Library’s study page, with notes on the Bhāsvatī. wisdomlib.org
Official and reference
- Max Planck Society (2023). How the brain decides what we perceive. Press release, 13 April 2023. mpg.de
Secondary: reviews and theory
- Blake, R., & Logothetis, N. K. (2002). Visual competition. Nature Reviews Neuroscience, 3, 13–21. doi:10.1038/nrn701
- Block, N. (2019). What is wrong with the no-report paradigm and how to fix it. Trends in Cognitive Sciences, 23(12), 1003–1013. doi:10.1016/j.tics.2019.10.001
- Dehaene, S., Changeux, J.-P., Naccache, L., Sackur, J., & Sergent, C. (2006). Conscious, preconscious, and subliminal processing: a testable taxonomy. Trends in Cognitive Sciences, 10(5), 204–211. doi:10.1016/j.tics.2006.03.007
- Lamme, V. A. F. (2006). Towards a true neural stance on consciousness. Trends in Cognitive Sciences, 10(11), 494–501. doi:10.1016/j.tics.2006.09.001
- Lau, H., & Rosenthal, D. (2011). Empirical support for higher-order theories of conscious awareness. Trends in Cognitive Sciences, 15(8), 365–373. doi:10.1016/j.tics.2011.05.009
- Michel, M., & Morales, J. Minority reports: consciousness and the prefrontal cortex. Mind & Language. doi:10.1111/mila.12264
- Overgaard, M., & Fazekas, P. (2016). Can no-report paradigms extract true correlates of consciousness? Trends in Cognitive Sciences, 20(4), 241–242. doi:10.1016/j.tics.2016.01.004
- Tsuchiya, N., Wilke, M., Frässle, S., & Lamme, V. A. F. (2015). No-report paradigms: extracting the true neural correlates of consciousness. Trends in Cognitive Sciences, 19(12), 757–770. doi:10.1016/j.tics.2015.10.002
Cite this work
Mahapatra, R. R. Binocular rivalry without button presses: how much prefrontal activity was only the report? Mind Frontier paper 38 (U0885). rajeshrmahapatra.com. DOI: to be assigned.
@article{mahapatra_mf_u0885,
author = {Mahapatra, Rajesh Ranjan},
title = {Binocular rivalry without button presses: how much
prefrontal activity was only the report?},
journal = {Mind Frontier},
number = {38},
note = {Topic U0885},
url = {https://rajeshrmahapatra.com/papers/mind-frontier/U0885-no-report-binocular-rivalry-prefrontal-bold.html},
doi = {10.xxxx/placeholder}
}