Graziano’s attention schema theory and the temporoparietal junction: does a model of attention make awareness?
Graziano’s theory says awareness is the brain’s rough model of its own attention, and the cleanest test, breaking the model while leaving attention intact, has been run only in halves.
TLDR
- Claim: no published study has disrupted the temporoparietal junction and measured both awareness reports and exogenous capture in the same people. So whether reports suffer more than capture is not yet known.
- Mechanism: attention schema theory says the junction holds a simplified model of attention, used to control it. Breaking the model should spare fast capture, impair reports, and weaken control-dependent effects such as inhibition of return.
- Implication: the evidence comes in halves. One TMS study found fewer awareness reports, another found disrupted reflexive orienting, and masking experiments reveal capture without awareness. Together they fit the theory, but none of them tests it directly.
- Boundary: the theory explains why we claim to be aware, not whether anything is experienced. A clean dissociation would show that the junction builds the claims, while leaving the question of experience exactly where it was.
- Decision: trust a junction study only if it measures report, short-interval capture and inhibition of return together, at the junction and at a control site. Anything less cannot tell the model from the spotlight.
The blip she caught and did not see
Ananya is a 38-year-old air-traffic controller who works the approach sector for Bengaluru. Seven weeks ago she had a small stroke in the right hemisphere of her brain, at the place where the temporal lobe meets the parietal lobe. Her neurologist, Dr Farhan Qureshi, has to decide whether she can return to the radar room. Ananya and Dr Qureshi are invented for this paper. The tests he runs, and the brain systems they probe, are real.
Ananya feels well. Her speech is normal, her arms are strong and her eyesight is unchanged. Yet in a simulator session last week something odd happened. A conflict alert flashed in the lower left of the screen. Her eyes jumped to it within a fifth of a second, and her hand moved towards the right console key. Then she paused and said, “Sorry, did something happen there?” The simulator log showed that her gaze had landed on the alert. Her own account said there had been nothing to see.
That small scene holds the question of this paper. Something in Ananya’s brain detected the alert and pulled her attention to it. Something else, which should have told her that she was attending to the alert, seems to have failed. Can damage to one region of the brain remove the sense of being aware of something while leaving the attention itself intact? Michael Graziano’s attention schema theory says yes, and it names the region that Ananya has damaged as a likely place for that sense to be built.
The theory’s central claim is easy to state. The brain controls attention, and like any good controller it keeps a simplified model of the thing it controls. That model is the attention schema. When you say “I am aware of the red cup”, you are reporting what the schema says: that your attention has a grip on the cup. The theory says the schema is computed in a network centred on the temporoparietal junction, often shortened to TPJ.
So the precise question is this. Does disrupting the TPJ impair subjective awareness reports more than it impairs objective, stimulus-driven attention? My answer, which the rest of the paper earns, is that nobody has yet run the experiment that would show it. The evidence exists in halves: one study disrupted the TPJ and measured reports, while another disrupted it and measured reflexive orienting. Both found effects, and neither measured what the other measured. Ananya’s pause is suggestive, but it is not yet evidence.
I will come back to Ananya in every part. Her stakes are concrete. A controller who catches every blip but cannot tell you she caught it is a different safety risk from one who misses blips and knows she missed them. Dr Qureshi needs to know which kind of patient she is, and the theory says he can find out with the right three tests.
A controller with a model of itself
Attention, in the sense used here, is the brain’s way of giving some signals more processing than others. Psychologists separate two ways it can move. Exogenous attention is pulled by the stimulus: a flash, a sudden sound, a moving object. It is fast, it happens whether or not you want it to, and it fades quickly. Endogenous attention is pushed by goals: you decide to watch the lower left of the screen because that is where trouble usually starts. It is slower to deploy and can be held for as long as you choose. Both can be covert, which means attention moves without the eyes moving.
Michael Posner’s cueing task, described in his 1980 paper “Orienting of attention”, became the standard way to measure covert attention. A cue appears at one place, and then a target appears either at the same place or elsewhere. If the target appears at the cued place, people respond faster. That speed-up is the signature of attention having moved there. With an uninformative flash as the cue, the benefit lasts only a few hundred milliseconds. After that it reverses, and responses at the flashed place become slower than elsewhere. That reversal is called inhibition of return. It is usually read as the system pushing attention away from a place it has already checked.
The exact timings vary across studies, so I will speak of a short interval and a long interval rather than fixed values. The short interval shows capture, and the long interval shows inhibition of return. Remember that pair, because the whole argument turns on it.
Graziano’s theory starts from an engineering principle. A controller works better if it has a model of the thing it controls. In 1970 Roger Conant and Ross Ashby gave this a formal statement in “Every good regulator of a system must be a model of that system”. The brain controls the arm, and it keeps a body schema: a simplified, constantly updated model of where the limbs are and how they move. Daniel Wolpert and colleagues showed in a 1995 Science paper that people estimate their hand position in a way that fits an internal forward model of the arm.
Graziano’s step is to say that attention is also something the brain controls, so it should also have a model. He and Sabine Kastner set out the idea in a 2011 paper in Cognitive Neuroscience, proposing that “awareness is a perceptual reconstruction of attentional state”. The same paper adds a second claim. The machinery that models other people’s attention is the machinery that models our own. That is why the theory locates the schema in regions known for social cognition, with the TPJ first among them.
Analogy · the body schema
Close your eyes and touch your nose. You can do it because your brain keeps a model of where your hand is, even when you cannot see it. The model is simplified. It has no muscles, tendons or nerve fibres in it, just a hand-shaped thing in a place. Graziano’s claim is that awareness is the same kind of model, built for attention. It has no neurons or signal competition in it, just a vague inner grasp on things. Where it breaks: a body schema can be checked against the body. You can open your eyes and see your hand. There is no independent way to see your attention and compare it with the schema, so the analogy assumes the very check the theory cannot supply.
The theory is precise about what the schema represents. It does not depict attention as neurons competing for signal strength, because the brain has no use for that detail in controlling attention. It depicts attention as a kind of invisible grasp: something in you that takes hold of an object, makes it vivid and prepares you to act on it. Graziano argues that this caricature is why people describe awareness as a non-physical essence. The description is accurate to the schema and inaccurate to the brain.
Here the theory makes a move that many readers miss. It is not a theory of how experience is produced from matter. It is a theory of why a brain would claim to have experience, and what that claim is about. In a 2022 PNAS perspective, Graziano presents it as an account that treats consciousness as the brain’s model of its own attention. Some readers call this view illusionist, in the sense developed by Keith Frankish. Graziano accepts part of that label and resists part of it.
Now apply this to Ananya. If her stroke damaged the schema and spared the attention system, the theory predicts a specific pattern. Her eyes and covert attention should still be pulled by the alert. But her report, “I saw it”, comes from the schema, so it should fail or become unreliable. A second prediction is less obvious. Because the schema helps control attention, anything that depends on control should also suffer. Inhibition of return should weaken, and attention held on a goal should wander more.
That gives three tests, and Dr Qureshi can run all three in one afternoon. The first is a short-interval cueing test, which measures capture. The second is a long-interval cueing test, which measures inhibition of return. The third is a near-threshold detection test with a confidence rating, which measures report. The theory predicts a specific profile: capture spared, inhibition of return weakened, reports reduced.
Worked case · minimal
Take the simplest possible version of the logic. A spotlight can point at one of two places, left or right. A schema stores one bit: “I am attending left” or “I am attending right”. A flash on the left pulls the spotlight left. With a working schema, the bit flips to “left”, the system can report “I noticed the left”, and a controller can push the spotlight back after the flash has proved uninformative. That push is inhibition of return. Now delete the schema bit: the flash still pulls the spotlight, so capture survives. But the report has nothing to read, and the controller cannot push back from a place it does not know it visited. Two of three outputs fail, from one deletion.
The minimal case shows why the theory’s predictions are not arbitrary. They follow from where the model sits in the loop. It also shows a weakness that later parts will press on. In the two-place system, the controller could have read the flash directly instead of reading the schema. It could have said “there was a flash on the left, so push away from the left”. That shortcut would preserve inhibition of return without any model of attention at all. Whether brains take such shortcuts is an empirical question, and the instrument in part six turns it into a test.
The evidence comes in halves
Correlation is cheap in this field. Many regions light up when people are aware of something, and the TPJ is one of them. The test that matters is causal: disrupt the region and see what changes. Two studies have done this with transcranial magnetic stimulation, or TMS, which uses a brief magnetic pulse over the scalp to interfere with the cortex beneath. They are the backbone of this part, and they measured different things.
Half one: reports go down
In 2014 Kelly, Webb, Meier, Arcaro and Graziano published a PNAS paper titled “Attributing awareness to oneself and to others”. It had two parts. First, 50 people in an fMRI scanner judged whether a cartoon face was aware of an object beside it. That judgement recruited the TPJ on both sides, with the strongest effect on the right. Second, the team used that location as a target for TMS in a separate group.
The TMS task was simple. A faint dot appeared on one side of a screen, and people were told to press a key when they saw it. A single TMS pulse arrived 0.2 seconds after the dot. Thirteen people received pulses at the TPJ site and fourteen at a control site about 2 cm away. At the TPJ, detection of dots on the opposite side fell, giving an interaction of F = 5.76, P = 0.03. At the control site there was no such effect, F = 0.11, P = 0.75.
This is the strongest causal support for the theory’s claim about reports. People reported seeing fewer dots on the side opposite the pulse. The authors are careful about the limits. A single pulse is a small perturbation, and the TPJ is not in the same place in every head. For this paper the decisive limitation is different: the study had no separate measure of attention. It cannot tell us whether the dot still captured attention while going unreported, which is exactly the pattern Ananya showed.
What a single pulse can and cannot do
It helps to be precise about the perturbation. A single TMS pulse delivered 200 milliseconds after a stimulus interferes with whatever the targeted cortex is computing at that moment, for a few tens of milliseconds. It does not remove the region. Its effect depends on coil position, scalp-to-cortex distance and the individual folding of the cortex, which is why Kelly and colleagues located their target from the group fMRI result and checked a nearby control site. A reduction in detection under these conditions is evidence that the TPJ contributes something time-locked to reporting a faint stimulus. It is not evidence about what the stimulus did to attention meanwhile.
That missing measurement is not a criticism of the study, which asked a different question. It does mean that the study’s result is equally predicted by two accounts. On the schema account, the pulse corrupted the model, so the report failed while attention still went to the dot. On a reorienting account, the pulse delayed the ventral network’s response to an unexpected event, so attention reached the dot too late for it to be reported. Only a concurrent attention measure could separate these accounts, and the design did not include one.
Half two: reflexive orienting goes wrong
Three years earlier, Ana Chica, Paolo Bartolomeo and Antoni Valero-Cabré had run the complementary experiment, reported in the Journal of Neuroscience in 2011. They used a peripheral cue that was either uninformative, to measure exogenous orienting, or predictive, to measure endogenous orienting. Then they used double-pulse TMS over the right intraparietal sulcus, the right TPJ, or a sham. Thirteen participants were analysed, at a short and a long cue-target interval.
The finding that matters here is about the long interval. The authors report that “the TMS interference of either IPS or TPJ prevented the expected contralateral inhibition at the long SOA when attention was reflexively oriented”. In plain terms, inhibition of return was disrupted at both sites. For goal-directed orienting the sites differed: “TMS on IPS but not TPJ induced abnormal facilitatory effects at long SOAs”. Their summary links the TPJ “with the orienting of exogenous but not endogenous spatial attention”.
Read through the theory, this is a mixed result. Disrupted inhibition of return at the TPJ is what the theory predicts, since inhibition of return depends on knowing where attention has been. But the study had no awareness measure, so it cannot say whether reports fell too. And it found the TPJ involved in exogenous orienting, which on a naive reading is the very thing the theory says should be spared. The resolution depends on which part of exogenous orienting you mean. The fast pull may survive while the later push-back fails.
Capture without awareness, in intact brains
If disrupting the TPJ cannot yet show the dissociation, perhaps hiding the stimulus can. Graziano’s group has done this repeatedly with masking, which makes a stimulus invisible by following it quickly with another. Webb, Kean and Graziano used metacontrast masking in a 2016 paper in the Journal of Cognitive Neuroscience. They found that “attention was drawn to the stimulus regardless of whether or not people were aware of it”. Without awareness, however, attention “was less stable over time”, inhibition of return “was reduced”, and attention “was more driven by the luminance contrast of the stimulus”.
That is the theory’s predicted profile, produced by removing awareness of the stimulus rather than by disrupting the TPJ. Wilterson and colleagues extended it in six experiments published in 2020. When people were not aware of the cue, “the exogenous attention effect remained and the endogenous effect was impaired”. A 2021 fMRI study by Wilterson and colleagues found that the right TPJ was active in the task where awareness of the cue let people control their attention, and the cue drew exogenous attention even without awareness.
Notice the shape of this evidence. The masking experiments demonstrate that capture and awareness can come apart, and that control suffers when awareness goes. The two TMS experiments indicate that the TPJ matters for reports and for inhibition of return. What no study shows is the combination the theory needs: TPJ disruption that removes reports while capture is measured and found intact.
What this is for: seeing which of the five 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:
- Kelly 2014: Measured reports after a TPJ pulse, with a control site. No measure of capture or inhibition of return, so it cannot show capture surviving.
- Chica 2011: Measured orienting at short and long intervals after TPJ and IPS stimulation, with sham. No awareness measure.
- Webb 2016 masking: Measured awareness, capture and inhibition of return together, but removed awareness by masking, not by touching the TPJ.
- Wilterson 2021: fMRI showed right TPJ activity when awareness enabled control. Correlational, no perturbation.
- Rees 2000: A patient with right inferior parietal damage and extinction. Reports measured, capture not separated. Generous to score as a TPJ case.
- Needed study: Report, capture and inhibition of return, with TPJ, IPS and vertex stimulation in the same people. Not yet run.
True story · attention in a blind field
The patient GY has damage to his left primary visual cortex (details of its cause are from the wider GY literature, to verify) and reports no ordinary vision in part of his right field. In a 1999 study, Kentridge, Heywood and Weiskrantz showed that attention still worked there. Cues told him where a target was likely to appear, and he responded faster and more accurately at the cued place, “without those stimuli entering consciousness”. The authors concluded that attentional selection “and its entry into conscious experience cannot be one and the same process”. GY is a case of attention without awareness, caused by damage far from the TPJ. What it shows and what it does not: it supports the theory’s claim that attention and awareness are separable. It does not show where the separation happens in an intact brain.
Return to Dr Qureshi. If he reads the literature, he finds support for each of his three tests separately. He finds no study that ran all three after TPJ disruption. That gap is the main finding of this part. It is also an opportunity, because Ananya is, in effect, a single-patient version of the missing experiment.
What the junction does besides
A theory that places awareness in one region has to deal with everything else that region does. The TPJ is a crowded neighbourhood. Kajsa Igelström and Graziano reviewed it in a 2017 Neuropsychologia paper, describing the inferior parietal lobule and the overlapping TPJ as a hub with “high inter-individual variability”. They list its roles in bottom-up attention, self-perception, undirected thinking, episodic memory and social cognition. Any TPJ disruption touches several of these at once.
Three of those roles matter here, and the first is reorienting. Maurizio Corbetta and Gordon Shulman wrote a 2002 review in Nature Reviews Neuroscience that separated two systems. A dorsal attention network sets goals. A ventral network, centred on the right TPJ and inferior frontal cortex, detects unexpected but relevant events. In a 2008 Neuron review with Gaurav Patel, they described the ventral network as a system that “interrupts and resets ongoing activity”. When attention is focused, “the ventral network is suppressed to prevent reorienting to distracting events”.
The second is change detection. Jonathan Downar and colleagues found in a 2000 fMRI study that sudden changes in visual, auditory and tactile stimuli all drove “a right-lateralized network including the temporoparietal junction”. They called it a network “for involuntary attention to events in the sensory environment”. The third role is social. Rebecca Saxe and Nancy Kanwisher showed in their 2003 NeuroImage paper that a TPJ region responds when people reason “about the contents of another person’s mind”, and not to physical descriptions of people.
Graziano reads this overlap as support. If the TPJ models attention, it should be involved in reorienting attention, in noticing that attention has been grabbed, and in modelling what others are attending to. The overlap is what a schema would produce, he argues, while a sceptic reads it the other way. A region involved in so many things gives any theory something to point at. The overlap fits the schema, and it fits equally well a general-purpose hub that the schema story borrows.
Analogy · the flight progress strips
Before radar screens took over, controllers tracked each aircraft on a paper strip in a rack. A strip held the call sign, level and route, and its position in the rack stood for the aircraft’s position in the sequence. Controllers moved strips to plan, and read strips to know what they had already cleared. The strips were a model of the traffic, used to control the traffic. They were also how a relieving controller learned what the outgoing one had been attending to. Where it breaks: strips were written deliberately and could be checked against radar. An attention schema, if it exists, is written automatically and cannot be checked against anything. Also, nobody ever claimed the strips were aware of the aircraft.
Neglect: the hardest neighbour
Damage to the right hemisphere near the TPJ often causes spatial neglect, in which patients fail to notice the left side of space. Hans-Otto Karnath, Susanne Ferber and Marc Himmelbach argued in a 2001 Nature paper that “the superior temporal cortex is the neural substrate of spatial neglect in humans”. David Mort and colleagues, in a 2003 Brain paper on the anatomy of visual neglect, noted that the critical regions “have become intensely disputed”. Either way, the lesions sit close to the TPJ, and neglect impairs both attention and awareness.
That matters for the question. If TPJ damage reliably produces neglect, and neglect is a failure of attention as well as of report, then TPJ disruption impairs both. The theory needs a different profile, in which report fails more than capture. Neglect patients often fail to orient to the left at all, which is a capture failure. So neglect is at best ambiguous evidence for the schema, and on a plain reading it counts against the predicted dissociation.
True story · the burning house
In a one-page 1988 Nature report, John Marshall and Peter Halligan described a patient, P.S., with right-hemisphere damage and left-sided neglect. They showed her two line drawings of a house, one of which had flames coming from its left side. She “judged that the drawings were identical”. Yet when asked which house she would prefer to live in, “she reliably chose the house that was not burning”. Something in her brain had processed the fire, although her report said there was no fire to process. What it shows and what it does not: it is a dissociation between report and use, which is the shape the theory predicts. But P.S. also failed to orient to the left, and the study did not measure covert capture, so it cannot separate a broken schema from a broken spotlight.
Extinction makes the same point from another angle. In extinction, patients see a single stimulus on either side but miss the left one when both appear together. Geraint Rees and colleagues scanned such a patient, with “circumscribed right inferior parietal damage”, and reported in a 2000 Brain paper that extinguished stimuli still activated visual cortex. Unseen stimuli were processed, so here again report fails while early processing survives. Yet the failure is driven by competition with the right-side stimulus, which is an attentional effect, so it is not the clean case either.
Awareness at matched attention
The most direct evidence for a TPJ role in awareness itself comes from a study designed to hold attention constant. In a 2016 PNAS paper, Webb, Igelström, Schurger and Graziano used a task in which a stimulus “drew a similar amount of spatial attention” whether or not people were aware of it. They compared aware with unaware trials. “The largest cluster of activity was obtained in the temporoparietal junction (TPJ) bilaterally.” In contrast, “no significant activity in association with awareness was found in the banks of the intraparietal sulcus”.
This is a good design, and it is still a correlation. It shows that TPJ activity tracks awareness when attention is matched. It does not show that the TPJ produces the awareness reports, or that disrupting it would remove them while sparing attention. It is the third half of an experiment, if such a thing can exist: the right contrast, without the causal manipulation.
There is one more strand, which concerns other people’s attention. Arvid Guterstam and colleagues showed in a 2019 PNAS paper that people judge an object as being gently pushed when someone is looking at it. The finding suggests that we model other people’s vision implicitly, “using the simplifying construct of beams coming out of the eyes”. The participants denied believing in such beams. To Graziano this is the social schema at work: a caricature of attention, used without being believed. It is evidence that the brain models attention simply. It is not evidence about the TPJ.
Return to Ananya, whose lesion is in a hub that does at least five jobs. If she misses alerts and denies seeing them, that could be neglect, a broken reorienting system, or a broken schema. Dr Qureshi’s three tests are designed to separate these. Neglect predicts impaired capture, a broken reorienting system predicts slow capture of unexpected events. A broken schema predicts intact capture with impaired report and impaired inhibition of return. Each reading predicts a different profile, and a single measure of report cannot choose between them.
Rival readings of Ananya’s pause
Attention schema theory is one of several theories that would explain Ananya’s pause, and each one predicts a different profile on Dr Qureshi’s three tests. Comparing them is the quickest way to see what a TPJ study must measure. If the rivals all predict the same thing, a study cannot choose between them, however careful it is.
Christof Koch and Naotsugu Tsuchiya argued in a 2007 Trends in Cognitive Sciences paper that attention and consciousness are “two distinct brain processes”. On their view, attention can occur without awareness and some awareness can occur with little attention. Attention schema theory accepts the first half and reframes it. Attention without awareness is attention without a working model of it. The second half, awareness without attention, is harder for the theory, because the schema models attention. If there is no attention, there is nothing for the schema to model.
Global workspace
Stanislas Dehaene and colleagues proposed in a 2006 paper a taxonomy with three states. A stimulus can be subliminal, too weak to drive much processing. It can be preconscious, strong enough to be processed but not attended, so it does not reach the global workspace. Or it can be conscious, attended and broadcast. On this view, Ananya’s alert drove her eyes through a fast sensorimotor route without being broadcast. Damage near the TPJ might weaken the route into the workspace for left-side events.
Workspace theory predicts that report and use for reasoning fail together, while fast reflexes survive. It does not obviously predict that inhibition of return will weaken. That gives the two theories one point of difference: inhibition of return. If TPJ disruption spares capture, removes reports and leaves inhibition of return intact, workspace theory fits better. If inhibition of return also fails, the schema story fits better, since only it ties report to a control function.
Higher-order theories
Hakwan Lau and David Rosenthal reviewed empirical support for higher-order theories in 2011. On these theories a state is conscious when there is a suitable representation of it, usually located in prefrontal cortex. Attention schema theory is a higher-order theory of a particular kind. Its higher-order representation is of attention rather than of a perceptual state, and it sits in the TPJ rather than the prefrontal cortex. A higher-order theorist would predict that Ananya’s report fails while first-order processing survives. That matches the schema story on report, and it says nothing about inhibition of return.
Local recurrence
Victor Lamme’s recurrent processing theory, the subject of paper 35 in this series, argues in “Towards a true neural stance on consciousness” that local feedback in sensory cortex is enough for phenomenal experience. Report and attention are added later. On that view, Ananya may have experienced the alert and failed to report it. The TPJ is then a reporting hub, and its damage removes access, not experience. Lamme’s theory predicts the same behavioural profile as the schema theory, with an opposite interpretation of what is missing.
That last point needs emphasis. Every theory in the figure predicts that Ananya’s reports will fail if the right pathways are disrupted. They disagree about whether anything was experienced and about whether control should suffer. Only the second disagreement is testable by behaviour. That is why inhibition of return and attentional stability, which look like technical details, are in fact the crux of the paper’s question.
Predictive processing and illusionism
Two further positions deserve a brief placement. Predictive processing, the subject of paper 36, treats attention as the optimisation of precision, the expected reliability of prediction errors. On that framework an attention schema would be a hyperprior over one’s own precision settings, a model of where the system is currently placing its confidence. That reformulation is attractive, because it makes the schema a specific variable inside an existing computational architecture rather than a free-standing module. It also inherits paper 36’s difficulty: precision-weighting operates in sleep and in inattention, so a model of precision is not obviously a model of awareness.
Keith Frankish’s illusionism, set out in a 2016 paper in the Journal of Consciousness Studies, holds that phenomenal consciousness is an introspective illusion: we represent our experiences as having qualitative properties they do not have. Attention schema theory supplies one candidate mechanism for that misrepresentation. The two positions are separable, however. One could accept that the TPJ computes a schema used for control while denying that the schema’s content is illusory. Behavioural experiments of the kind discussed here cannot decide between the illusionist and the realist reading of the schema, which is why I keep the empirical question and the metaphysical one apart.
Where the RTM layers sit
Placing the theory on the RTM layers clarifies what it does and does not claim. Attention, as the selective weighting of signals, is closest to L1: a system responding more to some inputs than to others. The schema, a model of that weighting used by the system itself, is a candidate for L3, a form of self-reference. The social extension, modelling other people’s attention with the same machinery, sits at L4, recognising a mind as a mind. The theory has nothing at L2 except the schema’s description of it, and it denies that L5 is anything beyond that description.
This placement produces this paper’s finding about RTM. Paper 36 suggested modelling L3 as same-order reflexivity, in which a state carries awareness of itself rather than being the target of a separate state. Attention schema theory offers the opposite design, since its L3 is a distinct model, computed elsewhere, which represents first-order attention. The GY evidence, attention working without reported awareness, is compatible with either design. A TPJ dissociation would favour the separate-model design. So the experiment this paper asks for would also test a design choice inside RTM.
Analogy · the sports commentator
A cricket commentator watches the same match as the players and describes it in a few short words: “he’s gone for the big shot”. The description is useful to listeners and leaves out nearly everything about the biomechanics. Silence the commentator and the match goes on, though the listeners lose their account of it. Where it breaks: the commentator does not control the match, and the schema is supposed to control attention. A better version is a captain calling field changes from a running commentary in his own head. Remove that and the match still goes on, but the field gets worse.
Return to Ananya once more. If the rivals are lined up against her three test results, only some profiles discriminate. Intact capture with failed reports is predicted by all of them. Intact capture with failed reports and failed inhibition of return is predicted mainly by the schema theory. Failed capture with failed reports is the neglect profile, which no theory needs a schema to explain. Dr Qureshi’s second test, the long-interval one, carries most of the weight.
What this is for: seeing which theories a two-measure study can and cannot separate. What it is not: data; region positions are predictions.
- Attention schema: reports fall, capture holds, inhibition of return weakens. Upper left, with IOR down.
- Global workspace: reports fall, capture holds, inhibition of return intact. Also upper left; only the IOR test separates it from the schema.
- Neglect: capture and reports both fall. Upper right. No schema is needed to explain it.
- Control site: nothing changes. Lower left. Every good study needs this point.
The charioteer and the crystal
Two old traditions have detailed accounts of how attention relates to awareness. I use them here as alternative models, to be tested against the same cases, and not as authorities. Each one sharpens a different part of the TPJ question. The Theravāda Abhidhamma treats attention as a factor present in every moment of mind. The Yoga Sūtra and its commentaries treat the sense of “I am aware of this” as a mistaken identification between two things that are close together.
Manasikāra, the charioteer
In the Abhidhamma, each moment of consciousness, a citta, comes with a set of mental factors. Seven of these are universal, present in every citta, and one of the seven is manasikāra, usually translated as attention. Nina van Gorkom’s study Cetasikas quotes the Atthasālinī, whose definition she notes matches Visuddhimagga XIV.152. Manasikāra “has the characteristic of driving associated states towards the object”, the function “of joining (yoking) associated states to the object”, and “the manifestation of facing the object”. It “should be regarded as the charioteer of associated states because it regulates the object”. I quote her rendering, and the citations should be checked against Ñāṇamoli’s printed translation of the Visuddhimagga.
The charioteer image fits attention schema theory surprisingly well. Manasikāra is described by what it does to the rest of the mind: it steers, it yokes, it faces. That is a control description. The same source adds that two kinds of citta are also called manasikāra. One is five-door adverting, the first moment of a sense-door process, “called ‘controller of the sense-door process’”. The other is mind-door adverting, which turns to the object through the mind-door and is “called ‘controller of the javanas’”, the impulsion moments that follow.
So the Abhidhamma has two levels of attention: a factor present everywhere, and a turning moment that controls what happens next. That is close to the distinction between attention and the control of attention. The tradition does not treat the turning moment as a model of attention. It is attention in the act of steering, and here the two accounts part company.
The tension is sharp on one point. In the Abhidhamma every citta is a moment of consciousness, so there is no attention outside consciousness. Kentridge’s blindsight patient GY, attending without reported awareness, would have to be described differently. Either his blind-field processing involves cittas he cannot report, or it is not mental in the Abhidhamma sense at all. Attention schema theory and the Abhidhamma therefore disagree about the very case that best supports the theory.
Primary text · Yoga Sūtra 1.4
वृत्तिसारूप्यमितरत्र · vṛttisārūpyam itaratra. In Rama Prasada’s 1924 translation on Wisdom Library: “Identification with modifications elsewhere”, that is, at times other than stillness the seer appears to take the form of the mind’s fluctuations. Vācaspati Miśra’s commentary explains that notions such as “I am calm” arise “by taking the will-to-know and the conscious principle to be one on account of proximity, as in the case of the white crystal and the Japa flower”. Use here: a model of how a description of the mind gets mistaken for the knower.
The crystal and the flower
Vācaspati’s image is a clear crystal placed beside a red japā flower, the hibiscus. The crystal looks red. Nothing about the crystal has changed, and nothing about the flower has moved into it. The redness is an appearance produced by nearness. In the Yoga reading, the mind’s states are like the flower and pure consciousness, puruṣa, is like the crystal. The thought “I am ruffled” attributes to consciousness a colour that belongs to the mind.
Set this beside Graziano. On his account, the schema describes attention as an inner grasp, and the brain then claims that it has a non-physical awareness. That claim is a misattribution: a model’s description is taken as a fact about an inner essence. Vācaspati also describes a misattribution, in the opposite direction. On the Yoga view the awareness is real and the colouring is borrowed. On Graziano’s view the colouring is all there is, and the awareness is the borrowed appearance.
The structural match is worth stating precisely. Both accounts say that the ordinary report “I am aware of this” combines two things and mislabels the combination. They disagree about which of the two things is real. That is exactly the disagreement RTM must settle at L5, and neither account offers an experiment that could settle it.
There is one testable consequence. If the schema is in the TPJ and open monitoring changes the schema’s state, then TPJ activity should change in characteristic ways during reports of objectless awareness. I am not aware of a study that has looked for this with TPJ-focused analysis. It is a modest test. It would not settle which tradition is right, but it would check whether the schema story extends to the contemplative case at all.
A further contrast concerns where each account puts the error. For Graziano, the error is built into an efficient engineering compromise: a detailed model of attention would be expensive and useless for control, so the brain settles for a caricature and then believes it. For Vācaspati, the error is a beginningless confusion, sustained by habit, which practice can loosen. For the Abhidhamma, there is no error in the attention factor itself; the difficulty lies in mistaking the stream of factors for a self that owns them. These are three different diagnoses of the same ordinary report, and each one implies a different remedy.
For Ananya the traditions offer one practical lesson. The Abhidhamma’s “controller of the javanas” is a moment that turns the mind towards an object and then hands control to the moments that act on it. If something like that turning is impaired, the eyes can still be pulled while the later moments that know and report the object never get going. That is a description in another vocabulary of the profile Dr Qureshi is testing for.
A toy spotlight with a schema
An instrument in this series is a small piece of code that makes an argument checkable. This one is a toy model of covert attention with an optional attention schema. It does not show that brains work this way. It shows which measurements could tell a broken schema from a broken spotlight, and how easily a system without any schema can mimic the profile. Its code is short enough to read, and the numbers below are its actual output.
What the model contains
Attention is a spotlight spread over eight locations, and at each 50-millisecond step three things drive it. A salience input pulls it towards any flash, which is the capture pathway. A goal input pushes it towards a chosen place. A controller adds corrections, but it never sees attention directly. It reads the schema, a copy of the attention state mixed with noise. A parameter called fidelity sets how good that copy is, from 1 for a faithful copy to 0 for an unrelated guess.
The controller does two jobs. After a flash, it tags whichever places the schema says attention visited and pushes attention away from them. That produces inhibition of return. During a goal-holding task with a flashing distractor, it suppresses wherever the schema says attention has strayed. That produces stability. The model also makes a report, which is the part that matters most for the question. A target counts as “seen” only when the target evidence is strong enough and the schema says attention was on it. A separate forced-choice judgement uses the evidence alone, as an objective measure.
I ran five conditions with random seed 37 and 400 trials per cell. The intact model has schema fidelity 0.95. The schema lesion drops fidelity to 0.25, the analogue of TPJ disruption on the theory’s own account. The capture lesion cuts the salience gain to a quarter, the analogue of damage to the stimulus-driven pathway. The control node changes nothing in the loop, the analogue of TMS at the vertex. Finally, the hint agent has no schema at all. Its controller tags the flashed place directly from the stimulus and reports from the evidence alone.
The core of the model (Python, about 90 lines in full)
def schema(rng, a, f):
# simplified, noisy copy of the attention state a
return f * a + (1 - f) * softmax(rng.normal(0, 2.5, N))
# inside each 50 ms step of a cueing trial
ahat = schema(rng, a, cfg['fid'])
tag += DT * 4 * np.clip(ahat - 1/N, 0, None) # where the schema says attention went
u += DT * (-u/0.3 + g_cap*40*sal - g_ior*tag) + noise
# at target time
rt = 420 - 140 * a[tgt] + noise # reaction time, ms
ev = 2.2 * a[tgt] + noise # target evidence
seen = ev > 0.35 and ahat[tgt] >= 0.6 * ahat.max()
# dissociation index
DI = (report_drop / report_intact) - (capture_drop / capture_intact)The output
condition capture IOR stab report acc DI
intact 97.5 8.2 0.75 0.64 0.77 +0.00
schema lesion (TPJ-like) 99.9 -1.3 0.68 0.28 0.77 +0.60
capture lesion 15.4 -1.7 0.87 0.49 0.68 -0.60
control node (vertex-like) 99.3 7.6 0.74 0.64 0.78 +0.02
hint agent, no schema 100.8 13.9 0.43 0.71 0.76 -0.07Capture is the short-interval benefit at the cued place, in milliseconds. IOR is the long-interval cost, so positive values mean inhibition of return. Stability is the share of time attention stays on the goal. Report is the share of correctly judged targets also reported as seen, and accuracy is the forced-choice score. The dissociation index, DI, is the proportional drop in report minus the proportional drop in capture. Positive values mean report suffers more than capture.
The positive case is the schema lesion, in which capture is unchanged at 99.9 ms against 97.5 ms. Inhibition of return disappears, falling from 8.2 ms to −1.3 ms. Reports fall from 0.64 to 0.28 while forced-choice accuracy stays at 0.77. Stability drops a little, from 0.75 to 0.68, and the dissociation index is +0.60. This is the theory’s predicted profile, and it is worth saying plainly why. I built a model in which reports read the schema, so breaking the schema had to break reports. The model confirms the logic of the theory, and says nothing about the brain.
The negative case is the capture lesion, in which capture collapses to 15.4 ms. Inhibition of return also vanishes, because there is nothing to inhibit if attention was never pulled. Reports fall too, to 0.49, because attention reaches the target less often. Accuracy falls to 0.68, and the index is −0.60, the mirror image. This is the neglect-like profile, and it shows why inhibition of return alone cannot diagnose a schema lesion. Both lesions abolish it. Only the capture measure tells them apart.
Negative control · the vertex node
The control-node condition runs the intact model with fresh random draws and no change to the loop, which is what TMS at a site outside the network should do. Its index is +0.02, and every measure sits within sampling noise of the intact model. That gives a scale for the other numbers: differences of a few milliseconds or a few hundredths are noise. The schema and capture lesions move the index by about thirty times as much.
This is the model’s counterexample, and it connects to the machine-learning literature. Lotta Piefke and colleagues trained an agent with spare resources it could use however it liked, reported in a 2024 CogSci paper. They found that “the more uncertain the agent was about the location of its attentional window, the more it benefited from these additional resources, which developed an attention schema”. When attention is easy to predict from the stimulus, a schema adds little. Wilterson and Graziano found in their 2021 deep Q-learning agent that disabling a learned schema left performance “greatly reduced”. The two results fit together, because a schema pays where hints fail.
The failure boundary
The second run sweeps schema fidelity from 1 down to 0, with seed 370 and 250 trials per cell. Capture stays near 100 ms throughout and accuracy stays near 0.76. Reports hold at about 0.65 down to a fidelity of 0.6, then fall to 0.56 at 0.4, 0.20 at 0.2 and 0.15 at 0. Inhibition of return holds at about 6 ms down to 0.6 and turns negative from 0.4 onwards. Stability drifts from 0.73 to 0.61.
fidelity sweep: fid capture IOR report acc
1.0 98.2 6.7 0.65 0.77 0.73
0.8 100.0 5.9 0.66 0.74 0.74
0.6 97.3 5.8 0.66 0.77 0.75
0.4 98.3 -3.1 0.56 0.77 0.68
0.2 98.8 -4.2 0.20 0.74 0.67
0.0 100.3 -5.0 0.15 0.76 0.61The boundary matters for real experiments. A perturbation that degrades the schema only modestly, from 1 to 0.6 in this toy, produces no measurable change at all. A single TMS pulse is a modest perturbation. So a null result from mild TPJ stimulation would say little, while a positive result like Kelly’s suggests a substantial disruption. Below 0.4, reports and inhibition of return fall together while capture holds. That joint fall is the theory’s signature.
What this is for: finding the fidelity at which reports and inhibition of return fail together. What it is not: a calibration of TMS strength; the mapping from pulses to fidelity is unknown.
With JavaScript off: fidelity 1.0, report 0.65, IOR 6.7 ms; fidelity 0.8, report 0.66, IOR 5.9 ms; fidelity 0.6, report 0.66, IOR 5.8 ms; fidelity 0.4, report 0.56, IOR -3.1 ms; fidelity 0.2, report 0.2, IOR -4.2 ms; fidelity 0.0, report 0.15, IOR -5.0 ms.
The last part of the instrument scores the published studies on five legs: report measured (R), capture measured (X), inhibition of return or control measured (I), TPJ perturbed (T) and a control site used (C).
study audit: R report, X capture, I IOR/control, T TPJ perturbed, C control site
Kelly et al. 2014, single-pulse TMS R..TC 3/5
Chica et al. 2011, double-pulse TMS .XITC 4/5
Webb, Kean and Graziano 2016, metacontrast RXI.. 3/5
Wilterson et al. 2021, fMRI RXI.. 3/5
Neglect and extinction lesions (Rees et al. 2000) R..T. 2/5
studies passing all five: 0No study passes all five. The Rees entry deserves a caveat: its patient had right inferior parietal damage, which overlaps the TPJ only loosely. Scoring it as a TPJ case is generous, and it still scores lowest.
Worked case · realistic
Suppose Dr Qureshi tests Ananya and three age-matched colleagues. These numbers are invented for illustration. The colleagues average a capture benefit of 40 ms, inhibition of return of 22 ms and a report rate of 0.70 on correctly judged near-threshold targets. Ananya shows capture of 38 ms, inhibition of return of 3 ms and a report rate of 0.41, with forced-choice accuracy equal to theirs. Her dissociation index is (0.70 − 0.41) / 0.70 − (40 − 38) / 40, which is 0.41 − 0.05 = 0.36. Her profile is capture spared, inhibition of return largely lost and reports down by two fifths. That is the schema profile. It is one patient, with no control site and an invented baseline, so it would pass three of the five legs and prove nothing alone. It would, though, justify the full study.
What this is for: comparing the five conditions on all five measures at once. What it is not: evidence about brains; the model was built to embody the theory.
- Intact: capture about 98 ms, inhibition of return 8 ms, report 0.64. The baseline.
- Schema lesion: capture unchanged, inhibition of return gone, reports down to 0.28, accuracy unchanged. The theory’s fingerprint.
- Capture lesion: capture down to 15 ms, inhibition of return gone, reports and accuracy both down. The neglect-like profile.
- Control node: every value within noise of intact. The negative control.
- Hint agent: capture, inhibition of return and reports all present with no schema at all. Only stability gives it away, at 0.43.
One honest summary of the instrument is that it is a measuring tape for an experiment nobody has run. It says the three tests are enough to separate a broken schema from a broken spotlight. It says inhibition of return is necessary to the argument and not sufficient for it. And it says a system without a schema can pass the standard tests, so the strongest single test is stability against unpredictable distraction.
The case against the schema
I have presented the theory as a clean, testable proposal with a missing experiment. Its critics would say that is too generous. Here is the strongest version of their case, in five steps, each one given in the form I find hardest to answer.
First, the TPJ evidence fits a simpler story. The right TPJ is part of a reorienting network that “interrupts and resets ongoing activity”, in Corbetta’s words. Disrupting a circuit breaker should impair reorienting, including the later push-back that produces inhibition of return. It should also impair reports of unexpected, faint events, since those are the events that need reorienting to reach awareness. Kelly’s dot-detection effect and Chica’s inhibition-of-return effect both follow, with no schema in the story.
Second, the masking evidence is not about the TPJ. Webb, Kean and Graziano’s profile, capture intact and control weakened, was produced by hiding the stimulus. That shows what happens when awareness of a cue is absent. It does not show that the TPJ computes the awareness, and other theories also predict weaker control without awareness. Workspace theory, for instance, says that unbroadcast information is not available to the systems that hold attention steady.
Third, the central identity claim may be unfalsifiable. It says reports of awareness are the content of the schema. Any report can be assigned to a schema, and any failure of report to a schema failure. What would count against the theory? Graziano’s answer is the control prediction: no schema, worse control. That is testable. But the claim that awareness is nothing beyond the schema’s content is not tested by any control experiment.
Fourth, schemas are cheap. The meeting-room camera from part three has a state variable for where it is pointing. Transformer networks compute attention weights and can be given a model of them. Kathryn Farrell, Kirsten Ziman and Graziano reported in a 2024 preprint that “an agent with an attention schema is better at categorizing the attention states of other agents”. If a simple network can have one, the theory must either grant awareness claims to many machines or draw a line it has not yet drawn.
Fifth, neglect points the wrong way. The most common natural TPJ-region lesion syndrome impairs capture and awareness together. If the schema lived in the TPJ and could be damaged alone, we might expect to meet patients like Ananya often. Clear reports of them are rare.
My replies are partial. The circuit-breaker story is a real rival, and only the stability measure, attention held against unpredictable distraction, separates it cleanly from the schema story. The unfalsifiability point is fair about the theory’s metaphysics and unfair about its neuroscience. Cheapness is a problem for the theory’s claims about machines, and not for its claims about the TPJ. Neglect is the hardest, and I think it lowers the prior that a pure schema lesion is common. It does not show that one is impossible.
There is a sixth objection that I raise against myself. My instrument was written to embody the theory, so its success in producing the predicted profile is unsurprising, and its usefulness lies only in the comparisons. The hint agent and the capture lesion are the informative conditions, because they show how the theory’s signature can be mimicked or confounded. A critic could reasonably ask for a version of the instrument built by someone sympathetic to the reorienting account. That would be a fairer test of whether stability really separates the accounts, and I would welcome it.
What would change my mind
My verdict is that the dissociation the question asks about is predicted, plausible and not shown. Four findings would move me.
- A TPJ perturbation study that measures report, short-interval capture and inhibition of return together, with an IPS site and a vertex control, and finds reports down by at least a third while capture stays within a fifth of control. That would move me to “likely”.
- The same design showing reports down, capture intact and inhibition of return intact. That would favour workspace or higher-order readings over the schema story, since it would dissociate report from control.
- Stability under unpredictable distraction falling after TPJ disruption, while it is preserved after IPS disruption. That is the measure my model says is hardest to fake without a schema.
- A series of right-hemisphere patients in whom report and capture are dissociated in the predicted direction, with lesion overlap at the TPJ rather than the superior temporal or inferior parietal cortex.
The decisive study, specified
Because the argument turns on a missing experiment, it is worth specifying that experiment closely enough for someone to run it. Participants would complete three interleaved tasks in each session. The first is a cueing task with an uninformative peripheral flash, with targets at a short and a long interval, giving a capture index and an inhibition-of-return index from the same trials. The second is a near-threshold detection task with a forced-choice location judgement followed by a four-point awareness rating, giving objective accuracy and subjective report separately. The third is a goal-holding task with unpredictable distractors, giving a stability index.
Stimulation would be delivered at three sites on separate days in counterbalanced order: the right TPJ, located individually from a theory-of-mind localiser; the right intraparietal sulcus, located from a dorsal-attention localiser; and the vertex. Double-pulse TMS timed to the cue and to the target would allow the analysis to ask which processing stage each site contributes to. The primary outcome would be the dissociation index at the TPJ relative to the vertex, with inhibition of return and stability as preregistered secondary outcomes.
Three results would each be informative. If the TPJ index is positive and inhibition of return and stability also fall, the schema theory gains its first direct causal support. If the index is positive while inhibition of return and stability hold, report has been dissociated from control, and workspace or higher-order readings fit better. If capture and report fall together, the TPJ is behaving as a reorienting hub, and the schema has to be looked for elsewhere. A design that cannot fail in any of these three ways is not worth running.
What I don’t know
I do not know whether the TPJ is one region or several for this purpose. Igelström and Graziano’s review stresses its subdivisions, and a schema might sit in only one. I do not know whether single-pulse and double-pulse TMS reach the same tissue in different heads, which matters when comparing Kelly with Chica. I do not know of a published study of open monitoring that analyses the TPJ for objectless-awareness reports. And I do not know whether the inhibition-of-return effect at the TPJ reflects a schema, a reorienting circuit or both, which is the question the decisive study would answer.
Claim Ledger
Each claim below is given with my confidence, the kind of evidence behind it and what would refute it. The IDs are stable across revisions.
| ID | Claim | Confidence | Evidence type | What would refute it |
|---|---|---|---|---|
| C1 | Attention and reported awareness can be dissociated in both directions of at least one kind. | Established | Published evidence | Reanalysis showing blindsight and masking effects are due to weak residual awareness. |
| C2 | Single-pulse TMS at a TPJ site reduces reported detection of faint contralateral targets. | Likely | One published study | A well-powered replication with neuronavigation finding no effect. |
| C3 | TPJ disruption interferes with inhibition of return after uninformative cues. | Likely | One published study | Replication finding intact inhibition of return after right TPJ TMS. |
| C4 | Without awareness of a cue, capture survives and control-dependent effects weaken. | Likely | Several published studies | Masking studies finding equal inhibition of return and stability with and without awareness. |
| C5 | No published study has measured report, capture and inhibition of return under the same TPJ perturbation. | Likely | Literature audit | Any such study, which I would be glad to be shown. |
| C6 | Inhibition of return alone cannot distinguish a schema lesion from a capture lesion. | Established | Toy model, logic | A capture lesion that leaves inhibition of return intact in the model or in patients. |
| C7 | Stability under unpredictable distraction is the best behavioural signature of a schema. | Speculative | Toy model | A schema-free controller matching schema stability across distractor types. |
| C8 | The right-TPJ reorienting account explains the causal data as well as the schema account. | Likely | Argument from published evidence | A TPJ effect on stability for predictable distractors, which reorienting does not need. |
| C9 | The theory places attention at RTM L1, the schema at L3 and social attribution at L4. | Speculative | Conceptual mapping | A clear case where the schema must represent manifestation itself, L2. |
| C10 | The Abhidhamma and Yoga accounts share the theory’s structure of misattribution, and they differ on what is real. | Likely | Textual reading | Scholarly readings of Vācaspati and the Atthasālinī that reject this structure. |
Dated prediction
By 31 December 2030, at least one peer-reviewed study will apply TMS or a lesion analysis to the right TPJ while measuring awareness reports, short-interval capture and inhibition of return in the same participants. I predict it will find reports reduced by more than capture, giving a dissociation index above 0.2 on the definition used here. I also predict inhibition of return will be reduced along with reports. If reports fall and inhibition of return is spared, the schema theory’s control claim is in trouble.
Glossary
- Attention schema
- In Graziano’s theory, a simplified internal model of the brain’s own attention, used to control attention and expressed in reports of awareness.
- Capture
- The fast, involuntary pull of attention to a salient stimulus, measured as a short-interval benefit at a cued place.
- Citta (चित्त)
- Pāli and Sanskrit, a single moment of consciousness in the Abhidhamma analysis.
- Cetasika (चेतसिक)
- A mental factor that arises together with a citta, such as contact, feeling or attention.
- Cit (चित्)
- Sanskrit, awareness as such. Layer L5 of the RTM hypothesis.
- Covert attention
- Attention that shifts without the eyes moving.
- Dissociation index
- In this paper, the proportional drop in awareness reports minus the proportional drop in capture. Positive values mean report suffers more.
- Endogenous attention
- Attention directed by goals and expectations. Slower to deploy and able to be sustained.
- Exogenous attention
- Attention drawn by the stimulus itself. Fast, automatic and brief.
- Fidelity
- In the instrument, how closely the schema copies the actual attention state, from 1 (faithful) to 0 (unrelated).
- Inhibition of return
- Slower responses at a recently cued place after a longer interval, read as the system pushing attention away from places already checked.
- Japā (जपा)
- Sanskrit name of the red hibiscus flower in Vācaspati Miśra’s simile of the crystal.
- Javana (जवन)
- Pāli, “impulsion”: the run of active moments in a cognitive process that follow the adverting moment.
- Manasikāra (मनसिकार)
- Pāli, attention: a universal mental factor, called “the charioteer of associated states”. Also the name of the two adverting cittas.
- Neglect
- A failure to notice or respond to one side of space, usually the left, after right-hemisphere damage.
- Prakāśa (प्रकाश)
- Sanskrit, “light” or manifestation. Layer L2 of the RTM hypothesis.
- Pratyabhijñā (प्रत्यभिज्ञा)
- Sanskrit, “recognition”. Layer L4 of the RTM hypothesis, and the name of a Kashmiri Śaiva school.
- Puruṣa (पुरुष)
- In Sāṅkhya and Yoga, the pure witnessing consciousness, distinct from the mind and its states.
- Sārūpya (सारूप्य)
- Sanskrit, “sameness of form”: in Yoga Sūtra 1.4, the seer’s apparent taking on of the form of the mind’s fluctuations.
- Spanda (स्पन्द)
- Sanskrit, “vibration” or pulsation. Layer L1 of the RTM hypothesis, read here as selective responsiveness.
- TMS
- Transcranial magnetic stimulation: brief magnetic pulses over the scalp that interfere with the cortex beneath.
- TPJ
- Temporoparietal junction: the cortex where the temporal and parietal lobes meet, a hub for reorienting, social cognition and, on this theory, the attention schema.
- Vimarśa (विमर्श)
- Sanskrit, reflexive awareness. Layer L3 of the RTM hypothesis.
- Vṛtti (वृत्ति)
- Sanskrit, a fluctuation or modification of the mind, in the Yoga Sūtra.
Three tests for a controller
Dr Qureshi runs the three tests on a Tuesday afternoon. In the version of this story I find most likely, the results are mixed. Ananya’s capture is a little slow on the left, her inhibition of return is weak, and her reports drop on faint left-side targets. Her profile sits between the neglect pattern and the schema pattern. That is what a small lesion in a crowded hub should produce. He does not clear her for live traffic yet. He books a second session with longer, less predictable distractor runs, because that is the test most likely to show whether the controller in her brain still knows where its attention is.
The question was whether disrupting the TPJ impairs subjective awareness reports more than objective exogenous attention. The honest answer is that it has not yet been shown. Kelly and colleagues found that a TPJ pulse reduced reports of faint dots, with no attention measure. Chica and colleagues found that TPJ stimulation disrupted inhibition of return after reflexive cues, with no report measure. Masking experiments find capture surviving without awareness while control weakens, and they did not touch the TPJ. Each half fits the theory, and together they have never been tested in one study.
The instrument adds three things. Inhibition of return falls after both a schema lesion and a capture lesion, so it can never decide the question alone. A system with no schema can match the standard cueing profile. And stability against unpredictable distraction is the measure most likely to reveal a schema, because it depends on knowing where attention actually is.
For the RTM hypothesis, the result is a placement and a choice. Attention belongs at L1, as selective responsiveness. The schema, if it exists, is a candidate for L3, and the theory designs L3 as a separate model rather than the same-order reflexivity proposed in paper 36. A clean TPJ dissociation would favour the separate model. The upper layers, L2 and L5, are what the theory says the schema describes, and no experiment in this paper can show whether that description is a caricature or an accurate report.
For anyone reading future TPJ papers, the decision rule is short. Ask whether the study measured report, capture and inhibition of return in the same people under the same perturbation, and whether it included a control site. If it measured fewer than all of these, read its conclusion as one half of the argument, however striking the effect. If it measured all of them, read the dissociation index first, then check whether stability was tested, because that is where a schema would be hardest to fake.
Ananya, in the end, is a reminder of why the distinction matters outside a laboratory. A controller who misses alerts is a known risk with known remedies. A controller who catches alerts and cannot say she caught them is a new kind of risk. If Graziano is right, her failure to report is a symptom of a damaged control model, and it may predict lapses that ordinary tests miss. If he is wrong, it is a failure of access and nothing more. The three tests, run together, are the cheapest way to find out which.
Sources
Primary research
- Chica, A. B., Bartolomeo, P., & Valero-Cabré, A. (2011). Dorsal and ventral parietal contributions to spatial orienting in the human brain. Journal of Neuroscience, 31(22), 8143–8149. doi:10.1523/jneurosci.5463-10.2010
- Downar, J., Crawley, A. P., Mikulis, D. J., & Davis, K. D. (2000). A multimodal cortical network for the detection of changes in the sensory environment. Nature Neuroscience, 3(3), 277–283. doi:10.1038/72991
- Farrell, K. T., Ziman, K., & Graziano, M. S. A. (2024). Testing components of the attention schema theory in artificial neural networks. arXiv:2411.00983. arxiv.org/abs/2411.00983
- Guterstam, A., Kean, H. H., Webb, T. W., Kean, F. S., & Graziano, M. S. A. (2019). Implicit model of other people’s visual attention as an invisible, force-carrying beam projecting from the eyes. PNAS, 116(1), 328–333. doi:10.1073/pnas.1816581115
- Karnath, H.-O., Ferber, S., & Himmelbach, M. (2001). Spatial awareness is a function of the temporal not the posterior parietal lobe. Nature, 411, 950–953. doi:10.1038/35082075
- Kelly, Y. T., Webb, T. W., Meier, J. D., Arcaro, M. J., & Graziano, M. S. A. (2014). Attributing awareness to oneself and to others. PNAS, 111(13), 5012–5017. doi:10.1073/pnas.1401201111
- Kentridge, R. W., Heywood, C. A., & Weiskrantz, L. (1999). Attention without awareness in blindsight. Proceedings of the Royal Society B, 266, 1805–1811. doi:10.1098/rspb.1999.0850
- Marshall, J. C., & Halligan, P. W. (1988). Blindsight and insight in visuo-spatial neglect. Nature, 336, 766–767. doi:10.1038/336766a0
- Mort, D. J., et al. (2003). The anatomy of visual neglect. Brain, 126(9), 1986–1997. doi:10.1093/brain/awg200
- Piefke, L., Doerig, A., Kietzmann, T., & Thorat, S. (2024). Computational characterization of the role of an attention schema in controlling visuospatial attention. CogSci 2024; arXiv:2402.01056. arxiv.org/abs/2402.01056
- Posner, M. I. (1980). Orienting of attention. Quarterly Journal of Experimental Psychology, 32(1), 3–25. doi:10.1080/00335558008248231
- Rees, G., Wojciulik, E., Clarke, K., Husain, M., Frith, C., & Driver, J. (2000). Unconscious activation of visual cortex in the damaged right hemisphere of a parietal patient with extinction. Brain, 123(8), 1624–1633. doi:10.1093/brain/123.8.1624
- Saxe, R., & Kanwisher, N. (2003). People thinking about thinking people: the role of the temporo-parietal junction in “theory of mind”. NeuroImage, 19(4), 1835–1842. doi:10.1016/s1053-8119(03)00230-1
- Webb, T. W., Igelström, K. M., Schurger, A., & Graziano, M. S. A. (2016). Cortical networks involved in visual awareness independent of visual attention. PNAS, 113(48), 13923–13928. doi:10.1073/pnas.1611505113
- Webb, T. W., Kean, H. H., & Graziano, M. S. A. (2016). Effects of awareness on the control of attention. Journal of Cognitive Neuroscience, 28(6), 842–851. doi:10.1162/jocn_a_00931
- Wilterson, A. I., Kemper, C. M., Kim, N., Webb, T. W., Reblando, A. M. W., & Graziano, M. S. A. (2020). Attention control and the attention schema theory of consciousness. Progress in Neurobiology, 195, 101844. doi:10.1016/j.pneurobio.2020.101844
- Wilterson, A. I., Nastase, S. A., Bio, B. J., Guterstam, A., & Graziano, M. S. A. (2021). Attention, awareness, and the right temporoparietal junction. PNAS, 118(25), e2026099118. doi:10.1073/pnas.2026099118
- Wilterson, A. I., & Graziano, M. S. A. (2021). The attention schema theory in a neural network agent: controlling visuospatial attention using a descriptive model of attention. PNAS, 118(33), e2102421118. doi:10.1073/pnas.2102421118
- Wolpert, D. M., Ghahramani, Z., & Jordan, M. I. (1995). An internal model for sensorimotor integration. Science, 269, 1880–1882. doi:10.1126/science.7569931
Primary texts and translations
- Atthasālinī and Visuddhimagga XIV.152 on manasikāra, as quoted in van Gorkom, N. (1999). Cetasikas. Zolag. vipassana.info/cetasikas9.html (to verify against Ñāṇamoli’s printed translation of the Visuddhimagga)
- Patañjali, Yoga Sūtra 1.4, with the Vyāsa Bhāṣya and Vācaspati Miśra’s Tattvavaiśāradī, tr. Rama Prasada (1924). Wisdom Library edition
- Encyclopedia of Buddhism, entry on yoniso manasikāra. encyclopediaofbuddhism.org
Official and reference
- Conant, R. C., & Ashby, W. R. (1970). Every good regulator of a system must be a model of that system. International Journal of Systems Science, 1(2), 89–97. doi:10.1080/00207727008920220
Secondary: reviews and theory
- Corbetta, M., & Shulman, G. L. (2002). Control of goal-directed and stimulus-driven attention in the brain. Nature Reviews Neuroscience, 3, 201–215. doi:10.1038/nrn755
- Corbetta, M., Patel, G., & Shulman, G. L. (2008). The reorienting system of the human brain: from environment to theory of mind. Neuron, 58(3), 306–324. doi:10.1016/j.neuron.2008.04.017
- 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
- Graziano, M. S. A., & Kastner, S. (2011). Human consciousness and its relationship to social neuroscience: a novel hypothesis. Cognitive Neuroscience, 2(2), 98–113. doi:10.1080/17588928.2011.565121
- Graziano, M. S. A., & Webb, T. W. (2015). The attention schema theory: a mechanistic account of subjective awareness. Frontiers in Psychology, 6, 500. doi:10.3389/fpsyg.2015.00500
- Graziano, M. S. A. (2022). A conceptual framework for consciousness. PNAS, 119(18), e2116933119. doi:10.1073/pnas.2116933119
- Igelström, K. M., & Graziano, M. S. A. (2017). The inferior parietal lobule and temporoparietal junction: a network perspective. Neuropsychologia, 105, 70–83. doi:10.1016/j.neuropsychologia.2017.01.001
- Koch, C., & Tsuchiya, N. (2007). Attention and consciousness: two distinct brain processes. Trends in Cognitive Sciences, 11(1), 16–22. doi:10.1016/j.tics.2006.10.012
- 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
- Frankish, K. (2016). Illusionism as a theory of consciousness. Journal of Consciousness Studies, 23(11–12), 11–39. author’s eprint
Cite this work
Mahapatra, R. R. Graziano’s attention schema theory and the temporoparietal junction: does a model of attention make awareness? Mind Frontier paper 37 (U0872). rajeshrmahapatra.com. DOI: to be assigned.
@article{mahapatra_mf_u0872,
author = {Mahapatra, Rajesh Ranjan},
title = {Graziano's attention schema theory and the
temporoparietal junction: does a model of
attention make awareness?},
journal = {Mind Frontier},
number = {37},
note = {Topic U0872},
url = {https://rajeshrmahapatra.com/papers/mind-frontier/U0872-attention-schema-theory-tpj-covert-attention.html},
doi = {10.xxxx/placeholder}
}