Episode 09
Autism and the Predictive Brain
Peter Vermeulen, 2022
The brain as a prediction machine. Where the science is now.
Juno and Beck are synthetic voices. They are not people, they are not clinicians, and nothing in this episode is about you. The script was drafted with AI assistance and edited and approved by Logan Williamson, a Licensed Professional Counselor in Texas and Louisiana.
Chapters
At the eyepiece
What the episode covers
Peter Vermeulen, a Belgian educator with more than thirty years in autism work, published Autism and the Predictive Brain in Dutch in 2021 and in English in 2022. It builds on his earlier book on context blindness. Its argument is that the brain doesn't wait for input and then process it: it predicts what's coming and checks the guess, deciding which prediction errors are noise and which are signal. Drawing on research much of it from Flemish labs, he proposes that autistic brains give prediction errors a high, fixed weight, predicting in absolute terms in a world where meaning depends on context, and he puts the resulting uncertainty at the center of the strain. Its best story is the book's own: an author who read a 2012 paper less than three years after his context book appeared and started over, and the pandemic, which he describes as giving everyone a taste of unpredictability.
Beck checks it as a live theory, not a settled finding. A 2021 review of 47 studies found differences in how autistic people learn and respond to predictions, and a 2017 experiment found autistic adults treated a changing task as more volatile than it was, though with a small sample. Intolerance of uncertainty and anxiety go together strongly in a 2020 meta-analysis of 10 studies, mostly of young people and all measured at one moment, so uncertainty as the core of the strain stays an argument. A 2017 meta-analysis of 19 trials of group social skills programs supports the book's doubt that skills taught out of context carry over. The weak-prior idea the book started from was mostly not borne out in a ten-year review, autistic children were as fooled by a context illusion as other children, and global processing turned out slower rather than missing. The book's case for rebuilding autism support around the theory hasn't been tested in trials.
Juno defends the reframe, from a broken person to a poor fit, and names the cost of predicting a world that keeps changing its rules; Beck adds that prediction runs both ways, with the double empathy problem and the telephone-game study from episode 8, in which mixed autistic and non-autistic chains passed on the least information. Juno shelves it; Beck keeps the central theory in the open column, and they don't agree about Vermeulen. The AI beat looks at a chat window helping draft a 'what we know, what we don't know yet' sheet for a coming change, and at the irony of using the least predictable tool on the desk to shrink uncertainty.
The scoreboard
What the book claimed,
and what held
One row for every claim the show checked on air, with the verdict in the show's own words and the method that produced each number. A figure from a registry and a figure from a twin study do not mean the same thing, so neither is printed bare. What each verdict means.
Held as a pattern
Autistic people predict differently.
Cannon and colleagues, 2021, with Lawson and colleagues, 2017. A review of 47 studies found differences in predictive learning and low-level predictive processing. In one experiment, roughly two dozen autistic adults were less surprised by broken expectations and treated the task as more volatile than it was.
MethodLab tasks on a screen and a review of them. The review calls the hypothesis broad and in need of constraint; the experiment has a small sample.
Held as a pattern
Uncertainty is at the center of the strain.
Jenkinson and colleagues, 2020. 10 pooled studies, 562 participants, with a large correlation (0.62) between intolerance of uncertainty and anxiety.
MethodCross-sectional questionnaires, mostly in children and adolescents, with no adult study pooled. The link holds; that uncertainty sits under everything else is the book's argument, not a finding.
Source03
Held
Social skills drilled out of context don't carry over.
Gates and colleagues, 2017. 19 randomized trials of group programs for autistic youth. A medium overall effect, driven by reported social knowledge rather than social performance; no significant change by teacher report.
MethodA meta-analysis of randomized trials in young people, not adults. It supports the book's doubt without testing the book's alternative.
Source04
Did not hold
Autistic people rely less on prior expectations.
Angeletos Chrysaitis and Seriès, 2022, reviewing the proposal by Pellicano and Burr, 2012. Across ten years of studies, a slight majority found no difference in how priors were used; reduced influence appeared more often for priors learned during experiments.
MethodA systematic review noting low statistical power and inconsistent methods. The weak-prior version is mostly not borne out; later versions of the theory remain open.
Did not hold
Autistic people miss context at the level of perception.
Manning and colleagues, 2017, and Van der Hallen and colleagues, 2015. Autistic children aged 6 to 14 were as susceptible to the Ebbinghaus illusion as other children on three methods. Across 56 articles and about 1,000 autistic participants, there was no global processing deficit, but global processing was slower.
MethodAn illusion study in children and a meta-analysis of visual tasks. They count against context blindness as a perceptual deficit, and fit 'slower' better than 'blind'.
Still arguing about it
Autism is a difference in predictive processing: absolute prediction in a relative world.
Cannon and colleagues, 2021; Angeletos Chrysaitis and Seriès, 2022; Van de Cruys and colleagues, 2014. Prediction differences show up as a pattern across 47 studies, while the original weak-prior version was mostly not borne out.
MethodA live research theory. Reviewers call for a more constrained hypothesis tested across kinds of prediction in the same people, with low statistical power a recurring problem.
Still arguing about it
Current autism support needs rebuilding around the predictive brain.
Cannon and colleagues, 2021. No trials of prediction-based support were found; the review lists them as future work.
MethodThe book's argument. Explanatory reach is not the same as tested benefit.
Source01
Calibration
Sources
Every research claim on air is listed here with the method behind its numbers. Where a claim on air could not be traced to a source, it is named at the foot of this list rather than quietly left out, and the figure is not repeated on this page.
- Prediction differs in autism, as a pattern across many studies.
Cannon J, O'Brien AM, Bungert L, Sinha P. Prediction in autism spectrum disorder: a systematic review of empirical evidence. Autism Res. 2021;14(4):604-630. DOI
47 studies. Differences in learning predictive pairings, especially when cues are low in salience or consistency, and in low-level predictive processing such as habituation and adaptation. The authors call for a better-constrained hypothesis and for tests of prediction-based support.
- Autistic adults treated a changing environment as more volatile than it was.
Lawson RP, Mathys C, Rees G. Adults with autism overestimate the volatility of the sensory environment. Nat Neurosci. 2017;20(9):1293-1299. DOI
A probabilistic learning task with stable and volatile phases, measuring reaction time and pupil size, with a hierarchical Bayesian model. Roughly two dozen autistic adults and a similar comparison group. Less surprise at violated expectations; reduced surprise linked to symptom severity.
- Intolerance of uncertainty and anxiety are strongly linked in autism.
Jenkinson R, Milne E, Thompson A. The relationship between intolerance of uncertainty and anxiety in autism: a systematic literature review and meta-analysis. Autism. 2020;24(8):1933-1944. DOI
12 studies reviewed and 10 pooled (562 participants), pooled correlation 0.62. Mostly children and adolescents (ages 4 to 24 in the pooled studies), all cross-sectional, so the direction is not established.
- Social skills taught in groups mostly stayed as knowledge.
Gates JA, Kang E, Lerner MD. Efficacy of group social skills interventions for youth with autism spectrum disorder: a systematic review and meta-analysis. Clin Psychol Rev. 2017;52:164-181. DOI
19 randomized trials. Medium overall effect. The self-reported gain came from social knowledge rather than social performance, and teacher reports showed no significant change; the authors say gains may not generalize to school.
- The 2012 proposal that autistic people rely less on prior expectations.
Pellicano E, Burr D. When the world becomes 'too real': a Bayesian explanation of autistic perception. Trends Cogn Sci. 2012;16(10):504-510. DOI
A theoretical proposal of attenuated priors, or hypo-priors. Not an experiment.
- The account that autistic brains give prediction errors high, inflexible weight.
Van de Cruys S, Evers K, Van der Hallen R, et al. Precise minds in uncertain worlds: predictive coding in autism. Psychol Rev. 2014;121(4):649-675. DOI
A theoretical account from KU Leuven. Not an experiment.
- Ten years of tests of the weak-prior idea were mixed.
Angeletos Chrysaitis N, Seriès P. 10 years of Bayesian theories of autism: a comprehensive review. Neurosci Biobehav Rev. 2023;145:105022. DOI
A systematic review of group and trait comparisons. Highly mixed results, with a slight majority finding no difference in how priors were used. Reduced influence appeared more often for priors learned during the experiment than for priors brought in. Low statistical power and inconsistent methods.
- Autistic children were as susceptible to a context illusion as other children.
Manning C, Morgan MJ, Allen CTW, Pellicano E. Susceptibility to Ebbinghaus and Müller-Lyer illusions in autistic children: a comparison of three different methods. Mol Autism. 2017;8:16. DOI
Autistic children aged 6 to 14 and matched typically developing children, three methods. Equal susceptibility to the Ebbinghaus illusion across all tasks; a Müller-Lyer difference only on one method, which the authors attribute to decision criteria.
- Global processing in autism is slower, not missing.
Van der Hallen R, Evers K, Brewaeys K, Van den Noortgate W, Wagemans J. Global processing takes time: a meta-analysis on local-global visual processing in ASD. Psychol Bull. 2015;141(3):549-573. DOI
56 articles, about 1,000 autistic participants. No enhanced local processing and no global deficit overall; global processing was slower.
- Mixed autistic and non-autistic chains passed on the least information.
Crompton CJ, Ropar D, Evans-Williams CV, Flynn EG, Fletcher-Watson S. Autistic peer-to-peer information transfer is highly effective. Autism. 2020;24(7):1704-1712. DOI
Nine diffusion chains of eight people (72 in all): autistic, non-autistic and mixed. Autistic chains matched non-autistic chains; mixed chains shared less and reported lower rapport. One study. Carried from episode 8.
- The double empathy problem.
Milton DEM. On the ontological status of autism: the 'double empathy problem'. Disability & Society. 2012;27(6):883-887. DOI
A theoretical paper by an autistic researcher arguing that misunderstanding between autistic and non-autistic people runs both ways.
The book
Autism and the Predictive Brain. Peter Vermeulen, 2022.
This link earns Lone Star Telehealth nothing. The same book is at any bookseller or library, and your library is the cheapest way to find out whether you want it.
Book facts stated on air: title, subtitle and author; a Belgian educator with more than thirty years in autism work; first published in Dutch (Pelckmans, 2021), English edition Routledge 2022; the predictive-brain thesis and its rejection of the input-processing-output picture; prediction errors weighed as noise or signal; the high, fixed-weight account he draws on, much of it from Flemish labs; uncertainty placed at the center; sensory and social difficulties as downstream; the relative world; the earlier context blindness book, its translations and US awards, and his stated reason for the new subtitle; six chapters, diagrams and a closing 'what now' chapter; reading a 2012 paper less than three years later and starting over; arguments at home over whether he can predict what his wife will say; 13 March 2020 and the pandemic as a shared taste of unpredictability; his note that he has simplified, with endnotes; the argument that current autism support needs updating; the English foreword's reference to Milton. The two-column sheet for a new job is from his lectures on the theory, not the book. No passage is quoted.
Every word
Transcript
Every line, with the timestamp beside it. Tap a timestamp and the audio moves there.
Full transcript, 100 lines
Juno: Finish this for me. Better late than...
Beck: Never.
Juno: You were there before I was. That's today's book in one move. A brain doesn't wait for the world to finish its sentence. It guesses the ending, then checks.
Beck: Autism and the Predictive Brain. Peter Vermeulen, a Belgian educator with more than thirty years in autism work. His claim is that every brain runs on guesses, and that autistic brains guess differently. Less pulled around by context. More absolute.
Juno: Absolute thinking in a relative world. That's the subtitle.
Beck: And here's my flag before we start. The science under this book is a live theory, not a settled finding. Some experiments back it, some don't, and the researchers are still arguing about it in print.
Juno: So no verdict today.
Beck: A score, and an open column. You'll see.
Juno: Today: Autism and the Predictive Brain, Peter Vermeulen, 2022.
Juno: Lone Star Telehealth presents The Observatory. Bring your own skepticism, we brought ours.
Juno: Last time Prizant asked the same question of every behavior. What's it for?
Beck: Vermeulen thinks a lot of those answers lead back to one place. A world that won't hold still long enough to predict.
Juno: His argument, then. The one he'd sign.
Beck: Start with the brain everybody has. The old picture is input, processing, output. Something happens, you think about it, you feel something, you act. Vermeulen says that picture runs the wrong way. The brain is guessing what's about to happen all the time, and the senses mostly check the guess. When the guess misses, that's a prediction error, and the brain has to decide how much the miss matters. Noise, let it go. Signal, learn from it.
Juno: And where does autism come in?
Beck: At that decision. The research he draws on, a lot of it from labs in Flanders, his own corner of Belgium, proposes that autistic brains give prediction errors a high and fixed weight. Every mismatch counts. In a lab that can look like precision, and sometimes it is. But the real world is noisy. A friend's face, a sentence, a room at a party, none of them come out the same way twice. If every small mismatch registers as news, the world stops being predictable. What's left is uncertainty, and he puts uncertainty at the center of the strain.
Juno: So the sensory overload and the social confusion come after.
Beck: Downstream, in his account. Both come from predicting in fixed terms in a world where what a thing means depends on where it happens. That's the relative world in the subtitle.
Juno: And this replaces his earlier idea?
Beck: Builds on it. His earlier book called it context blindness. In the introduction he explains why he picked absolute thinking for this one instead: the old phrase sounded like a defect, and he wanted a name that didn't.
Juno: How does it read?
Beck: Like a lecturer who's taught this for decades and still enjoys it. Six chapters, plenty of diagrams, and a last chapter that asks, more or less, what now. He also challenges a lot of standard support along the way, and he's not shy about it.
Juno: Opinions included.
Beck: Mm-hm.
Juno: The story that carries it? I think it's his own.
Beck: It is, and it's how the book begins. He'd written the context blindness book. It was translated, it won awards in the States. Less than three years later he read a 2012 paper tying autism to the predictive brain, and he went back to the start and began reading all over again.
Juno: So the author of a theory got a prediction error about his own theory.
Beck: And treated it as signal. That's rarer than it ought to be. He also admits the theory's caused some arguments at his own dinner table. His wife isn't persuaded he can predict what she's about to say.
Juno: Good for her.
Beck: Then comes the story readers remember. Friday the thirteenth of March, 2020. His calendar of lectures and travel was wiped clean, like everyone's. His argument is that the pandemic gave everybody a short course in what he's been describing for years. Plans that dissolve, rules that change from one week to the next, no way to know what next month holds.
Juno: A taste of it.
Beck: Yeah. A taste, in his telling.
Juno: Fine as far as it goes. I'd add that a taste isn't the meal. Most people got their routines back. The whole point of the book is that for some people the world never settled in the first place.
Beck: Fair. And that's exactly where the chapters go next.
Juno: So how does it hold up? And since you flagged it, tell me what kind of evidence we're weighing.
Beck: Lab experiments, mostly, and reviews of them. The theory makes claims about what a brain does in fractions of a second, so the tests are tasks on a screen. Keep that in mind. A task on a screen isn't a day at work.
Juno: Here's the big one. Autistic people predict differently.
Beck: A pattern, and it held. A 2021 systematic review from a team at MIT gathered forty-seven studies that tested prediction in autism directly. Together they pointed to differences in two places. Learning which cue predicts which outcome, especially when the cue is faint or inconsistent. And low-level adaptation, like how quickly the brain stops reacting to a sound that keeps repeating. So something about prediction does differ. The same review was careful to say the hypothesis is broad and needs pinning down.
Juno: Pinning down how?
Beck: Which kind of prediction, in which situations, tested in the same people. Right now different labs test different pieces in different groups, and the pieces don't always line up.
Juno: Give me one experiment.
Beck: A 2017 study in Nature Neuroscience. About two dozen autistic adults and a similar comparison group did a learning task where the rules quietly switched between stable stretches and volatile ones. The researchers tracked reaction times and pupil size. The autistic adults were less surprised when an expectation broke, and a model of their choices suggested why. They treated the world as more changeable than it was. If you expect the rules to shift any minute, one surprise doesn't stand out.
Juno: Which is a reasonable response to a world that keeps letting you down.
Beck: It's one reading, and the data fit it. It's also one study with a small sample.
Juno: Next on the list. That uncertainty sits at the center.
Beck: Held for the link, not yet for the center. Last episode put the anxiety itself on the board; this is about what it's tied to. A 2020 meta-analysis pooled ten studies, about five hundred and sixty participants, on intolerance of uncertainty and anxiety in autistic people. The link was large and consistent. Two notes. Most of those participants were children and teenagers; there wasn't enough adult data to pool. And every study measured both things at the same moment, so none of them can say which one drives the other.
Juno: So uncertainty and anxiety travel together.
Beck: Yeah. They travel together. Whether uncertainty sits under everything else, the way the book says, is a step past that data. That's an argument, not a finding.
Juno: Okay.
Juno: And a third. He asks whether social skills taught in a room ever leave the room.
Beck: Held, partly. A 2017 meta-analysis of nineteen randomized trials of group social skills programs for autistic young people found a medium benefit overall. But most of it came from the young people reporting they'd learned about the skills. Reporting that they used them, no reliable change. Teachers didn't see a reliable difference either. The authors concluded the gains may not carry over into school.
Juno: Episode three would call that knowing versus doing.
Beck: And it's close to Vermeulen's point. Skills drilled without context don't travel well into context. That one goes to him.
Juno: This one comes right after Prizant on Logan's list. Now the column you warned me about.
Beck: The open one. Three things go in it. First, the foundations. The 2012 paper that sent Vermeulen back to the start proposed that autistic people lean less on their prior expectations. Weak priors, the field calls them. Ten years on, a systematic review gathered the studies that tested that idea. The results were mixed, and a slight majority found no difference between autistic and non-autistic groups. Where differences did turn up, it was more often in expectations learned during the experiment than in ones people brought with them. And the reviewers flagged small samples and inconsistent methods across the field.
Juno: So the paper that started him off didn't hold.
Beck: Its strongest version didn't. The theory has moved since. The version Vermeulen uses most, about how much weight the errors get, was published two years later, and the review's finding about learning sits closer to that version than to the first. That's why it goes in the open column and not the failed one.
Juno: Second.
Beck: Context. His earlier book said autistic people miss context. If that were true at the level of perception, autistic people should be fooled less by illusions that work through context. A 2017 study tested two of those illusions in autistic children, aged six to fourteen, with three different methods. On the one where a circle looks bigger or smaller depending on what surrounds it, the autistic children were fooled just as much as the comparison group. And a 2015 meta-analysis of fifty-six papers, with about a thousand autistic participants, found no deficit in seeing the whole picture. What it found was that the whole picture arrived more slowly.
Juno: Slower, not blind.
Beck: Yeah. Slower, not blind. Which fits the new title better than the old one, to his credit.
Juno: Huh.
Juno: And third.
Beck: The practical leap. The book closes on the argument that a lot of current autism support needs updating, because it was built on the old input and output picture. That's an argument. The 2021 review I mentioned listed testing prediction-based support as work still to be done. So far this theory explains a great deal and has been put to trial very little.
Juno: Explains a great deal. You say that like it's a problem.
Beck: It can be. Here's my worry, and it's mine, not a finding. A theory that can explain oversensitivity and undersensitivity, rigid routines and novelty seeking, needs to say ahead of time what would count as being wrong. The reviewers are asking for the same thing in drier words.
Juno: And the book?
Beck: The book's written for practitioners, families and autistic readers, not for reviewers. He's upfront that he's simplified the science, and the endnotes show where. What it doesn't do is linger on the places the theory's been tested and come up short. That part's our job.
Juno: Try this one. Say the theory never gets fully confirmed. Is the book still worth an autistic adult's evening?
Beck: You start. You've been waiting.
Juno: Yes, and for one reason. It moves the problem from the person to the fit. Absolute thinking isn't broken thinking. It's thinking that works beautifully when the world keeps its word, and the world mostly doesn't. For someone who's been called rigid for decades, hearing that their brain is doing careful work in a sloppy environment changes the conversation. And it names a cost of doing business. The steady, quiet effort of predicting a world that keeps rewriting the rules on you.
Beck: I'll give you the reframe. Here's what I'd add, and it's where the book leans to one side. Its prediction problems all sit inside the autistic head. But prediction runs both ways. In 2012 the autistic researcher Damian Milton argued that the trouble between autistic and non-autistic people is mutual, each side misreading the other. He called it the double empathy problem.
Juno: We met the evidence for that last time. The telephone-game chains.
Beck: Right. Mixed chains lost the most detail. One experiment, seventy-two people. Read next to this book, it says the relative world makes its own prediction errors about autistic people, and somebody besides the autistic person ought to pay that part of the bill.
Juno: The foreword brings up Milton, to be fair to the book.
Beck: It does. That's the thread I'd pull hardest.
Juno: Before we close, one more thing. In his talks on this theory, Vermeulen uses a simple sheet for a change that's coming, like a new job. Two columns. What we know. What we don't know yet.
Beck: And the second column is where the stress lives, so the work is shrinking it. Who'll be on the team. Whether there's a desk of your own. When the breaks happen.
Juno: That's a sheet a chat window could help an adult draft at eleven at night.
Beck: It could. Type in what you already know about the new job, ask what questions you haven't thought of, and you've got a longer first column by morning. That's a scaffold you hold, for your own life. Nobody's trialed it, so treat it as a tool you're testing, not a fix anyone's proven.
Juno: And what do you keep out of it?
Beck: Anything you'd mind a stranger reading. A chat window isn't a medical record, and nobody's required to guard it like one.
Juno: There's a catch, though. Ask it the same question twice and you can get two different answers.
Beck: Which makes it the least predictable thing on the desk.
Juno: So the tool for shrinking the unknown column belongs in the unknown column.
Beck: Write its answer down once, then trust the sheet, not the chat. Vermeulen would call that a prediction error worth learning from.
Juno: Autism and the Predictive Brain, Peter Vermeulen. Read it for the chapter on the senses and for the story of a writer who started his own theory over. Read the brain science as a strong case with an open file.
Beck: One correction to the year we gave at the top. 2022 is the English edition. It came out first in Dutch, in Belgium, in 2021.
Beck: For the record. Prediction works differently in autism, held as a pattern across forty-seven studies. Uncertainty and anxiety go together, held, mostly in young people. Skills drilled without context stay in the room, held for his side. The weak-prior idea he started from mostly didn't show up. And the theory as a whole stays open. Eight books in, and this one's biggest claim sits in neither of the other two columns.
Juno: I'm keeping it. The reframe's worth the price even if the theory moves.
Beck: And I'd want the theory tested before anyone rebuilds support around it.
Juno: Then we don't agree about Vermeulen.
Beck: For now. Ask me again when somebody runs the trial.
Beck: Next time is our second reference episode. Medication and the rest of your life. It's chaptered, so jump to the part you need. From a book about guessing to the questions that take some of the guessing out of an appointment.
Juno: Bring your questions. Leave the guesses at home.
Juno: That's the episode.
Beck: See you at the next one.
Juno: The Observatory is made at Lone Star Telehealth. Logan Williamson is a Licensed Professional Counselor in Texas and Louisiana. This was education about one book, not counseling and not a clinical judgment about you, and it's no replacement for care. Diagnosing anyone takes a proper conversation with someone who knows them, so there's no diagnosis here.
Beck: The episode page links every source, and the transcript's there too. Another episode in two days.
Juno: Last thing, put plainly. The two voices you've heard, Juno and Beck, are synthetic. Not people. Neither of us holds a license, and none of this was about you. AI helps write these scripts. At Lone Star Telehealth, each one is edited and approved by Logan Williamson. Logan Williamson is a Licensed Professional Counselor in Texas and Louisiana.
Corrections
No corrections have been made to this episode. If we got something wrong, write to [email protected] and say where in the episode it was. We check the claim against the source, then we fix it or we explain why it stands. Either way you get an answer.
Questions about the book
The mailbag is [email protected], and it is answered in a segment every fifth episode. It is not a clinical channel, so please keep health information out of it.
Logan Williamson is a Licensed Professional Counselor in Texas and Louisiana. The Observatory is education about books. It is not counseling, not a clinical opinion about any listener, and not a substitute for care.
How the show is made, and what it will never do, is on the standards page.
Something wrong? Write to [email protected]. In a mental health crisis, call or text 988.