Chapter 26: False-Belief Understanding and Predictive Processing

Welcome to Last Minute Lecture.

This free chapter overview is designed to help students review and understand key concepts.

These summaries supplement not replaced the original textbook and may not be redistributed or resold.

For complete coverage, always consult the official text.

Welcome back to the Deep Dives where we take the densest academic material and turn it into the knowledge you need to feel, well, truly ahead of the curve.

Today we are really getting into it.

We're tackling one of the biggest, most fundamental debates in all of cognitive science.

Which is, how in the world does one person understand another?

How do we predict what someone's going to do next?

It's a huge question.

And this deep dive, which is based on a chapter by Leon de Bruijn, is really a cross -examination of how we think we do that.

The core disagreement is about the mechanism itself.

Right.

Is it because we have some kind of special internal machine for simulating other people's minds?

Or is it something much more direct, something that happens out in the world in the interaction itself?

For a long, long time, the answer was pretty much settled.

It was this idea called theory of mind or the mind -dreading account.

Can you just lay out the basics of that for us?

Sure.

The traditional view, you'll hear it called theory theory or simulation theory.

It basically works on one big assumption, that when we interact with people, we're relying on this specialized internal cognitive thing.

We're constantly tracking and mentally reading another person's hidden internal states.

So we're attributing things to them.

Beliefs, desires, intentions.

Exactly.

So if I see you grabbing your keys, I'm not just seeing an action.

I'm internally running a little simulation that says, he desires to leave and he believes his car's in the parking lot.

I use that little model of your mind to figure out what you're doing.

And this whole idea, this entire framework, was really built on the back of one key piece of evidence.

One key experimental measure, yeah.

Research into false belief understanding or FBU.

Okay.

So why false belief?

Why isn't just understanding what someone truly believes enough?

Because that's the acid test.

If you can understand that someone holds a belief that is false, that's different from what you know is true about the world.

Ah.

So you have to separate your own knowledge from their mental state.

Precisely.

If my colleague thinks the meeting is at three, but I know it was moved to four, my ability to predict that she'll show up at the wrong time proves I'm operating on her mistaken belief, not my correct one.

So FBU was seen as like the gold standard for proving you have this mind -dreading capacity.

But like any big idea in science, this one got a major challenger.

A huge one.

The 4E cognition movement that's embodied, embedded, and active and extended cognition.

And this is where the showdown really begins.

It is.

Because the 4E camp just fundamentally disagrees with the whole premise.

They say social understanding is not primarily about decoding hidden internal states.

It's direct, it's situated, it's something that arises from how we interact with the world and with each other.

Okay.

So they're not just saying the results of these false belief tests are wrong.

They're saying the entire design of the experiments is flawed from the get -go.

Exactly.

They argue that it completely misunderstands what social life is actually like.

It focuses so much on what's hidden inside the head that it misses all the social information that's right there in front of us, in gestures, in posture, in the context of the situation.

And this is where the author, Leon de Bruijn, sets up the mission for our deep dive today.

It's sort of a two -part goal.

Right.

First, we need to really walk through the FBU findings, especially the new surprising evidence, and see how the 4E criticisms hold up.

And second, he's going to introduce a pretty radical alternative framework,

something called predictive processing, or PP.

And we're going to use that framework to reinterpret all of these findings and see if it can maybe bridge the gap between these two Warren camps.

So we're kind of moving from this idea of static internal cogs and wheels in the brain to a much more dynamic interactive view.

A view of the brain as this situated error minimizing system.

And we're going to ask a really interesting question.

What if failing a classic test isn't a failure of social skill, but actually a sign that the brain is just doing its fundamental job, which is trying to make sense of the world?

That's a huge shift in perspective.

Okay.

Let's start at the beginning.

Let's look at the evidence that built this whole mind -reading empire in the first place.

Let's do it.

So the story of theory of mind has a very clear starting point, right?

A 1978 paper that asked a very provocative question.

Does the chimpanzee have a theory of mind?

That study by Premack and Woodruff is the origin story for this entire field.

And they have this chimpanzee, Sarah.

What did they have her do?

They showed her videos of a human actor who was, you know, facing some kind of problem.

Like they couldn't reach a banana that was hanging from the ceiling.

Okay.

And then Sarah was given a choice of photographs and she had to pick the one that represented the solution to the problem.

So in this case, she'd have to pick the picture of a chair.

And she was good at it.

She was very good at it.

And Premack and Woodruff took that as evidence that she understood the actor's goal, their intention, that she had a theory of mind.

But the philosophers at the time,

they weren't so convinced.

Not at all.

People like Bennett, Dennett, and Harmon immediately pointed out a big flaw in the logic.

They said, look, Sarah might just be choosing based on her own knowledge of how the world works.

She knows chairs are for climbing on.

She doesn't need to know what's in the actor's head.

Exactly.

She's not representing the actor's internal mental state.

She's just solving a physical problem.

And that's where the demand for a false belief measure came from.

It became the absolute requirement to prove mind -reading.

Which brings us to the classic test, the one that really set the standard for decades, the maxi test.

Developed by Wimmer and 1983,

this became the gold standard for what's called an elicited response test, meaning the child has to verbally tell you the answer.

Okay, walk me through the setup.

What actually happens in the maxi test?

So it's a little story.

A character named Maxi puts his chocolate in, let's say, the green cupboard.

Then he leaves the room.

Tempolo.

While he's gone, his mother comes in and moves the chocolate to a new spot, maybe a blue drawer.

And Maxi, of course, doesn't see this happen.

He comes back wanting his chocolate.

And then comes the big question.

The big question.

Where will Maxi look for his chocolate?

And to get this right, the child has to totally ignore what they know to be true, that the chocolate is in the drawer.

An answer based only on Maxi's outdated false belief.

It's a pure test of mental perspective taking.

You have to inhibit your own knowledge.

And the results were just incredibly consistent.

You saw the same pattern over and over.

They did variations like the famous Smarties test.

Where the kid is shown a Smarties tube, but it's full of pencils.

Right.

And then you ask them, what will your friend who hasn't looked inside think is in the tube?

And what was the pattern?

Three -year -olds consistently fail.

They make what's called the false belief error.

They answer based on reality.

They'll say Maxi will look in the drawer or that their friend will think there are pencils in the tube.

So they just can't get out of their own head.

They can't inhibit their own knowledge of reality.

That's the interpretation.

Yes.

But then around age four, something clicks.

Four -year -olds consistently pass.

And so for decades, the conclusion in cognitive science was that theory of mind, this core social machinery emerges around the age of four.

Okay.

So let's pivot back to the 40 cognition critique because this conclusion is exactly what they take issue with.

They're not just questioning the age.

They're questioning the entire setup of the test.

Yes.

Thinkers like Gallagher, Hutto, Zahavi, they argue that the Maxi test is built on a completely misleading picture of what social understanding is.

They see us as these, you know, detached little scientists observing the world.

So let's break down their objections.

How do the big ideas of Fourier embodiment embeddedness and so on apply here?

Okay.

Let's start with the perspective issue.

The test forces the child into a purely observational third -person perspective.

They're a passive audience member watching a puppet show, and then they're asked for this abstract explanation about a third party.

But the 4E folks say that's not how we normally understand people.

Not at all.

They insist that our primary mode of understanding is second -person interaction.

It's what we do with each other face to face in a shared situation.

That detached third -person analysis is something that comes much later, if at all.

It's a specialized skill.

Okay.

That makes sense.

We learn by doing by interacting.

Then there's embodiment.

The 4E view is that we understand others directly through their bodies, through their actions, their expressions.

So the intention isn't some hidden thing in the head.

It's expressed in the behavior itself.

If I see a toddler reaching with all their might for a toy, I perceive the intention in the action.

I don't need to run some complex internal simulation about their belief, desire, psychology.

The meaning is right there on the surface.

For the most part, yes.

And that leads to the third point.

Embeddedness or situatedness.

Social understanding never happens in a vacuum.

It's always in a rich context, constrained by our shared history and the practical demands of the situation.

The maxi -test strips all of that away.

So it puts this artificial cognitive load on the child.

It does.

It forces them to reason in this abstract, context -free way that's totally unlike real life.

So the 4E critique is basically that this sophisticated verbal mind -reading that the maxi -test measures,

well, if it exists, it's a late -developing secondary skill.

And they would predict that a more basic non -mentalistic understanding based on direct interaction should show up much, much earlier.

That was their prediction.

But that's exactly what the next wave of studies, the spontaneous response measures, would put to the test.

And the results, they were not what the 4B camp expected.

Right.

So the traditional story puts FBU at age four.

Let's see how these newer non -verbal studies just completely upended that timeline.

This is where things get really interesting.

The big shift here was moving away from asking for a verbal answer, the elicited response, which seemed to be the big stumbling block for younger kids.

Right.

The new methods measure spontaneous non -verbal behaviors, things like where a baby looks or whether they try to help someone.

It was a direct response to the criticism that the old tests were just too demanding in terms of language and executive function.

So they were trying to strip the task back to its pure social core.

Exactly.

Let's start with the violation of expectation or VOE paradigm.

How does that even work?

The basic idea is really clever.

It's based on the fact that babies, like all of us, look longer at things that are surprising or unexpected.

So longer looking time is taken as a sign that their expectation has been violated.

So you can kind of infer what they were expecting to happen.

That's the logic.

And a huge study here was by Onishi and Belarjan in 2005.

They tested infants at just 15 months old.

15 months.

So that's an infant who might not even be walking steadily yet.

Right.

And what they did was show these infants the classic object transfer scenario.

An experimenter hides a toy, but then the toy gets moved while the experimenter isn't looking, creating a false belief.

Okay.

Then the experimenter comes back and reaches for the toy.

The key finding was that the infants looked significantly longer when the experimenter reached for the actual location of the toy.

Wait, so they were surprised that the person acted according to reality.

Exactly.

They were surprised that the agent didn't act according to her false belief.

It implies that the 15 -month -olds were expecting her to look where she last saw it, not where they knew it was.

Wow.

So that's evidence for FBU almost three years earlier than the maxi test showed.

A massive difference.

And it wasn't just about location.

Another study by Song and Belarjan found the same thing with object identity.

Okay.

So looking is one thing.

But can infants actually, like, predict where someone's going to go?

That brings us to the anticipatory looking paradigm.

Yeah.

This uses eye tracking cameras to see where a child looks in anticipation of an action.

And the first hint that something was weird with the maxi test results came from a study by Clements and Perner.

What did they find?

They found that three -year -olds, the very kids who were failing the verbal test, were actually looking at the correct false belief location.

Their eyes knew where the character would look, even if their mouths gave the wrong answer.

So the knowledge was in there somewhere.

They just couldn't say it.

It was a stunning finding.

The failure wasn't a competence failure, but some critics pointed out that the setup still used some verbal prompts, which might have confused things.

Just where Salve -Kate and her colleagues came in.

Right.

In 2007, they tested 25 -month -olds, so just over two years old, and they stripped out all the verbal priming.

And what happened?

In their setup, the agent would return after missing the toy being moved and just look toward the boxes.

And even with no questions, no prompts, the toddler's eyes immediately darted to the location where the agent falsely believed the toy was.

So by age two, they're spontaneously predicting actions based on false belief.

That's solid.

But this raises a problem for the 4E view, doesn't it?

These are still passive third -person tasks.

That's a great point and a huge challenge to their theory, which is why the next paradigm is so incredibly important, the active helping paradigm.

Because now we're moving squarely into that second person, interactive space that 4E cares so much about.

This is where the infant has to actually do something.

A study by Butlerman and colleagues in 2009 tested 18 -month -olds.

An agent would try and fail to open a box where he thought a toy was.

But the toy had been moved and the agent had a false belief.

Correct.

Now to truly help, what does the infant have to do?

They can't just help him open the empty box he's struggling with.

No, they have to infer his real goal, which is the toy, and realize he's going to the wrong place.

And that's what they did.

The 18 -month -olds would bypass the box the agent was trying to open, go to the correct box, get the toy out, and give it to him.

That's incredibly sophisticated.

Its goal inference, action, prediction all rolled into one.

It's very rich.

And what's more, another study showed they could do this even when verbal requests were involved, which really undermines the idea that it's just a language problem in the maxi test.

Okay, so there's one last study we should cover here, which kind of tried to bridge the gap, the Duplo test.

Yeah, by Rubio Fernandez and Jertz.

This was basically a souped -up, highly supportive version of the maxi test, designed to give three -year -olds the best possible chance to succeed by making it more interactive and situated.

How did they make it more situated?

A couple of key changes.

First, the character, a Duplo girl, stayed in the scene.

She just had her back turned, which made it really obvious that she couldn't see the switch.

So the constraint was right there, visually.

Right.

And the experimenter would even say things like, she can't see what I'm doing, can she?

To reinforce it.

And what were the results?

When they asked the three -year -olds to act out what the girl would do,

an incredible 80 % of them passed.

They made the Duplo girl look in the false belief location.

80 % compared to, what, 20 % on the standard test?

A huge difference.

And if they remove those interactive, supportive elements, the success rate just plummeted again.

So this really seemed to support the 4E idea that the interactive nature of the task is critical.

It does.

So now we're left with this big developmental paradox.

Infants pass, three -year -olds fail verbally, but pass when the test is more interactive.

So we have to figure out why.

What is actually going on?

This is really the heart of the whole debate.

The spontaneous results show FBU is there way earlier than we thought.

Mind -reading folks can say, see, it's a basic early capacity, just like we said.

But the big question is still on the table.

Is what we're seeing in these infants really mentalistic?

Are they attributing a belief or are they just tracking something simpler, like behavior?

And to try and solve this paradox, early success, later failure,

the field is kind of split into a bunch of competing camps.

About five main ones, yeah.

We need to walk through them to see how they each try to explain this weird pattern of data.

Let's start with the two camps that still believe in an internal specialized mind -reading system.

First up, the dual system accounts.

Right.

These accounts propose that we don't have one theory of mind system, we have two.

And they're functionally separate.

How are they different?

Well, according to Aperly and Butterfill, you have system one, which is a minimal theory of mind, and system two, which is the full -blown propositional one.

Okay, what's a memelier of mind?

It's early, it's fast, and it's non -propositional.

It doesn't use language -like concepts like belief.

Instead, it just tracks what they call off -target registrings.

So it's basically just noting that agent registered the object at that location, and it knows that registration is now out of date or off -target.

Exactly.

It's a simpler, more direct tracking system, and that's enough to explain why infants pass the spontaneous tests.

System two is the slow, effortful, language -based system you need for the verbal maxi -test, and it just isn't fully developed in three -year -olds.

So that's one dual system view.

Bailarge and inner -colleagues have a slightly different take.

They also propose two subsystems, but they slice it a different way.

Subsystem one handles early stuff, like goals and reality congruent beliefs.

Things that match what the child knows is true.

And subsystem two.

That's what you need for reality in congruent states, like false beliefs.

But for them, the failure on the maxi -test isn't about the belief itself, it's about processing overload.

The verbal format just puts too many demands on the child's brain at once.

What are those demands?

The child has to.

One, represent the false belief.

Two, select it as the right thing to base the prediction on.

And three, inhibit their own true knowledge.

Doing all three at once under the pressure of having to give a verbal answer just crashes their system.

So they default to the easiest answer, which is their own knowledge.

Exactly.

It's a resource problem.

Okay, so that's the dual system idea, but what about the single system accounts?

Proponents like Carruthers argue that we don't need two different systems.

We have one core mind -reading mechanism, and the performance gap is all down to how much demand is placed on our general purpose executive functions.

So if it's the same system,

why the huge difference in results between the tests?

Carruthers breaks it down by task complexity.

He says a spontaneous task, like looking, is a single mind -reading task.

The child just has to inhibit their own belief and select the agent's false belief.

It's hard, but manageable.

But the maxi -test is much harder.

He calls it a triple mind -reading task, because you have the first task inhibition and selection, plus you have to process the experimenter's question, plus you have to formulate a verbal answer.

It's the cumulative load.

It's the cumulative load, especially the language production part.

It's so demanding that the system just defaults to the path of least resistance, which is the child's own true belief.

It's an executive function breakdown, not a mind -reading failure.

Okay, that's a powerful explanation.

Now let's move to the non -mentalistic views, the ones that align more with 4E.

The first of these are the behavior rule accounts.

The idea here is that infants aren't tracking mental states at all.

They're tracking observable behavioral principles.

Give me an example of a rule.

A simple rule could be, people tend to look for an object where they last saw it.

Or maybe, being ignorant of something leads to making mistakes.

The infant just learns these statistical regularities about how situations and behaviors are lumped.

So they don't need to posit a hidden belief in the middle.

Exactly.

It seems much simpler.

But there's a big philosophical problem with this view.

Which is?

It's almost too flexible.

Critics like Pavinelli and Vonk argued that you can always just invent a new, more complicated behavior rule to explain any piece of data.

It becomes impossible to design an experiment that could ever prove it wrong.

So it's hard to test against the mentalistic theories.

Very difficult.

Which brings us to the most radical 4E view.

The interactionist accounts.

And these really minimize the role of anything happening inside the head.

Almost entirely.

People like Gallagher argue that children understand actions in terms of social affordances.

What opportunities for interaction and action provides.

So in that active helping test, how does that work without beliefs?

The infant sees the agent struggling with the empty box.

That frustrated action just signals that the interaction is blocked, that the social affordance is wrong.

The infant doesn't need to represent a hidden belief.

They just perceive the need to change the situation to make the interaction work.

So they go get the toy.

The information is in the interaction itself.

That's the core idea.

And it explains the duplo test's success perfectly.

The supportive interactive setup made the social affordances clear so the child could succeed.

Okay, one last account to cover, and it's the most skeptical one.

The low -level novelty accounts.

Researchers like Hayes argue that the looking time results have nothing to do with social understanding at all.

It's just basic domain general stuff like perception and memory.

The babies are just tracking low -level novelty.

So in looking tasks, the baby just remembers where the hand appeared last time and expects it to appear there again.

It's a simple memory trick, not FBU.

That's the argument.

And this critique highlights the huge problem we have in trying to adjudicate between all these accounts.

How do you decide?

They all seem to explain some part of the data.

It feels like we're kind of stuck.

We are.

The debate gets locked into these philosophical arguments about parsimony and what counts as evidence.

And this is why the author says we need to step back and change the entire frame of reference.

Okay, so let's do that.

Let's unpack this.

If all these traditional interpretations are kind of stuck in a stalemate, what happens when we look at the data through a completely different lens?

We shift our framework to predictive processing, or PP.

PP is this huge idea that's really shaking up neuroscience and psychology.

So just to start, how does PP see the brain's main job?

At its core, the predictive processing framework sees the brain as an active prediction machine.

Its one single constant goal is to build models of the world to predict the sensory input it's about to receive.

And its main job is to minimize prediction error.

And prediction error is that feeling of surprise, right?

The gap between what you expected and what you actually got.

Exactly.

The brain is basically a hierarchical guessing machine.

And its guesses, or models, are judged not just on the current evidence, but also on their prior probability, how likely they were based on all your past experience.

And these models are organized in a hierarchy.

Right.

Based on time scale, low levels are for fast changing things like edges and colors.

High levels are for slow changing stable things, like the fact that this is a table.

Or, crucially for us, that an agent has a stable preference for a certain toy.

Okay, and this is where it gets really, really interesting.

Because PP unites perception and action.

There are two ways the brain can fix that prediction error.

This is the absolute core of its explanatory power.

When there's a mismatch, a surprise, the brain has two options.

Option one.

Perceptual inference.

This is what we normally think of as perception.

You change your internal model to better match the world.

I thought the banana was in the bowl, and look, it's not there.

I update my model.

Okay, no banana.

The model changes.

And option two is the action pathway.

Active inference.

Instead of changing your model, you change the world, or your relationship to it, so that the world matches your model.

Okay, using the banana example.

My high level model is strong.

The banana should be in the bowl.

Look, I don't see it.

Error signal.

What do we do?

You perform an action.

You reach out.

You move the apple that's blocking your view.

You are actively changing the sensory input you receive until it matches your prediction.

When you see the banana, the error is gone.

You didn't change your belief.

You changed the world to fit your belief.

Okay, that's a powerful idea.

So now let's apply this new framework to the FBU data.

How does PP see those spontaneous infant tests?

Well, the looking time tests, the VOE and AL ones, they just show that the infants have formed a stable high level model about the agent's goal.

When the agent's action violates that model, you get a big prediction error and they look longer.

Simple as that.

But the active helping tests are much more revealing under this model.

Oh, much more.

Because the active helping tests are a perfect example of active inference in the social world.

The infant has a strong model.

This person wants that toy.

When the person starts fumbling with the empty box, that creates a huge prediction error.

And to fix that error, the infant uses active inference.

They don't update their model.

They change the world.

They actively intervene.

They go get the toy from the correct box and give it to the person.

They perform an action that makes reality conform to their high level prediction about the agent's goal.

They minimize the social prediction error.

And this gives us a totally new way to explain that big developmental paradox.

Why infants pass and three -year -olds fail.

This is the synthesis.

The PP explanation isn't just that the maxi test is more complex.

It gives a deep mechanistic reason for why it's so hard.

So let's break it down.

Why do three -year -olds fail the classic test, according to PP?

In all the spontaneous tests, active inference is an available option.

The child can look, they can help, they can interact.

And that's the brain's preferred lower cost way to reduce error.

But in the classic maxi test, active inference is explicitly blocked.

The kid just has to sit there and give a verbal answer.

Right.

So they're forced to use the only other tool they have.

Perceptual inference.

They have to revise their model.

Then what goes wrong?

A young child's brain, with its powerful drive to minimize error and weaker inhibitory control, takes the most efficient route.

It tries to update the agent's model to match reality.

They know the chocolate is in the drawer, so their brain's default is to assume that maxi's belief model should also reflect that reality.

That would minimize the overall error in the system.

But updating the agent's belief to match reality is the exact definition of the false belief error.

It is.

The failure is a direct result of the child's inability to inhibit this fundamental error minimizing tendency.

They can't honor the constraint that maxi was absent and couldn't have seen the change.

Older kids with better executive function can inhibit that impulse and maintain two conflicting models, which leads them to the right answer.

That is a really elegant explanation.

It's not a social failure.

It's the misapplication of a fundamental cognitive process.

Exactly.

It's rooted in the very architecture of how the brain works.

Okay, so last big question.

How does this PP framework fit with 4E cognition?

Where do they line up and where do they clash?

This is a really tricky area.

There's a lot of synergy on the one hand, but a deep philosophical conflict on the other.

Where's the common ground?

PP fits beautifully with embeddedness and situatedness.

The whole system runs on context and prior experience.

And the central role of active inference is a huge point of overlap.

It validates the 4E focus on action and interaction.

The key distinction might not be second person versus third person, but rather active inference versus perceptual inference.

But there's a big point of conflict around embodiment and representation.

This is the sticking point.

It's the R word, representation.

Radical 4E thinkers reject the idea that the brain contains representations or models of the world at all.

So for them, the fact that PP relies on a hierarchical probabilistic generative model is a deal breaker.

For the most radical proponents, yes.

Because it sounds like we're right back to having a little picture of the world inside the head, which is what they fought so hard to get away from.

So is there any path to reconciliation?

The philosopher Andy Clark has offered one.

He suggests that the representations needed for PP aren't the old fashioned static picture -like ones.

He calls them action -oriented.

Meaning their main job isn't to be a map of the world, but to guide action in the world.

Exactly.

Their purpose is to enable the dynamic coupling between the agent and the environment.

He even suggests they might not have content in the traditional sense.

But even this is a tough sell for the most radical 4E folks.

So the debate continues.

But it seems like PP offers this incredibly powerful new tool for understanding social cognition.

It does.

It provides a unified mechanism error minimization that can explain the situated interactive social understanding that 4E champions.

And it finally gives us a plausible solution to that decades -old puzzle of false belief understanding.

So if we just pull all of this together,

the story of this deep dive has been a real scientific journey.

We started with this established theory, mean dreading, and its key evidence, the age form milestone.

Then we saw that theory get completely turned on its head by new evidence from spontaneous tests, showing competence in infants.

Which led to this messy, complicated debate with all these different accounts trying to explain the data.

But the real insight from the chapter comes from shifting the entire frame to predictive processing.

By focusing on the brain's basic drive to minimize prediction error, PP solves the paradox.

It explains the 3 -year -old's failure not as a lack of social ability, but as a systematic error.

The test blocked their natural path for fixing error's active inference and forced them onto a path, perceptual inference, that led them to make a mistake by updating the agent's belief to match their own.

It's a much deeper, more mechanistic explanation.

So what's the final takeaway for you, the listener?

Well, the big insight is that our very human tendency to reduce cognitive dissonance, to want the world and other people to match what we expect might be the source of a lot of our everyday social mistakes.

Think about that difference between active and perceptual inference.

When you're in a real interactive situation, you can often solve a social mix -up just by acting, by clarifying, by helping.

That's active inference.

But when you're forced to just sit back and guess what someone else is thinking, or why they did something strange, say you're trying to figure out why a colleague made a weird decision in a meeting.

You're stuck using perceptual inference.

And when you do that, your brain is under constant pressure to minimize error.

And you might just be failing to stop yourself from updating their belief model to match your knowledge.

You assume they must have seen what you saw or that they know what you know, because that's the simplest, most error -free model for your brain to handle.

That failure to apply the constraint that maybe they didn't know is a source of constant misunderstanding.

It suggests that sometimes being good at understanding others isn't about having bitter theories, but just having better inhibition.

Something to think about.

Food for thought indeed.

Thank you so much for joining us on this deep dive into how we understand each other.

And thank you for sharing your source material with us.

From the entire team, thank you for listening, and we'll catch you next time for the next deep dive.