Chapter 44: Scaffolding Intuitive Rationality

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 Dive.

If your mission is to gain knowledge quickly but thoroughly, to shortcut your way to being truly well -informed, you are absolutely in the right place.

Today we're doing a deep dive into a chapter by Cameron Buckner called Scaffolding Intuitive Rationality.

And this one is, well it's a big one.

We're tackling a concept that's right at the heart of intelligence, but it's also maybe the most difficult to pin down.

We're talking about intuitive rationality.

Exactly.

And the core idea here is pretty challenging to the old ways of thinking.

Buckner's argument is that our intuitive judgment, you know that fast non -linguistic gut feeling kind of thinking, is actually rational.

But there's a catch.

There's a big catch.

You can't understand it by trying to sort of force it into the box of human logic.

Instead, we need a totally different approach, a buildup model that starts simple and adds complexity.

Okay, let's unpack that.

I think the best way to start is by mapping out the cognitive terrain because the whole problem here is that intuitive judgment lives in this huge confusing middle ground.

That's a decent handle on the two extremes of behavior.

Right.

So up at the top, you have what the chapter calls classical cognition.

Right.

This is the stuff we usually think of as thinking.

It's explicit.

It's effortful.

It uses language.

Like doing your taxes or debating philosophy.

Or trigonometry.

Yeah.

It's the kind of thinking you can usually describe step by step.

And we have a clear way to evaluate it.

Which is classical rationality.

Exactly.

Are your beliefs consistent?

Do they follow the laws of probability?

Are you maximizing your utility?

It's all very well defined.

Okay.

So that's the high ground.

Then you've got the low ground.

The low ground is basic behavior.

This is your reflexes, your basic conditioning.

All the stimulus driven stuff we see in non -linguistic animals.

And we evaluate that with biological rationality.

Does it help the organism survive and reproduce?

Is it maximizing fitness in its environment?

It's very direct, very stimulus bound.

So we know the rules for the super logical language based stuff.

And we know the rules for the basic biological survival stuff.

But the vast territory in between.

That's the problem.

That's the murky zone.

And that's where we spend what like 95 % of our cognitive lives.

Pretty much.

And this murky zone is intuitive judgment.

Buckner's definition is that it's the system that allows for really flexible behavior.

But it's implicit.

It's tacit.

It's non -linguistic.

So think about riding a bike or instantly reading someone's facial expression.

Exactly.

Or the classic example of driving a familiar route on autopilot.

You're making incredibly complex, flexible decisions in a dynamic environment.

But if someone asks you how you did it, you can't explain the rules.

You just did it.

It feels like a hunch or a seeming.

It's completely opaque to introspection.

And this isn't just a weird human quirk.

No, this is assumed to be the main form of flexible thinking in animals and pre -verbal infants.

So if we want a real science of cognition, a complete one, we have to figure out how this murky zone works.

And more importantly, how we can possibly call it rational.

Which brings us right to the conflict.

I mean, this is the engine of flexible intelligence for most creatures on earth.

But the rationality wars have been raging over this for decades.

So why is this specific thinking so controversial?

Well, the controversy really comes from two places.

First,

nobody can agree on the underlying model.

I mean, 50 years after Kahneman and Tversky were still arguing.

You've got all these different theories flying around.

You do.

You have these qualitative labels like anchoring or the availability heuristic, which just describe what happens.

Then you have connectionist networks trying to model the neurology.

And the fast and frugal algorithms guys who say it's all about simple rules for specific environments.

So there's this big disagreement about whether our intuitions are these complex opaque network processes or just a collection of simple rules of thumb.

And that disagreement feeds right into the bigger, more philosophical fight.

The fight over rationality itself.

Because when you bring these intuitive processes into the lab, they almost always violate the rules of classical probability.

Which leads one side to say, see, they're irrational.

They're bugs in the system.

Right.

They're heuristics that lead to systematic errors.

But the other side says, hold on.

They're not bugs.

They're features.

They argue that these processes are fundamentally rational because they work.

They're efficient.

They're reliable.

And they succeed in the real world environments where they evolved.

That distinction, what works in a sterile lab versus what works in messy reality, that seems to be everything for this chapter's argument.

It is.

How you study the depends entirely on which standard of success you choose.

And this leads to two completely opposite ways of trying to solve the problem.

Okay, so let's get into those two methodologies.

The dominant one historically is what the chapter calls the revised down strategy.

Right.

And this approach starts at the very top with a peak of human achievement,

formal, explicit logic, and then it tries to work its way down.

So you take adult human logic as the gold standard, and then the question becomes, well, can we find little shadows or analogs of this logic in animals or infants?

Exactly.

And this often uses a really high standard, something called the Neo -Fregion conception.

Basically, it requires thought to be like language with systematic concepts, propositions, and rules that care about truth.

So if you're a philosopher who thinks all rational thought has to be structured like a sentence, you pretty much have to take this revised down approach.

You're stuck with it.

Can we get a concrete example of this?

How does someone actually try to revise down logic for an animal?

A great example is Jose Bermudez's work on protologic.

He knows animals don't have sentences, so he looks for cognitive structures that work like logical parts.

So not a real if -then statement, but maybe a proto -if -then.

A codoconditional, yeah.

An instrumental belief about means and ends.

Or he'll look for something like proto -negation in the relationship between contrary concepts, which might let an animal reason by exclusion.

He's hunting for the ghosts of logic in non -linguistic minds.

But the burden of proof there just seems gigantic.

You have to somehow prove that a creature without language can master these abstract domain general rules that govern classical logic.

And that's the core critique from Buckner.

The revised down strategy just can't bridge that gap.

How does an animal, without the kind of general purpose representations language gives us,

possibly internalize the abstract rules of logic?

You're trying to fit a square peg into a round hole.

You're trying to cram this ecologically -tuned associative machine into a highly -refined, truth -preserving structure that it was never built for.

It's a non -starter.

Okay.

So if starting from the top and working down is flawed, that brings us to the chapter's alternative, the build -up strategy.

And this just flips the whole thing on its head.

Instead of starting with logic, we start with the most basic thing we know all agents have, associative learning.

Simple cause -and -effect learning.

The most ancient, evolutionarily conserved mechanism.

And from there, we ask,

what are the absolute minimal additions we need to scaffold onto this base to get the kind of flexible, intelligent behavior we see in the real world?

This feels immediately more in line with the whole 4E cognition movement embodied, embedded.

Absolutely.

Yeah.

Because it starts with the organism in its environment, learning and adapting.

It doesn't start with an abstract set of rules floating in a vacuum.

The big advantage here seems to be that it forces you to acknowledge how much our intelligence depends on the world around us.

That's it.

You have to account for the ecological scaffolding and the developmental shaping that guides learning.

You avoid this massive philosophical headache of explaining where logic came from, and instead you focus on how the environment structures the learning process to produce smart outcomes.

So our mission, then, is to see how this buildup actually works.

And to call this bottom -up intelligence rational, we have to first clear a pretty basic philosophical bar for what even counts as an inference.

Right.

The classical definition of inference, the high -altitude one, is a rule -based transition between propositions.

That's obviously not going to work for a non -linguistic animal.

So we need something more fundamental.

We need to satisfy what's called the taking condition.

It's a really foundational idea in philosophy of mind about why we say an action was rational.

So what is the taking condition?

In its strong form, it says that to infer something, the thinker has to actually take their premises to support their conclusion.

And they have to act because of that perceived support.

So the agent has to be aware, on some level, of the connection between their reason and their action.

Right.

You can't just say, well, here's a good reason they could have used.

You have to show the agent was actually guided by the content of their internal representations.

And this is the big problem for intuition, right?

If it's just a hunch and the process is totally opaque to me, how can I be taking a premise to support a conclusion?

It feels automatic.

That is the whole problem.

If taking means you have to consciously apply a formal rule, then intuition is dead on arrival.

It's irrational by definition.

So Buckner proposes a revision.

A revision, exactly.

The condition still has to be met.

The agent has to act because of the content of their thoughts, but not through formality, not by following rules.

The key is to meet the condition through intentionality.

Okay.

This is a big one.

We need to pause here and make sure this concept is crystal clear.

Let's break down the difference between intention with an S and extension with an X.

Good idea.

So the extension of a concept is just the set of all the things in the world that the concept refers to.

For example?

The extension of creature with a heart is every single animal on earth that actually has a heart.

It's just the set of objects.

Okay.

So what's the intention?

The intention is the meaning.

It's the semantic content, the specific features you use to define the concept.

It's the mode of presentation.

And the classic philosophy example here is?

Creature with a heart versus creature with a kidney.

Now, in the real world, these two concepts have the exact same extension.

They refer to the very same group of animals.

But their meaning, their intention is totally different.

One is about a circulatory organ.

The other is about a filtering organ.

Exactly.

They are extensionally identical but intentionally distinct.

And for intuitive judgment to be rational, the inference has to be causally sensitive to the intentions, the features, the meaning, not just the set of things in the world.

Okay.

So we've lowered the bar.

We're not saying the agent must follow a rule about all creatures with hearts.

We're saying the agent must be internally guided by the specific features they associate with creature with a heart.

Yes.

And this is why similarity models are a perfect fit.

A judgment based on similarity, like Trisky's feature matching model, is by its very nature causally sensitive to the structure and features, the intentions of the things the agent is comparing.

This is where that simple restaurant analogy really helps make this concrete.

Let's use it.

So you're on vacation, walking down a street, trying to pick a restaurant.

You're not running some complex algorithm.

No, you're just getting a vibe, a gut feeling.

Right.

You're relying on your internal prototype of a good choice.

And that prototype, that's your intention.

It's made up of features.

It has local people eating there.

The tables are clean.

It smells good.

Maybe the menu is handwritten.

So if my decision is driven by how similar a place is to that internal prototype, then your choice is being causally governed by those specific features.

If a restaurant had different features, loud music, only tourists, fluorescent lights, you'd make a different choice.

Your action is rational because it's sensitive to your internal reasons, your intentions.

Even if I can't consciously list them all out at the moment.

Even then, it satisfies the revised weaker taking condition.

Okay.

So this gives us the first half of the puzzle.

Our intuitive judgments are rational in this internal psychological sense.

And that leads to the chapter's hybrid rationality conclusion.

Right.

Because now we have to connect it back to the outside world.

This creates two distinct ways to talk about the rationality of intuition.

The first is explanatory rationality.

Which is an internalist view.

It answers the question, why did the agent do what they did?

And the answer is because they judged the option to be sufficiently similar to their desired category based on their internal set of features.

It's a satisfying psychological explanation for the behavior.

And then there's the second part, justificatory rationality, an externalist view.

This asks a different question.

Was the action good?

Was it justified?

Was it the right move?

And the answer there is that the action is justified when it fits the external environment.

When it's ecologically rational.

And this is where the theory takes a very sharp turn away from classical logic.

Yeah.

Let's really dig into this because the concept of overfitting is so important here.

Why does the classical idea of justification fail so badly when it comes to intuition?

Well, in classical decision theory, the ideal is usually to use all the evidence you have to make the most optimal decision.

You build the most detailed model possible.

Which sounds good in theory.

It sounds great.

But in noisy, unpredictable, high -variance environments, which is just a description of real life, that approach is a disaster.

Why?

What happens?

It backfires because you end up treating random noise as if it's a meaningful signal.

You build this incredibly complex model that perfectly fits the random blips and quirks of the past data.

That's overfitting.

So when a new situation comes along, your perfectly -tuned, complex model just falls apart.

It fails spectacularly.

Yeah.

Because it was built on a foundation of irrelevant details.

It's like if you tried to decide on your tennis serve by factoring in the opponent's shoe color and the exact time on the clock, you'll do worse than if you just focus on the two or three cues that actually matter.

So trying to be perfectly logical and account for everything can actually make you less successful.

Way less successful.

Yeah.

And this is the source of that great analogy in the chapter.

Even an optimizing demon with infinite computing power can be reliably beaten in a noisy world by an appropriately lazy dog with the right epistemic bet.

And that lazy dog is using intuition, a simple frugal heuristic.

Exactly.

So the norms for justifying intuition have to be external and ecological.

We're not looking for truth preservation.

We're looking for ecological validity.

Does this strategy have a good fit with the structure of the information out there in the environment?

That's the question.

Rationality becomes a successful long -run gamble achieved by focusing on a few valid cues and ignoring all the noise.

So we have our hybrid model.

Intuitive judgments are rational on the inside because they rely on intentional similarity.

And they're rational on the outside because they're ecologically valid.

So now, how do we get from a simple similarity match to something as complex as, say, a raven outsmarting a competitor?

Right.

This is the heart of the buildup strategy.

We start with that basic similarity, but as you said on its own, it's pretty weak.

A simple feature matcher would say a penguin isn't a bird.

Because it doesn't fly, it swims, it's black and white, it doesn't match the prototype.

So we need mechanisms to look past those surface features and reshape the similarity space itself.

We need the first piece of scaffolding.

Scaffold one.

Abstraction via resistance to nuisance variables.

The whole game here is moving the agent from relying on stimulus -bound features, like a specific color, to matching more abstract higher -order patterns.

The pigeon experiments are a perfect illustration of this.

Let's walk through those levels of the pigeon hierarchy.

Okay, level one is super simple.

You train a pigeon to only peck at one very specific shade of red.

Its response is incredibly stimulus -bound.

If you change the hue even a little bit, its performance just collapses.

Very little flexibility.

Now, level two.

At level two, you train the pigeon to recognize a triangle.

Now the triangle could be red, blue, green, big, small, rotated.

The pigeon has to ignore all that surface level stuff.

It has to learn to ignore color, scale, and rotation to get at the underlying structural principle.

Three connected lines forming a shape.

That's a huge leap in flexibility.

And then level three is even more complex.

Much more.

At level three, you train the pigeon to do something really configurable, like telling the difference between an impressionist painting and a cubist one.

That requires picking up on very subtle high -order statistical patterns in the input.

This gives us a nice gradient of abstraction, but the chapter gives us a more rigorous technical definition of what abstraction actually is, borrowing from AI.

It does.

And this is a key move.

It links this idea to why modern AI, especially deep learning convolutional networks or condonets, are so successful.

Technically, abstraction is defined as a representational format that is more resistant to nuisance variation.

Okay.

So nuisance variable, what is that exactly?

A nuisance variable is any feature of the sensory input that changes a lot but gives you zero useful information for the task at hand.

Like envision the position of an object in the frame or its rotation or the lighting.

Exactly.

If you're trying to identify a cat, it doesn't matter if the cat is on the left or the right or if it's slightly tilted.

Those are nuisance variables.

An embodied agent moving through a 3D world is constantly being bombarded by this nuisance variation.

The angle, the distance, the lighting, it's always changing.

Right.

So a huge part of what the cognitive system has to do is filter all that out.

It has to transform the raw sensory data into a higher -order format that is insensitive to those nuisance variables so it can find the deeper taxonomic principles that stay the same.

So that filtering process, that's our first layer of scaffolding.

That's the first step up from basic similarity matching.

So once we have the architecture to see all triangles as triangles, regardless of their color or size, we need the next scaffold to help us figure out which of these deep patterns actually matter.

Scaffold 2.

Predictive learning and reshaping similarity space.

This is where we move beyond just recognizing static patterns and start grouping things together based on whether they predict the same future outcomes.

So things that look totally different on the surface.

But lead to the same good or bad result?

Should start to be treated as if they're similar on a deeper, more abstract level.

This feels like the mechanism that allows for genuine insight.

It's all about prediction error.

The chapter points to models like the Gluck and Myers model of the hippocampus.

The system isn't just learning from the input.

It's learning from the mismatch between what it predicted would happen and what actually happened.

So if two very different situations consistently predict the same reward.

The learning mechanism physically reshapes the internal similarity space.

It compresses those two disparate points, moving them closer together in this new transformed feature space.

It creates what's called an abstract equivalence class.

They become equivalent because they predict the same thing.

Precisely.

And this is the mechanism that gives us a really plausible explanation for the amazing Raven theory of mind experiment.

Okay, we absolutely have to detail this one because it's such a clear example of non -linguistic insight.

What was the experiment set up?

So Ravens are master cashers.

They hide food for later and they've evolved these complex strategies to protect their caches from competitors who might be watching.

The experimenters created a chamber where they could manipulate whether the Raven thought it was being watched.

And there were three main conditions.

Right.

First, the observed condition.

The Raven is hiding food and another Raven is clearly visible watching through a window.

End result.

The Raven is super careful.

It guards its cache.

It re -hides it.

Classic defensive behavior.

Okay, condition two.

The non -observed condition.

The Raven is caching completely alone and predictably it's much more relaxed.

No guarding.

And now for the critical one, the people condition.

This is the genius part.

The Raven is caching alone again.

The window is closed so it can't see a competitor but there's an open people in the wall and they're playing the sounds of another Raven.

And the crucial piece of background information here is that these specific Ravens had been previously trained to use a people themselves.

To pilfer.

To steal food that an experimenter was hiding.

So they knew what a people was for.

So in the test, the Raven can't see another bird but it knows the people is open and it can hear one.

What does it do?

It behaves exactly as if it was being fully observed.

It aggressively guards its cache.

That's incredible.

So a revised down theorist might try to explain this by saying the Raven is running a logical syllogism in its head.

If the people is open, someone can see.

If they can see, they can steal.

Therefore, I must guard my cache.

The very clunky language -like explanation.

So what's the much more elegant buildup explanation?

The Raven used predictive learning.

It had mapped its own successful pilfering experiences into an abstract equivalence class.

For the Raven, seeing a competitor in the open and seeing a competitor through a people were two perceptually different situations.

That predicted the same outcome.

The same outcome.

Yeah.

You get the food.

So it created this abstract class and then when it was the one caching, it applied that abstraction in reverse.

The people plus sound cue triggered the prediction of an opportunity for others.

So it's not reasoning about seeing as a concept.

It's generalizing the structural opportunity for theft regardless of who is in the role of the thief and who is in the role of the hider.

That's it exactly.

It's an amazing example of insight that looks like rule -based logic but can be explained by associative learning over these abstract, predictively relevant features.

Which takes us to the theoretical limit of this kind of abstraction.

Scaffold three.

The Tintimoresque slot.

Right.

This is Buckner's hypothesis for what happens when you optimize this similarity space so much that it starts to functionally resemble a free gen logic function.

A Tintimoresque refers to those cardboard cutouts at a fair where you stick your head through a hole for a photo.

Ah, so it's a placeholder.

A head and a hole cutout.

It's an unsaturated function, exactly.

The idea is that this highly abstract feature space creates a representational slot.

In the Raven example, the equivalence class has a slot for actor A, the one watching, and actor B, the one being watched.

The system doesn't care about the specific features of who is in that slot.

It doesn't need to know who or what A or B is.

The only thing that matters is that if an individual fits into that functional role, a certain consequence is predicted.

So the Raven system has abstracted the situation to something like if placeholder X can perceive placeholder Y's cache, then placeholder Y should engage in guarding behavior.

Yes, and this kind of placeholder abstraction allows an animal to generalize a behavior it learned in one role like being the pilfer to others who might occupy that same role later.

It's a powerful approximation of true logical abstraction.

But even with these three scaffolds, there's still one major hurdle left, mental time travel, planning for the future.

Right.

If these systems are ultimately driven by current stimuli,

how can an agent plan for a goal that isn't present right now?

Humans do this all the time with language.

We use internal speech, calendars, written notes, all this symbolic scaffolding to represent future goals.

How can an animal do it?

The chapter suggests they might achieve a limited form of it by chaining these abstract associations.

The classic scrub J caching experiments are the key example here.

Right, the J's that seem to plan for future hunger.

Exactly.

A scrub J that is currently full will still hide a certain kind of food in a specific chamber where, in the past, it learned it would be hungry for that food the next day.

So if we don't think it's telling itself, I'll be hungry tomorrow, I should save this, how does the buildup model explain this?

It's a chain of predictive inference.

The Q is the environment itself, that specific chamber.

That Q triggers a simulation of a future state, specifically the future absence of food in that location.

Ah, so the simulated absence of food then acts as the immediate trigger for the caching behavior.

It becomes the associative trigger for the desire to cache.

So the system is still chaining Qs to predictions to actions, but it allows for a kind of transcendence over the immediate here and now.

It's more fragile than our language -based planning, for sure, but it's a real limited form of foresight.

And with that, we've successfully built up from simple similarity judgments to complex, insightful, and even forward -looking behavior, all without needing to import classical logic.

Which brings us to the final section, which is all about defending this model from the philosophical pressure to conform to those old logical standards.

Right, this is where we need to address the constant pushback from the revised -down camp.

And the chapter gives us the philosophical tools to resist what Buckner calls anthropofabulation.

I love that word.

Let's define it.

What is anthropofabulation?

It's this constant pressure on researchers to define cognitive abilities based on an artificially inflated, idealized version of human logical capability.

It's a mix of anthropocentrism, human -centeredness, and confabulation.

And it shows up in these endless debates.

These endless debates about whether an animal is really reasoning by exclusion, or just as -if reasoning.

Or if it's really optimizing utility, or just as -if optimizing.

It's this skeptical challenge that's always holding psychology to a standard that's not even appropriate for describing human behavior, let alone animal behavior.

Exactly.

It's forcing a descriptive science to conform to a normative, idealized standard.

And the solution is to go back to a very old, but often forgotten, idea.

We have to reestablish the Friedgen boundary.

Let's go back to the source.

Gottlob Friedge, the father of modern logic.

What did he actually say about the relationship between logic and how the mind works?

He was incredibly clear about this.

Friedge insisted that logic is not the science of how minds actually reason.

Logic is a normative science.

Its whole point is to discover the laws of truth.

It tells us how we ought to think if our goal is to preserve truth.

Psychology, on the other hand, is descriptive.

It has to explain how we actually think, including all our biases or mistakes.

In Friedge's words, error and superstition.

He was adamant that confusing the two was a massive mistake.

But critics often misuse a specific quote from Friedge about inference to attack models like this one.

They do.

The quote is something like, to make a judgment because we are cognizant of other truths as providing a justification for it is known as inferring.

And they say, see, intuition doesn't involve being cognizant of truths.

So what's the mistake in that reading?

The subtle but critical mistake is that Friedge is building a success grammar into his definition.

By saying cognizant of other truths, he's conflating the general act of inference with sound inference.

Inference that proceeds only from true premises.

Which would make the study of unsound inference, which is most of what psychology studies, conceptually impossible.

It writes it out of existence.

That's why we needed that weaker, intentional version of the taking condition earlier.

It gives us a minimal definition of inference that allows for mistakes.

OK, so once we reject that idealized standard, the anthropofabulation challenge kind of melts away.

But there's still one question left.

We know that humans can, sometimes, grasp logical entailment.

So doesn't our psychological theory need to explain that peak performance?

It does, but we have to be very careful about the context.

Humans only grasp logical entailment in our most highly scaffolded and deliberate activities.

Like when we're doing formal science or math or collaborative philosophy?

Right, and this is the crucial point.

The laws of logic were not discovered by studying how people think in their day -to -day lives.

So where did they come from?

They were induced and systematized by observing the laudable inferences.

The best of the best within the theoretical sciences.

You can think of them as the diamonds that emerge after weeks or months of slow, painful, linguistically, and socially scaffolded work.

So writing things down, using external tools, arguing with colleagues, revising, and forcing coherence?

All of that.

The laws of logic are a product of that highly structured technological process.

They are not a description of the raw processing of the mind.

That totally reframed what logic is.

It's not the brain's operating system.

It's a high -tech piece of software for verifying a very specific kind of output.

That's the perfect analogy.

Classical logic and decision theory are scientific technologies.

They're intellectual tools we invented to evaluate our most idealized reasoning.

They have the same relationship to everyday cognition that quantum mechanics has to our everyday physical intuition.

You don't use quantum mechanics to catch a baseball.

And you don't use predicate logic to decide which checkout line to join at the grocery store.

So this leads to what the chapter calls the vanishingly weak constraint on any psychological theory.

Since we know humans can produce these logical end products, our theory of how the mind works only needs to allow for the possibility that this effortful scaffolded process can sometimes result in linguistic artifacts, papers, proofs, theorems that satisfy logical norms.

That's all it has to do.

That's it.

It's a huge weight off the shoulders of comparative psychologists.

It frees them up to study the murky zone, the similarity, the abstraction, the predictive learning on its own terms, without having to apologize for the fact that it doesn't look like an idealized logical proof.

That old skeptical challenge, the demand that everything live up to this inflated Phrygian standard, we can just reject it.

We can, and we should.

The buildup model gives us a way to see intuitive judgment as robustly rational in its own right, respecting its ecological and developmental origins, and freeing it from the tyranny of an inappropriate logical standard.

What a journey.

So we've done a really massive deep dive today into the nature of intuitive rationality.

We started by mapping out that murky zone of intelligence, setting it apart from both classical logic and simple reflexes.

And the really crucial move from Buckner's chapter was defining the rationality of that murky zone with a hybrid model.

It's rational on the inside because it's explanatory.

It's driven by intentional similarity.

And it's rational on the outside because it's justified.

It's ecologically valid and successful in the messy real world.

Then we walked through the buildup approach, seeing how you can scaffold complex cognition on top of basic associative learning.

We saw how abstraction helps agents filter out nuisance variables to see deeper patterns, like the pigeons learning to see all triangles as triangles.

We looked at predictive learning, which reshapes that similarity space based on what outcomes are predicted and how that explains the incredible insight of the ravens who generalize their own experience as a thief to predict the threat from a competitor.

And finally, we put up a philosophical wall.

We reestablish that phreagin boundary between logic as a normative standard and psychology as a descriptive science.

We recognize that classical logic isn't a model of our minds, but a highly specialized intellectual technology, a tool we use to check our best, most careful work.

So what does all this really mean for you, the listener, trying to navigate the high -speed chaos of daily life?

You are living and operating almost entirely in that intuitive murky zone.

We've just established that classical logic is a technology built by observing our most pristine, slow, collaborative, painstaking reasoning.

It is absolutely not our default mode of thought.

So given the lodging is a spiralized tool we use to evaluate our most idealized output rather than a description of how our brains actually work.

Here is the final provocative thought for you to take away.

If your intuitive, ecologically rational mind is specifically designed to be fast, frugal, and successful in noisy, uncertain environments, how much energy should you really be spending trying to force your daily high -speed decisions to conform to the slow, effortful, and often over -fitted standards of classical logic?

Perhaps trusting the power and the necessary imperfection of your own scaffolded intuition is the most rational strategy of all.

Something to think about.

Thank you for joining us for this deep dive.

We'll see you next time.