Chapter 12: Reasoning: Deductive, Inductive & Everyday Logic

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, the place where we take a mountain of source material and distill it into the essential knowledge you need to be well -informed.

And today we are undertaking a deep exploration into one of the most fundamental cognitive processes our brains perform,

reasoning.

Right.

And at its simplest, reasoning is the process that transforms a set of given information, what psychologists will call premises,

into meaningful conclusions.

It's the engine of deliberate thought.

It's the process you rely on constantly, often, you know, without even realizing it.

Give us a quick example of that.

Well, think about a classic everyday scenario.

You're waiting for a friend who is always, always on time, and they're late.

Okay.

You know they're driving and you know it's rush hour, so you combine those premises.

She is usually on time, she hasn't called, and she is driving in peak traffic.

What do you conclude?

That she's stuck in traffic.

Exactly.

You infer that.

That entire transformation,

that structured movement from these established facts to a likely inference, that's reasoning in action.

It's how we predict, problem solve, and navigate, well, pretty much everything.

You know, I often hear people use reasoning and thinking interchangeably, but cognitive psychology makes a pretty important distinction here, doesn't it?

It does, and it's a helpful one.

Thinking is sort of the broad category.

It includes daydreaming, imagining, just letting your mind wander, but reasoning is usually seen as a more specific, more focused,

and goal -directed kind of thinking.

Like solving a puzzle or a mystery.

Precisely.

It often involves applying principles of logic or carefully weighing evidence to get to an answer.

And you mentioned that a lot of this happens without our awareness.

Oh, absolutely.

The sheer speed of human inference is just astounding.

Take language, for example.

If you read a headline that says, an athlete managed to break the high jump record, your brain just proffervate.

Yeah.

It automatically infers that it was difficult, that it required a lot of effort.

Because of that one word, managed.

Yes.

That word carries a built -in premise, and your mind processes the conclusion of high effort instantly.

You never have to stop and consciously think, okay, rule of linguistics, managed implies difficulty.

It just happens.

That structural complexity sort of humming along in the background is what makes this topic so critical to understand.

So for our deep dive today, what's our mission?

Our mission is to systematically unpack human inference based on the source material from Chapter 12 on reasoning.

We're going to start with the bedrock, the formal definition separating deduction and induction.

Okay.

Then we'll move through the common, pervasive, and highly predictable errors we all make, and we'll analyze why they happen.

And finally...

We'll dedicate some serious time to examining the competing cognitive theories, confidential rules and mental models, that try to explain the actual mental machinery our brains use to manage all of this.

Okay.

Let's get into it.

Let's unpack that foundational split first.

Deductive reasoning versus inductive reasoning.

These two sound similar, but they're defined by fundamentally different principles.

That's right.

And we can really distinguish them based on three critical conceptual differences.

The first one is scope.

Go.

Deductive reasoning always moves from the general to the specific.

It starts with a broad rule or a general truth, and then applies it to a narrow particular instance.

Can you give us the classic example there?

Sure.

If we start with the premises,

all engineering students are mathematically proficient,

and Eliza is an engineering student.

Then the deductive conclusion has to be Eliza is mathematically proficient.

It must be.

We started with the general class, all engineering students, and we moved down to a specific member of that class, Eliza.

So conversely, inductive reasoning reverses that flow.

It moves from the specific to the general.

Precisely.

If I observe that Dr.

Jones, a professor, drives a hybrid car, and then Dr.

Smith, also a professor, drives a hybrid car, I might induce a generalization.

Something like all professors drive hybrid cars.

Exactly.

I took a few specific instances, and I used them to form this big, sweeping general conclusion about a whole class of people.

Which leads us right to the second difference.

Information content.

In the Eliza example, the conclusion that she's proficient, well, that was already contained or implicit in that first premise.

Right.

The process of deduction adds no new knowledge to the system.

It just makes the hidden information explicit.

But the inductive process, that's different.

By definition, induction expands knowledge.

When I generalized from two professors to all professors, I've just generated a massive amount of new information.

A hypothesis that goes way beyond the two specific cases I actually saw.

Induction is how we form rules, categories, and make predictions about the future.

And the third difference is probably the most crucial one for understanding how reliable the conclusion is.

Conclusion certainty.

With deduction, you get deductive validity.

And deductive validity is really the gold standard.

An argument is deductively valid if, and only if, assuming the premises are true and the logic is sound, the conclusion cannot possibly be false.

The truth is guaranteed.

So if the rule all A or B is true and C is A is true.

Then C is B.

Must be true.

No exceptions.

But that level of certainty is pretty rare in the real world, right?

Which is why most of our day -to -day reasoning is probably more in the inductive camp.

Exactly.

Inductive reasoning gives you what's called inductive strength.

This means the conclusion is highly probable, or strongly supported by the evidence.

But it is not guaranteed.

It's improbable, but not impossible, for the conclusion to be wrong.

That's the key.

Even if every single professor I ever meet drives a hybrid, it remains possible that the very next one I meet drives a monster truck.

Inductive conclusions are always provisional.

They're always subject to being proven false by new evidence.

They can be strong or weak, but never necessarily true.

Alright, let's delve into that world of guaranteed truth then.

We'll start with the most formal structure.

Propositional reasoning.

This is all about drawing conclusions from propositions.

Assertions that are strictly binary, either true or false.

We just abbreviate them as DePaul and dollars.

And to link these simple assertions into more complex arguments, we rely on what are called logical connectives.

There's Daybird, V, Nag, and Right Arrow.

The really essential feature of these connectives is something called truth functionality.

Meaning the truth or falsity of the whole statement depends only on the truth or falsity of its little component parts.

Yes.

And this is where we run into problems, because this rigid logical structure often violates

the social and conversational rules of English.

Okay, let's start with the easy ones.

The conjunction, which is the unger symbol, pretty much mirrors our word and.

Logically P and Q is the same as Q and P -O.

It's logically equivalent, yes.

But in conversation, sequence is implied.

If I say, she finished her thesis and she celebrated, that sounds totally normal.

But if you reverse it, she celebrated and she finished her thesis, it sounds like she celebrated way too early.

Exactly.

It implies a weird temporal distortion.

But to a logician, the truth value is identical.

They are truth functionally the same, regardless of any implied sequence or causality.

Okay, what about the disjunction, the V symbol or or?

This one is maybe more troublesome.

We tend to use or in English almost exclusively in the exclusive sense.

Like you can take the train or the bus, which usually means you can't do both.

Right.

But in formal logic, the or is inclusive.

P -B -Q is true, if B -Kill is true, if B -Kill is true, or if both are true.

The only way for the whole thing to be false is if both parts are false.

So if I tell you I will travel by train or by bus and you find out I took both, well, in logic, my statement is still true, even if it feels a little odd in conversation.

And then we get to the one that gives human reasoners the most headaches, the material implication, the arrow or the if, then connective.

This one is tricky because it's defined in a completely abstract way and it's divorced from any necessary cause and effect relationship, which is, you know, it's a primary function in everyday speech.

That lack of causality is key, you're saying?

It is.

Logically, the E -P -E -R -R -Q is defined as being false in only one situation.

When the antecedent, the if clause is true and the consequent, the then clause is false.

In all three other possible combinations, it is defined as true.

I think we need to linger on that for a second because it's so counterintuitive.

There's this other way of phrasing it, right, as not be or oh, girl.

Yes.

And that equivalence can help make sense of it.

Let's look at the failure condition.

For P -R -I -T -E -R -O -Q to fail, you need a true biller and a false dollar.

Okay.

Now let's look at not P or or gore.

For that statement to be false, both parts have to be false.

Not P has to be false and thin all has to be false.

And if not P is false, then P must be true.

Exactly.

So the only way for not P or dollar to be false is if P or is true and to or is false.

The failure condition is identical, so the statements are logically equivalent.

That helps, but it's still pretty abstract.

Let's go back to that bizarre example from the source material about the false antecedent.

If the first part, the if part is false, the whole statement is automatically true no matter what the second part says.

Right.

The grandmother example illustrates this perfectly.

If I state,

if my maternal grandmother lived to be 569 years old, the sky is green.

Which is just a nonsense statement.

It is.

But logically, the antecedent my grandmother lived to be 569 is false.

And since the first part is false, the condition for the implication to fail, which requires a true play, is impossible to meet.

Therefore, the entire statement, as bizarre as it is, is logically true.

The lack of an implied connection between the two parts is what trips everyone up.

Because in English, we assume relevance.

So this complexity means we need tools other than just making these huge truth tables which get way too big, way too fast.

They grow exponentially, yeah.

So we rely on valid shortcuts.

The two foundational valid rules are modus ponens and modus tollens.

Modus ponens is the affirming mode, right?

It is.

If play, write, arrow, cue, and we know pity, we can validly conclude tiller.

If it is raining, the street is wet.

It is raining.

Therefore.

The street is wet.

Simple enough.

And modus tollens is the denying mode.

It's equally valid, but it's one that people often miss.

If p, arrow, shuri, or cue, and we know neg cue, the consequence is false, we can validly conclude neg pu, the antecedent, is false.

So if it is raining, the street is wet.

The street is not wet.

Conclusion, it is not raining.

Perfect.

But our brains constantly try to apply shortcuts that aren't valid.

The fallacies.

These are reasoning errors that can give you false conclusions even when your premises are true.

And the two main ones here are affirming the consequent and denying the antecedent.

Let's use an analogy to show why they fail.

Premises.

If a person is running for office, then they must wear a suit.

Okay.

So affirming the consequent would be if we see someone wearing a suit and then we fallaciously conclude they must be running for office.

Right.

But that's just wrong.

Because the original premise allows for plenty of other people who wear suits but are not running for office like people at a wedding or waiters.

Just because pre -keyed is true doesn't guarantee jeopardy is true.

The inference is invalid.

And denying the antecedent is the other side of that coin.

If we know someone is not running for office, we might fallaciously conclude they are not wearing a suit.

And again, that fails for the same reason.

People who aren't running for office still wear suits.

A true neg -P doesn't guarantee a true neg -Key.

These fallacies have a strong intuitive pull because we tend to treat if P, then 30, as if it also means it's converse, if Ted is then another, which is logically wrong.

This clash between our intuition and formal logic is tested perfectly by the Wason Selection Task, the famous four -card problem.

It is the ultimate diagnostic tool for human logic's failures.

So you're shown four cards, A, D, 4, 7.

The rule you have to test is if a card has a vowel, then it has an even number.

And the goal is to turn over the minimum number of cards to test if that rule is true.

The cards represent Teddy, P, Me, and Teddy Q.

That's right.

Now the overwhelming majority of people correctly choose A.

Which is Tito.

Why is that the right move?

Because that's modus ponens.

You have to check $1 to see if it delivers dollars.

If that A card has a 7 on the back, the rule is immediately broken.

But the critical failure point is the 7 card.

Most people either ignore it completely or they choose the 4 card instead.

And choosing the 4 card, which is dollars,

is the mistake of affirming the consequent.

Right.

Exactly.

If you turn over the 4 and find a vowel, great, you've confirmed the rule.

But if you find a consonant, the rule is still safe.

The rule only says what must follow a vowel, not what must precede an even number.

So the card you absolutely must choose is the 7, which represents negative two dollars.

Right.

Because if that 7 has a vowel on the back, the rule is instantly falsified.

Checking the 7 is modus ponens.

And it's the single most effective way to check for a violation.

The reason people miss it is that they are naturally drawn to seeking confirmation turning over the A and the 4 rather than seeking falsification, which is what turning over the 7 does.

And that inability to search for the crucial counterexample is a pattern we're going to see again and again.

Okay, let's move to another structure of deduction.

Syllogistic reasoning.

Specifically, categorical syllogisms, which deal with classes of entities using quantifiers like all, none, and some.

This form is usually represented by three statements.

Two premises and a conclusion.

All dealing with the relationships between different classes like A, B, and C.

Again, clarity really hinges on the precise logical definition of those quantifiers.

All and none are fairly unambiguous.

But some is a cognitive trap.

It absolutely is.

Logically, some means at least one and perhaps all.

This is fundamentally different from how we use it in conversation, where some x or y is often taken to mean explicitly that some x are not y.

This confusion drives so many syllogistic errors.

The formal rules of logic also tell us that any syllogism with two negative premises or two some premises will always be invalid.

It has no necessary conclusion.

And we need to keep that performance reality in mind.

The majority of categorical syllogisms you can construct actually have no valid conclusion.

Logically, sound thinking requires recognizing when a necessary relationship just doesn't exist.

But people are significantly slower and much more error -prone when premises include that ambiguous quantifier some or when they're stated negatively.

Can you give us a compelling example of that error specifically with the difficulty of the word some?

Sure.

Take this set of premises.

Some businessmen are Republicans, some Republicans are conservative.

The vast majority of people will incorrectly conclude that some businessmen are conservative.

They just create a chain in their head, businessmen to Republicans to conservative, and assume there has to be an overlap.

They do.

But logically, nothing necessarily follows.

And to prove it's invalid, you have to search for a counterexample that satisfies the premises but falsifies the conclusion.

Okay, what would that look like?

Imagine this.

The group of businessmen who are Republicans are all moderates, not conservative.

And the group of Republicans who are conservative are all lawyers, not businessmen.

In that scenario, both premises are still true, but the conclusion that some businessmen are conservative is false.

The failure here stems from the inability to consider and construct all the possible ways those three classes could be arranged.

So now we switch gears from the guaranteed world of deduction to inductive reasoning.

This is where we hypothesize, classify, and predict.

It's about expanding knowledge in the face of uncertainty.

Right.

And a primary function of induction is analogical reasoning.

This is reasoning based on comparison, formatted as A is to B, as C is to blank.

So the task is to figure out the relationship between A and B, and then apply that same relationship structure to C to find D.

Exactly.

You see this in verbal analogies, like dogs to cocker spaniel as cat is to Persian.

The relationship is the category to a specific subtype.

It requires you to access your semantic memory and apply that structure.

Or more complex ones, like Washington is to one as Jefferson is to what?

Right.

If you encode Washington and one is first president, you then map Washington to Jefferson They're both presidents.

And you apply that same relationship to Jefferson, who is the third president, which gives you three.

The difficulty of any analogy is directly related to how complex the terms are, how much, you know, how easy it is to find that core relationship, and whether there are other plausible answers competing for your attention.

And we know from our deep dives on problem solving that this kind of analogical reasoning is often the key to tackling new problems.

You map the structure of a known solution onto a new unsolved problem.

It is a core cognitive tool.

And another critical inductive task, which really illustrates our predictable failures, is Peter Watson's 2 -4 -6 task.

Right.

For those who aren't familiar, you're given the numbers 2 -4 -6 and told they follow a specific secret rule.

You have to figure out the rule by generating your own three number sets, and the experimenter just tells you yes or no.

And the problem is just insidious.

In the original study, only a tiny fraction of participants got the right rule, which was simply any three increasing numbers without first confidently guessing an incorrect rule.

They almost always start with a very narrow specific idea, like numbers increasing by two.

And their method of testing that narrow hypothesis reveals that bias we touched on earlier with the card task, confirmation bias.

Confirmation bias is this powerful tendency to seek out and focus only on evidence that confirms what you already believe, rather than looking for counter examples that could disprove it.

So in the 2 -4 -6 task, a person who thinks the rule is increasing by two will generate triplets like 10, 12, 14, or 20, 22, 24.

And the experimenter says yes to all of them, which just reinforces their wrong belief.

They're operating under the logical assumption that if their hypothesis is true, then the data should confirm it.

But when they get that dollar as a yes, they fallaciously conclude the dollar is proven.

Which is, once again, the fallacy of affirming the consequent, just playing out in an inductive context.

Exactly.

And the logical pitfall, as Watson explained, is that an infinite number of hypotheses are consistent with that initial 2 -4 -6.

If you only generate examples that follow your narrow rule, you never force yourself to test the boundaries.

So the reason people fail is that they never generate a counter example, like 3 -5 -7.

That's it.

If they tried 3 -5 -7 and the experimenter said yes, that follows a rule, the participant would instantly know their hypothesis about even numbers or increments of two was wrong.

That failure to actively attempt falsification, to try and prove your own hypothesis false, is the core limitation of human inductive reasoning and the defining method of the scientific process.

Okay, so the tasks we've discussed so far are these beautiful clean examples of formal reasoning.

But how well do these findings actually apply to the messy reality of solving problems in our own lives?

This brings us to everyday reasoning.

Let's use that simple mozzarella cheese example.

You realize you need mozzarella for dinner, you look in the fridge, you don't find any, and you conclude you have to drive to the store.

That decision, which seems so simple, is actually a mix of formal and inductive steps.

The first part, the search, is deductive.

If I had mozzarella, it would be in the fridge.

It is not in the fridge, therefore I have no mozzarella.

That's a perfect, rapid use of modus tollens.

It is.

But the next step, driving to the store, is an act of induction, and it's full of uncertainty.

You induce.

In the past, the store has always had mozzarella, therefore it is highly probable that the store will have it today.

The mental processes are recognizable, but the context is just vastly different from a lab syllogism.

And cognitive psychologists Collins and Michalski highlighted this difference, noting that everyday life requires these processes like plausible deduction or induction that don't really fit the rigid rules of the lab.

Their example of checking a fact illustrates this really well.

If you're asked if Uruguay is in the Andes Mountains, you might not know the answer directly.

But you might reason plausibly.

Most typical South American countries have parts in the Andes.

Uruguay is a typical South American country, therefore it's plausible that Uruguay is in the Andes.

This is probabilistic, evidence -based reasoning, relying on prototypes and general knowledge.

It leads to a high -probability conclusion that could still be wrong.

Which suggests a really stark contrast between these lab tasks and real -life problem -solving.

Let's lay out the key structural differences that separate formal reasoning from everyday reasoning.

Okay, so the first difference is the nature of the premises.

In formal tasks, all the premises are explicitly supplied.

The problem is completely self -contained, often using meaningless terms like daxes or wugs to strip away any content effects.

Whereas in everyday reasoning, the premises are often implicit or even missing, and you have to generate them yourself, like remembering if the store has ever run out of mozzarella before.

Exactly.

The second difference is the answers.

Formal deduction dictates one, unambiguously correct answer, valid or invalid.

Everyday problems rarely have a single right answer.

They result in multiple potential answers that vary in quality.

Do I use cheddar instead?

Do I drive 10 minutes?

Do I order expensive takeout?

Right.

And finally, there's the goals and relevance.

Formal content is academic.

You solve it for its own sake, often with no personal consequence.

But everyday content is intensely personally relevant.

Solving a relationship conflict, getting dinner on the table, making a budget decision, it's always a means to an end.

We have to keep this contrast in mind as we analyze why our performance so often fails to meet formal, logical standards.

We've established that human reasoners are systematically error -prone, even on these simple formal tasks.

So let's dedicate some serious time to understanding the root causes.

What are the predictable patterns of error that cognitive psychologists have identified?

Well, the first is simple.

The effects of premise phrasing.

Just the specific words in the syntax used can dramatically influence how long it takes you to process something and how accurate you are.

The complexity of the syntax puts a load on our cognitive resources.

Precisely.

Premises that contain negatives, using words like no or not, are consistently harder to process.

They require longer response times, and they generate significantly more errors.

That's a finding that's been robust for decades.

And we've seen the same thing with quantifiers.

Yes.

The quantifier sum is much harder to manipulate than the categorical extremes of all or none, largely because it's so ambiguous and complex to interpret.

Even the order matters.

A syllogism presented a -b, then b -c, is much easier to solve than one presented a -b, then c -b.

Right, because in the second case, the reasoner has to spend valuable working memory resources just rearranging the elements before they can even attempt the actual logical operation.

It all really boils down to working memory capacity, doesn't it?

It does.

Complex syntax, a jumbled order, negatives, ambiguous quantifiers.

It all consumes significant working memory just for encoding and representation.

And if your prefrontal cortex is already taxed trying to hold and organize the premises, there are just fewer resources left for the really challenging task of checking the conclusion and searching for counter examples.

Okay, so a second major source of error is a cognitive shortcut.

The alteration of premise meaning.

We basically misinterpret the problem because we automatically impose our conversational assumptions onto these formal logic problems.

We are constantly assuming more than is explicitly stated.

For instance, when presented with the premise, all daxes are wugs, many people immediately commit the conversion fallacy and mistakenly interpret that to also mean all wugs are daxes.

They fail to consider that the class of wugs might be much bigger than the class of daxes.

Exactly.

And we already covered the conversational misinterpretation of some people assume that some x or y must imply some x or not y.

We confuse the logical statement with our expectation of social conversational rules.

That misinterpretation then leads directly into the third pervasive error, the failure to consider all possibilities.

This is perhaps the most significant structural failure in human reasoning.

When you're solving complicated syllogisms, especially those involving sum,

the logical possibility space is just enormous.

For the simple two -premise problem, some A's are B's, some B's are C's, logicians using Venn diagrams have identified 41 possible spatial arrangements of those classes.

41 possibilities.

That's an overwhelming cognitive load.

It is.

The average human reasoner typically constructs only one or two initial mental models, settles on a conclusion that works for those models, and then just stops searching for alternatives.

If a conclusion only holds true for one out of 41 models, it's logically invalid.

We saw this in the Wausen 246 task too.

The participants stop searching after finding one simple rule that gets confirmed, failing to imagine the vast number of other possible rules.

Right.

And we also have to discuss the role of context and belief.

The powerful content and believability effects.

The content of the premises can completely override the logical structure.

This is where that Wausen selection task variant comes in, from Griggs and Cox.

Yes.

The most famous demonstration.

When the task was abstract vowels and even numbers, performance was terrible.

But when the identical logical structure was wrapped in concrete, familiar content, the rule if a person is drinking a beer, then the person must be over 19 years of age,

performance just soared.

Something like 75 % of college students solved it correctly.

So why does familiar content suddenly make us logically competent when abstract content leaves us floundering?

The primary explanation is memory queuing.

Familiar content, like laws or social norms about drinking age,

activates relevant real -world schemas or scripts that you have in your long -term memory.

And those schemas basically give you an instruction manual for how to think about the problem.

They do.

They cue the reasoner that violations are the crucial thing to look for.

The rule violation in the drinking scenario, checking the 16 -year -old, the negative consequent, is a real -world concern that our memory easily flags.

And that triggers the necessary modus tollens operation.

And this also extends to the believability effect, which shows that our existing world knowledge can filter and validate conclusions even when the logic is totally flawed.

That's the classic finding from Evans and his colleagues.

People are far more likely to judge a conclusion as valid if it just reinforces their established beliefs or stereotypes, regardless of the logical derivation.

Like the syllogism, all college professors are intellectuals.

Some intellectuals are liberals, therefore some college professors are liberals.

Right.

People often accept that conclusion because it aligns with a common stereotype, even though it doesn't logically follow.

But if we change the content to something where the conclusion clashes with reality… A garrer rate drops.

Precisely.

If we substitute,

some men are teachers, some teachers are women.

If you followed the same flawed logic, you'd conclude, some men are women.

Which is obviously false.

And because that conclusion is patently false based on your world knowledge, you reject it, even if you couldn't logically explain why.

This shows that we often check validity against our beliefs first, rather than purely against the logical structure.

And finally, we saw evidence that emotions also degrade our capacity for sound reasoning.

Research has shown that when emotional words like punishment or hurt are inserted into these conditional reasoning tasks, people's performance decreases significantly.

They draw more invalid inferences compared to tasks using neutral language.

Emotional salience, just like complex syntax, seems to consume cognitive resources needed for careful logical assessment.

This all ties back to that fundamental confirmation bias.

Whether it's the Wesson task or judging a syllogism, the tendency to seek supporting evidence and ignore falsifying information is a major driver of systemic error in human It absolutely is.

Okay, so we've established the errors.

Now we need to look at the three major theories that try to explain the cognitive architecture behind our reasoning.

Let's start with the confidential approach, which was pioneered by Robert Sternberg.

The confidential approach treats reasoning like a computer program.

It argues that we can understand high -level intellectual performance by breaking it down into a sequence of small, measurable component cognitive processes, or mental subroutines.

The goal is to identify these components, figure out their function, and measure the time each one takes.

So let's use that verbal analogy we talked about earlier.

Washington is to one as Jefferson is to what?

To trace the components Sternberg identified.

The process starts with encoding.

That's the time it takes to mentally represent each term, retrieve its meaning from memory, and assign temporary labels.

Then you move to inferring.

Yes, inferring the relationship between the first two terms, A and B.

For example, finding that Washington was the first president.

Third is mapping.

Finding the relationship between the first term, A, and the third term, C.

In this case, Washington and Jefferson were both presidents, both founders.

And once that structured relationship is established, you get to the actual solution finding.

Which is applying.

You take the inferred relationship you found between A and B, you map it onto C, and you generate the necessary D term.

So if the AB relationship was positioned in office, applying that to Jefferson, the third president yields three.

Okay, but how did Sternberg experimentally isolate and measure the time for these separate lightning fast components?

He used a really clever methodology with a tachistoscope, which briefly flashes images.

He'd show participants parts of the analogy before the full problem was presented.

A technique called pre -queuing.

So you might show them just the Washington part first?

Exactly.

And if you pre -queue only term A, the participants total solution time is reduced by exactly the amount of time it takes to encode A.

If you pre -queue A and B, the time savings gives you an estimate for both the encoding and the inference time.

By systematically manipulating what was pre -queued, Sternberg built a quantitative model of the reasoning process.

But Sternberg's model also recognized that reasoning isn't just about these performance components.

Absolutely.

He defined three types.

The performance components are the ones we just detailed.

Encoding, inferring, applying.

But then you have meta -components.

These are the executive control, functions planning, monitoring, strategy selection, resource allocation.

They would decide, OK, I'll try the number in office strategy first.

Or I should check my work for bias.

And the third type relates to learning.

Those are the knowledge acquisition components, which we use when we encounter and internalize new information.

They include things like selective encoding, deciding what's relevant and what's noise.

And selective combination, integrating new facts into a coherent whole.

Under this model, errors come from either poor encoding or a failure to use your meta -components effectively to monitor and correct your reasoning.

OK, moving to the second framework, the rules heuristics approach.

This perspective suggests that we don't necessarily break everything down into tiny subroutines, but that reasoning relies on specialized, inherent mental rules, almost like a mental logic or an internal grammar.

Right.

Proponents, like Martin Brain, argue that we possess a set of abstract inference rules that we access implicitly, just like we use grammatical rules without consciously knowing their formal definitions.

When premises are presented, our mind automatically matches them to these abstract rules.

For example, the rule.

If $10 and not Pendol, then $10.

Brain believed these fundamental rules are automatic and mostly errorless.

But this abstract rules theory ran into trouble explaining that massive content effect we saw in the Wasson selection task, didn't it?

If our logic rules are abstract and context -free, why did performance jump so dramatically when we switched from letters and numbers to beer and age?

Exactly.

And that's where Patricia Cheng introduced the idea of pragmatic reasoning schemata.

Schemata.

What are those exactly?

They're sort of intermediate between abstract logical rules and specific personal memories.

There are sets of context -sensitive rules that get activated by specific real -world situations like situations involving permission or obligation or causation.

So the permission schema, for example.

It has four distinct rules, like if the action is to be taken, then the precondition must be satisfied.

This schema gets evoked by the drinking age problem because our brain recognizes it as a permission structure.

But it would not be evoked by a structurally identical but abstract corporate travel policy problem.

This concept beautifully explains why context matters so much.

And the evolutionary approach adds another layer to this.

Leda Cosmides argues that our reasoning abilities aren't general purpose logic machines but are highly specialized by evolutionary pressures.

Cosmides' idea is that humans are just exquisitely skilled at reasoning about social contracts,

situations involving cooperation, benefits, costs, because of the intense evolutionary pressure to detect cheating.

Because being cheated threatens survival.

And reproductive success.

The drinking age Walton task works so well because the brain interprets it as a contract violation.

Someone is taking the benefit, drinking beer, without paying the cost, being over 19.

This specialized cheater detection mechanism overrides our general logical failures.

So under this whole rules heuristics framework,

errors aren't about slow processing but about failing to match a problem to the right rule or schema.

That's the idea.

Now let's move to the third major framework.

The mental models approach, championed by Philip Johnson Laird.

This theory makes a pretty radical departure, right?

It suggests we don't need special distinct inference rules at all.

It does.

Instead, reasoning is seen as an extension of language comprehension.

Meaning that the process of drawing a conclusion is fundamentally the same as the process of just understanding a complex sentence or a story.

Precisely.

Reasoning involves constructing and manipulating mental models, these spatial or visual representations, that depict the state of affairs described by the premises.

If I tell you the spoon is next to the mug, you don't use a formal rule, you construct a little mental image or representation of that relationship.

So if I have two premises, I just construct a model that integrates both pieces of information.

But how does that prevent the errors we've been talking about, like the failure to consider all possibilities?

This is the critical innovation of the mental models approach.

Good reasoning requires the reasoner to actively search for and construct alternative models,

counterexamples that are consistent with the original premises, but inconsistent with the tentative conclusion you drew from your first model.

So you're actively trying to falsify your own conclusion by using your imagination.

Yes.

Let's use that confusing syllogism again to show the power of this.

All of the beekeepers are artists, none of the chemists are beekeepers.

Okay.

The common initial model would probably separate all three classes.

Beekeepers are a circle inside the artist circle, and chemists are a separate circle over here.

Right, model A.

And that often leads to the false conclusion, none of the chemists are artists.

But the premises don't prohibit chemists from being artists, only from being beekeepers.

Exactly.

So to reason correctly, you have to generate model B.

Beekeepers are artists, but some of non -beekeeper artists are also chemists.

And the model C beekeepers are artists, and all the chemists are also artists, as long as they aren't beekeepers.

And only by checking your conclusion against all three of those logically distinct models can you find the single valid conclusion.

Which is,

some of the artists are not chemists, because those artists must be the beekeepers, who are explicitly excluded from being chemists by the second premise.

So the necessity of generating models B and C highlights the creative, imaginative aspect of this approach.

It really does.

Reasoning ability is often limited by a reasoner's inability, or just unwillingness to imagine the possibilities that might exist beyond that first, most plausible model they constructed.

This neatly explains the source of errors under this framework.

Yes, errors stem primarily from three sources.

First, a failure to construct any relevant models at all.

Second, a failure to assess the implications of the models you do find.

But most importantly, third, the failure to search for and construct enough alternative models.

This explains pretty much everything from syllogistic reasoning failures to the woollen tasks.

So if we synthesize these three robust theories, Componential, Realistic, and Mental models, it's clear they each explain different facets of human reasoning really well.

They offer complementary strengths.

The Componential approach is unparalleled in providing that microscopic, detailed analysis of the steps and time requirements.

The Rules Heuristics approach excels at explaining the powerful role of context and content, why we suddenly become logical when a problem touches on social contracts or permission.

And the Mental Models approach is praised for its general applicability, easily extending from formal logic to those messy everyday reasoning situations because it's built on the fundamental process of language comprehension and visual representation.

Our dive wouldn't be complete without looking inside the brain.

What does the neuropsychological evidence, particularly fMRI studies, tell us about the biological basis of all this?

Gole and his colleagues used fMRI to examine brain activity while people solved these three -term relational problems, like Karen is in front of Larry, Larry is in front of Jane.

And they made a fascinating discovery.

Different brain areas were activated depending on whether the premises were concrete, using named people or abstract, using letters like K, L, and J.

And the specific areas activated when reasoning with that abstract material lend some credence to one of our theories in particular.

Precisely.

When people reasoned with the abstract, non -name -based material, brain areas associated with visual and spatial perception were highly active.

This strongly suggests that when context and meaning are stripped away, the brain defaults to constructing and manipulating spatial representations, mental models, to solve the problem.

It's significant support for the Mental Models approach.

And there's further evidence that comes from studying patients with specific brain damage, particularly to the prefrontal cortex, or PFC.

Yes.

Walt and his colleagues found that patients with PFC damage had a very specific, crucial deficit.

They performed relatively normally on simple logical problems, but they struggled intensely with integrating multiple propositions, especially when the premises were presented in a non -optimal, non -direct order that required them to rearrange and hold the pieces of information together mentally.

So it wasn't a general intelligence deficit, but a specific problem with synthesis.

Exactly.

Their general IQ, memory, and comprehension were intact, but their ability to handle complexity that required synthesis, putting two or three premises together to form one unified mental representation,

was catastrophically hampered.

This points to a really specific vital role for the prefrontal cortex in our ability to reason complexly.

The evidence concludes that the prefrontal cortex is specialized for the integration of relations.

It is the vital area responsible for putting together disparate pieces of information, whether they're steps in a formal syllogism or facts in an everyday problem,

into a comprehensive, coherent mental picture.

This ability to synthesize is the cornerstone of robust, complex reasoning.

Okay, that was a lot.

Let's try to bring it all together.

To summarize this extensive deep dive, we've learned that reasoning is the goal -directed thought process that moves from premises to conclusions.

It's defined by the strict guarantee of deduction and the high probability of induction.

And we've learned that human reasoning, while powerful, is prone to these systematic and predictable errors.

They're linked to syntactic complexity, the conversion fallacy, our failure to search for all possibilities, and our tendency toward confirmation bias, which is driven by believability and memory queuing.

And we map the cognitive landscape by exploring those three major frameworks.

The confidential approach, focusing on microprocesses, the rules heuristics approach, emphasizing context and evolution, and the mental models approach, which views reasoning as the imaginative construction and critical testing of alternative possibilities.

And finally, the neuro findings suggest that complex reasoning really hinges on the prefrontal cortex's ability to integrate relations, to see how separate pieces of evidence fit together into a unified whole.

Right.

And this raises an important question for you, the listener, to mull over.

If the ultimate performance constraint in robust reasoning is this synthetic ability, the creation and testing of integrated mental models, what strategies, beyond just learning abstract logical rules, might we deliberately employ in our daily lives to strengthen our ability to synthesize separate isolated facts into a comprehensive, robust, and cohesive mental picture?

That's a powerful thought to take forward.

Thank you for joining us for this deep dive into the fascinating world of human reasoning.

We hope you feel well equipped to tackle your next logical challenge.