Mental Models

A mind does not merely react.

It anticipates.

It constructs simplified internal models of:

• places,
• people,
• causes,
• possibilities.

These are often called mental models.

A mental model is not a perfect copy of reality.

It is a working representation used to predict, explain, and act.

Models in the Mind

Suppose you are planning how to move a sofa through a narrow doorway.

Before acting, you may imagine:

• rotate it,
• tilt it,
• try another angle.

You are running an internal simulation.

Representation for Action

A mental model is useful because it supports:

• prediction,
• planning,
• inference.

Its value lies in what it lets the mind do.

Not a Photograph

A mental model is selective.

You may know the layout of your home without remembering:

• exact wall texture,
• every object,
• precise dimensions.

The model preserves relevant structure.

Compression

Mental models compress reality.

Instead of representing every detail, they encode:

• regularities,
• causal relations,
• useful distinctions.

This reduces cognitive load.

Spatial Models

People can form internal representations of space.

Examples include:

• routes,
• room layouts,
• relative positions.

These models support navigation.

Cognitive Maps

Edward Tolman’s work popularized the idea of cognitive maps.

Animals appeared to learn spatial structure rather than merely fixed stimulus-response chains.

This suggested internal representation of environment.

Hippocampal Support

Place cells, grid cells, and related neural systems support spatial navigation.

They provide biological evidence for structured internal representations of space.

But a cognitive map is more than one cell type.

Causal Models

Mental models also represent causation.

If:

switch → light

we expect flipping the switch to alter illumination.

Causal knowledge supports intervention.

Counterfactuals

A causal model lets us ask:

What if?

What would happen if the switch failed?

What if I took another route?

Counterfactual reasoning depends on models of alternatives.

Temporal Models

We also model sequences.

Examples:

• cooking steps,
• project plans,
• social interactions.

A model can represent how states change over time.

Social Models

Humans build models of other minds.

We infer:

• beliefs,
• goals,
• intentions.

This supports prediction of behavior.

Theory of Mind

Theory of mind refers to capacities for representing mental states in others.

We understand that another person may:

• know something,
• believe something false,
• want something different.

This is a model of an agent.

False-Belief Tasks

Classic developmental tasks test whether children understand that another person can hold a false belief.

Success requires separating:

my knowledge

from:

their belief.

Mental models become nested.

Nested Models

We can reason:

“I think that she believes that he intends…”

This recursive structure is central to:

• social strategy,
• deception,
• cooperation.

It can become computationally expensive.

Models of Self

The mind also models:

• body,
• abilities,
• personality,
• intentions.

Self-models help coordinate behavior.

They need not be perfectly accurate.

Body Schema

A body schema represents the body’s configuration for action.

It supports movement without requiring conscious calculation of every limb position.

The body is modeled dynamically.

Body Image

Body image is related but more conscious and evaluative.

It concerns how one perceives and thinks about one’s body.

Schema and image are not identical.

Internal Models in Motor Control

Motor-control theories often distinguish:

• forward models,
• inverse models.

A forward model predicts consequences of an action.

An inverse model estimates what action could produce a desired outcome.

Forward Model

Given:

current state + motor command

predict:

future sensory state.

This helps rapid control.

Prediction Error

Compare predicted sensory outcome with actual input.

The difference is a prediction error.

The system can update:

• model,
• action.

Feedback and prediction work together.

Mental Models in Reasoning

Philip Johnson-Laird proposed that reasoning often uses mental models of possible situations rather than purely formal symbolic derivation.

To test an inference, people construct situations satisfying the premises.

Example

Premises:

All poets are readers.

Some teachers are poets.

A mental model makes it easy to see:

Some teachers are readers.

The conclusion is supported by the represented relation.

Model-Based Errors

Humans may fail when too many possible models must be considered.

We often construct:

one plausible scenario

rather than:

all logically possible scenarios.

This can produce reasoning errors.

Models and Bias

A mental model can be wrong.

If someone believes:

“all economic problems are caused by one factor,”

new evidence may be forced into that model.

Models guide perception and can distort it.

Updating

Good cognition requires model revision.

When prediction repeatedly fails, the model should change.

This resembles scientific theory revision at smaller scale.

Predictive Processing

Predictive-processing frameworks treat the brain as maintaining hierarchical generative models.

The system predicts sensory input and updates from error.

This gives mental modeling a central role.

Generative Model

A generative model represents how hidden causes could produce observations.

Instead of only mapping:

input → label,

it models:

cause → expected data.

This supports inference backward from observation to cause.

Bayesian Framing

The brain may combine:

• prior expectations,
• sensory evidence.

In Bayesian language:

[ P(H\mid D)\propto P(D\mid H)P(H) ]

This is a powerful formal analogy.

The brain need not explicitly manipulate this equation symbolically.

Priors

Prior expectations can improve perception when data is noisy.

But strong priors can also bias interpretation.

Perception balances:

expectation, evidence.

Illusions

Perceptual illusions reveal assumptions built into internal models.

The visual system makes useful guesses based on normal environments.

In unusual displays, those guesses can fail.

Models and Expertise

Experts have richer mental models.

A novice sees isolated facts.

An expert sees:

• causal structure,
• dependencies,
• likely consequences.

Expertise is partly better internal representation.

Mental Models and Education

Learning is not only adding facts.

It is restructuring models.

A student may memorize equations but misunderstand how concepts relate.

Deep learning changes the internal map.

Models in Science

Scientific models and mental models share features.

Both:

• simplify,
• select,
• predict.

Scientific models are public and formalized.

Mental models are often private and implicit.

Models Are Not Reality

A mental model can feel obvious.

That does not make it true.

The map is not the territory.

Cognition depends on useful approximations.

Multiple Models

The same situation can support multiple models.

An organization can be viewed as:

• hierarchy,
• communication network,
• incentive system.

Different models answer different questions.

Model Selection

A good thinker asks:

Which model is useful here?

Not:

Which single model is reality itself?

Flexibility matters.

Simulation

Mental simulation lets us test actions without performing them physically.

This saves:

• time,
• risk,
• energy.

Imagination becomes a control technology.

Imagination and Memory

Mental simulation draws on stored experience.

We recombine remembered elements into possible futures.

Memory and imagination are deeply connected.

Episodic Future Thinking

Humans can construct detailed imagined future events.

Some of the same brain systems support:

• remembering past episodes,
• imagining future ones.

Memory is not only backward-looking.

Planning

Planning uses a model to search possible action sequences.

A plan is a route through imagined state space.

This connects cognition directly to AI planning.

Model-Based Reinforcement Learning

In reinforcement learning, a model-based agent learns or uses a model of environmental transitions.

It can mentally evaluate outcomes before acting.

This resembles deliberative planning.

Model-Free Learning

A model-free agent can learn values or habits without explicitly representing the environment’s transition structure.

Brains may combine both styles.

Habit vs Deliberation

Habits can be fast and efficient.

Deliberation is flexible but costly.

Cognition balances stored policies with model-based simulation.

The Philosophical Lesson

Mental models are simplified internal structures used to:

• predict,
• infer,
• imagine,
• control.

They are powerful because they omit detail.

They are dangerous because omitted detail can matter.

A mind lives partly through models of worlds that are never identical to the world itself.

The Next Question

Mental models require information from the past.

How does the mind preserve experience?

Why do we remember some things and forget others?

And why is memory often reconstructed rather than replayed?

The next topic is:

Memory.