What Is Intelligence? Sensing, Thinking, and Acting

Intelligence is one of those words that seems obvious until we try to define it.

We call:

• a person intelligent,
• an octopus intelligent,
• a chess program intelligent,
• a colony intelligent.

But are we talking about the same thing?

Part XV begins with the question:

What is intelligence?

Intelligence Is Not One Ability

Intelligence can involve:

• sensing,
• learning,
• reasoning,
• planning,
• acting,
• adapting.

Different systems excel in different combinations.

There may be no single essence.

Goal Achievement

One influential idea defines intelligence in terms of achieving goals across environments.

A more intelligent system succeeds:

• more reliably,
• in more varied situations.

This emphasizes flexible competence.

Sensing

An intelligent agent needs information about its environment.

Sensors may include:

• eyes,
• ears,
• cameras,
• microphones.

Without sensing, action is blind.

Perception Is Not Raw Input

Sensing produces signals.

Perception organizes those signals into useful structure.

A camera gives pixels.

An intelligent system may need:

• objects,
• motion,
• depth,
• relevance.

Thinking

Thinking can include:

• inference,
• prediction,
• problem solving,
• planning.

But intelligence should not be equated only with internal deliberation.

Some intelligent behavior is fast and automatic.

Acting

Intelligence becomes visible through action.

An agent must choose what to do.

Even a perfect world model is useless if it cannot guide behavior.

Adaptation

A rigid system can perform one task brilliantly.

But general intelligence requires some ability to adapt when:

• rules change,
• goals change,
• environment changes.

Flexibility is central.

Learning

Learning changes future behavior based on experience.

A learning system can improve without being explicitly reprogrammed for each case.

This distinguishes adaptive intelligence from fixed competence.

Generalization

A system is more impressive when it succeeds on cases it has not seen before.

Memorizing examples is not enough.

It must extract regularities.

Transfer

Transfer means knowledge learned in one context helps in another.

Humans are often good at transferring abstract structure.

Machine systems vary widely.

Problem Solving

A problem exists when:

• current state,
• desired state

are separated by unknown steps.

Intelligence searches for a path.

This connects directly to algorithms.

Search

Many forms of intelligent behavior can be framed as search over possibilities.

Examples:

• chess moves,
• routes,
• plans,
• proofs.

The difficulty is avoiding combinatorial explosion.

Heuristics

Heuristics guide search.

They help prioritize promising options.

A strong agent does not examine every possibility equally.

Planning

Planning constructs action sequences before execution.

It requires:

• model of consequences,
• goals,
• evaluation.

Planning converts imagination into control.

Reactive Intelligence

Not all intelligent behavior requires long planning.

A skilled driver reacts rapidly.

A fly avoids collision.

Fast sensorimotor loops can be highly adaptive.

Deliberative Intelligence

Deliberative intelligence uses:

• explicit models,
• memory,
• simulation.

It is slower but more flexible.

Many intelligent systems combine reactive and deliberative modes.

Embodied Intelligence

An agent’s body shapes what intelligence requires.

A bird solves flight control.

A human solves manipulation and social communication.

Intelligence is relative to embodiment.

Ecological Intelligence

A capability valuable in one environment may be useless in another.

Intelligence should be evaluated relative to:

• goals,
• environment,
• constraints.

There is no context-free performance.

Domain-Specific Intelligence

Some systems are highly intelligent in narrow domains.

A chess engine can outperform every human at chess.

It cannot necessarily:

• cook,
• teach,
• navigate a room.

Narrow excellence is not general intelligence.

General Intelligence

General intelligence involves competence across many tasks.

It includes:

• transfer,
• abstraction,
• adaptation.

Human intelligence is relatively general compared with most engineered systems.

Intelligence and Knowledge

Knowledge helps intelligence.

But storing facts is not the same as reasoning.

A database can contain enormous information without acting intelligently.

Intelligence and Speed

Fast calculation can contribute to performance.

But speed is not intelligence by itself.

A pocket calculator multiplies faster than most people.

It does not possess broad cognition.

Intelligence and Memory

Good memory helps.

But perfect memory is neither necessary nor sufficient.

Human intelligence depends heavily on:

• abstraction,
• forgetting,
• compression.

Intelligence and Rationality

An intelligent system may reason well.

But humans often violate formal rationality while remaining clearly intelligent.

Intelligence and rationality overlap.

They are not identical.

Intelligence and Consciousness

A system can perform intelligent tasks without obvious evidence of consciousness.

Chess engines are an example.

Therefore:

intelligence ≠ consciousness.

Intelligence and Understanding

A system may produce correct answers through:

• lookup,
• pattern matching,
• reasoning.

Whether this counts as understanding is a deeper question.

Performance alone may underdetermine mechanism.

Intelligence and Creativity

Creating useful novelty may be one dimension of intelligence.

But creativity depends on:

• representation,
• exploration,
• evaluation.

It will receive its own treatment later.

Intelligence as Compression

A powerful system often discovers compact regularities.

Instead of memorizing every case, it learns a model.

Compression supports prediction.

Intelligence as Prediction

Predicting consequences is central to adaptive behavior.

An agent estimates:

If I do X, what happens?

Prediction supports planning.

Intelligence as Control

From another perspective, intelligence is effective control under uncertainty.

The system maintains desirable states despite:

• noise,
• change,
• limited information.

This connects intelligence to cybernetics.

Cybernetics

Cybernetics studied:

• control,
• communication,
• feedback.

Intelligent behavior can be understood as regulation through information.

The historical connection to AI is deep.

Intelligence and Feedback

An intelligent system acts.

Then observes consequences.

Then updates.

This closed loop is essential.

Open-loop behavior cannot adapt.

Agent

An agent is a system that:

• perceives,
• decides,
• acts.

AI often defines intelligence through agent performance.

This avoids requiring human-like internal processes.

Rational Agent

A rational agent selects actions expected to maximize performance relative to:

• goals,
• information,
• constraints.

This became a dominant AI framework.

Bounded Rationality

Real agents have limited:

• time,
• memory,
• computation.

Therefore optimal action may be impossible.

Intelligence includes managing constraints.

Satisficing

An agent may seek a solution that is:

good enough.

This can be more intelligent than spending infinite resources chasing perfection.

Social Intelligence

Humans live in social worlds.

Intelligence includes:

• cooperation,
• deception,
• negotiation,
• empathy,
• norm understanding.

Pure puzzle-solving misses this dimension.

Emotional Intelligence

Emotion recognition and regulation also contribute to effective action.

The concept is broader and less precisely measured than traditional cognitive intelligence.

Still, social and affective skill clearly matters.

Collective Intelligence

A group can outperform its members.

Examples include:

• scientific communities,
• markets,
• teams.

Intelligence can emerge from organization.

Swarms

Ant colonies solve routing problems without central control.

The colony exhibits intelligent behavior.

No individual ant contains the full solution.

Intelligence Without Language

Animals demonstrate planning and learning without human language.

Therefore language is not required for all intelligence.

It may greatly expand abstraction and communication.

Intelligence Without Consciousness?

Machines make this question unavoidable.

If a system:

• learns,
• plans,
• adapts

but has no experience, is it intelligent?

Functionally, yes in many definitions.

Phenomenal consciousness is separate.

IQ

IQ tests measure some human cognitive abilities.

They do not define intelligence universally.

They are designed for:

• humans,
• specific psychological constructs.

They cannot simply be applied to machines or animals.

The g Factor

Human psychometrics finds a statistical general factor, often called g, across many cognitive tasks.

This is an important empirical regularity.

It does not prove intelligence is one indivisible faculty.

Multiple Intelligences

Howard Gardner proposed multiple intelligences such as:

• linguistic,
• spatial,
• musical.

The theory became culturally influential.

Its status as a psychometric theory is controversial.

Intelligence as a Family of Capacities

A cautious view treats intelligence as a structured cluster of abilities.

Different systems can occupy different profiles.

Comparison may require several dimensions.

Legg–Hutter Definition

Shane Legg and Marcus Hutter proposed a formal definition of universal intelligence based on expected performance across environments weighted by simplicity.

The framework is mathematically interesting.

It is not a practical universal intelligence test.

No Free Lunch

In optimization, No Free Lunch results show that no algorithm performs best across all possible problems without assumptions about problem distribution.

This reinforces a key point:

intelligence depends on structure in environments.

Priors

Every intelligent system has inductive biases.

Humans have:

• evolved priors,
• developmental biases.

Machines have:

• architectures,
• training objectives.

Learning never begins from absolute neutrality.

Intelligence as Adaptation to Structure

Perhaps intelligence is the ability to exploit regularities in a world.

Where there is structure, models can help.

Where there is no exploitable structure, intelligence has nothing to learn.

The Philosophical Lesson

Intelligence is not just calculation.

It is the capacity to:

• perceive,
• learn,
• model,
• reason,
• act,
• adapt

under uncertainty and limited resources.

Different systems realize these capacities differently.

The Beginning of Part XV

We now enter:

Intelligence and Artificial Intelligence.

The central question is no longer only how minds work.

It is whether intelligence can be understood, formalized, and built.

The Next Question

Humans are not the only intelligent systems.

Animals solve ecological problems.

Machines solve computational ones.

How should we compare these forms without making humans the universal standard?

The next essay asks:

Human, Animal, and Machine Intelligence.