What Is Intelligence, Really? A Working Framework

Before we can judge whether an AI is intelligent, we need a framework for what intelligence even means. Draw on Chollet, Dennett, and modern evals.

45 min · Reviewed 2026

The Definition Problem

You cannot evaluate something you cannot define. Psychologists, philosophers, and computer scientists have all taken cracks at intelligence, and none of their answers fully agree. That ambiguity quietly infects every AI debate.

For our purposes, start with François Chollet's working definition from his 2019 paper: intelligence is a measure of skill-acquisition efficiency over a scope of tasks, with respect to priors, experience, and generalization difficulty. That sounds dense, but it pays off.

Unpacking Chollet's definition

Skill-acquisition efficiency: how fast you learn new tasks
Scope: the variety of tasks the agent can handle
Priors: what the agent already knows going in
Experience: the data and practice it has had
Generalization difficulty: how far the new task is from anything seen before

ARC-AGI: a benchmark built on this idea

ARC is a set of visual reasoning puzzles designed so each task requires inferring a rule from a handful of examples. Humans solve most of them. For years, the best AI systems could not pass 30 percent. Reasoning models have closed the gap but still struggle compared to humans.

Input grid → Output grid . . X . . . X . . . . . → . . X . . . . . . . X . Infer: fill a vertical line below the X. Apply to a new grid the model has never seen.ARC tasks measure learning a rule from 2-3 examples and applying it to novel inputs.

Competing frameworks

Framework	Core claim
Turing (1950)	If you cannot tell it from a human, call it intelligent
Russell & Norvig	Intelligence is rational action maximizing expected utility
Chollet	Intelligence is efficient generalization to novel tasks
Dennett	Intelligence is intentional stance at scale
LeCun's world models	Intelligence requires internal causal simulation

Why it matters for product decisions

If you believe intelligence is skill count, scale wins
If you believe it is generalization, you need new architectures
If it is rationality, alignment becomes the central problem
Your framework shapes which benchmarks you trust

The measure of intelligence is the ability to change.
— Albert Einstein (attributed)

The big idea: intelligence is not a single dial. Choose a working definition, then judge every AI claim against that lens. You will cut through most of the noise.

End-of-lesson check

8 questions · take it digitally for instant feedback at tendril.neural-forge.io/learn/quiz/end-creators-what-is-intelligence-really

What is the main idea of "What Is Intelligence, Really? A Working Framework"?
1. Before we can judge whether an AI is intelligent, we need a framework for what intelligence even means. Draw on Chollet, Dennett, and modern evals.
2. Use AI as the final authority for the whole decision
3. Avoid checking the answer once it sounds polished
4. Focus only on speed instead of judgment
Which concept is most central to "What Is Intelligence, Really? A Working Framework"?
1. generalization
2. intelligence
3. ARC
4. Chollet
Which use of AI fits this topic best?
1. Let the AI decide what matters without your review
2. Use the answer before checking whether it fits the situation
3. Skill-acquisition efficiency: how fast you learn new tasks
4. Treat the AI output as automatically correct
What should a careful learner remember about "The key shift"?
1. Use AI to draft or organize ideas about intelligence, then verify before acting.
2. Skip the context so the tool can guess faster
3. Treat the output as private even after sharing it online
4. Use the answer without checking the source
You want to use AI after this lesson. What is the safest next step?
1. Act immediately because the AI answer is written clearly
2. Use AI for drafting and comparison, but verify before publishing or relying on it.
3. Hide uncertainty so the final answer looks cleaner
4. Use private or sensitive details before checking permission
How should AI output about intelligence be treated?
1. As proof that no other source is needed
2. As a replacement for context, consent, or expert review
3. As a draft or helper output that still needs human judgment and verification
4. As something that becomes correct when it sounds confident
Name one way to verify an AI answer about intelligence.
Which action would help you apply "What Is Intelligence, Really? A Working Framework" responsibly?
1. Use the tool to avoid thinking through the tradeoff
2. Keep going even if the output conflicts with a trusted source
3. Treat the AI output as automatically correct
4. Scope: the variety of tasks the agent can handle

← Back to interactive lesson

Tendril · Creators · AI Foundations

What Is Intelligence, Really? A Working Framework

Before we can judge whether an AI is intelligent, we need a framework for what intelligence even means. Draw on Chollet, Dennett, and modern evals.

45 min · Reviewed 2026

The Definition Problem

Unpacking Chollet's definition

Skill-acquisition efficiency: how fast you learn new tasks
Scope: the variety of tasks the agent can handle
Priors: what the agent already knows going in
Experience: the data and practice it has had
Generalization difficulty: how far the new task is from anything seen before

ARC-AGI: a benchmark built on this idea

Input grid → Output grid . . X . . . X . . . . . → . . X . . . . . . . X . Infer: fill a vertical line below the X. Apply to a new grid the model has never seen.ARC tasks measure learning a rule from 2-3 examples and applying it to novel inputs.

Competing frameworks

Framework	Core claim
Turing (1950)	If you cannot tell it from a human, call it intelligent
Russell & Norvig	Intelligence is rational action maximizing expected utility
Chollet	Intelligence is efficient generalization to novel tasks
Dennett	Intelligence is intentional stance at scale
LeCun's world models	Intelligence requires internal causal simulation

Why it matters for product decisions

If you believe intelligence is skill count, scale wins
If you believe it is generalization, you need new architectures
If it is rationality, alignment becomes the central problem
Your framework shapes which benchmarks you trust

The measure of intelligence is the ability to change.
— Albert Einstein (attributed)

The big idea: intelligence is not a single dial. Choose a working definition, then judge every AI claim against that lens. You will cut through most of the noise.

End-of-lesson check

8 questions · take it digitally for instant feedback at tendril.neural-forge.io/learn/quiz/end-creators-what-is-intelligence-really

What is the main idea of "What Is Intelligence, Really? A Working Framework"?
1. Before we can judge whether an AI is intelligent, we need a framework for what intelligence even means. Draw on Chollet, Dennett, and modern evals.
2. Use AI as the final authority for the whole decision
3. Avoid checking the answer once it sounds polished
4. Focus only on speed instead of judgment
Which concept is most central to "What Is Intelligence, Really? A Working Framework"?
1. generalization
2. intelligence
3. ARC
4. Chollet
Which use of AI fits this topic best?
1. Let the AI decide what matters without your review
2. Use the answer before checking whether it fits the situation
3. Skill-acquisition efficiency: how fast you learn new tasks
4. Treat the AI output as automatically correct
What should a careful learner remember about "The key shift"?
1. Use AI to draft or organize ideas about intelligence, then verify before acting.
2. Skip the context so the tool can guess faster
3. Treat the output as private even after sharing it online
4. Use the answer without checking the source
You want to use AI after this lesson. What is the safest next step?
1. Act immediately because the AI answer is written clearly
2. Use AI for drafting and comparison, but verify before publishing or relying on it.
3. Hide uncertainty so the final answer looks cleaner
4. Use private or sensitive details before checking permission
How should AI output about intelligence be treated?
1. As proof that no other source is needed
2. As a replacement for context, consent, or expert review
3. As a draft or helper output that still needs human judgment and verification
4. As something that becomes correct when it sounds confident
Name one way to verify an AI answer about intelligence.
Which action would help you apply "What Is Intelligence, Really? A Working Framework" responsibly?
1. Use the tool to avoid thinking through the tradeoff
2. Keep going even if the output conflicts with a trusted source
3. Treat the AI output as automatically correct
4. Scope: the variety of tasks the agent can handle

← Back to interactive lesson