1. The Four Definitions of AI

Source: AIMA 4th Ed, §1.1 — What Is AI?

The Core Question

“What is AI?” is not simple to answer. Researchers have historically pursued four distinct goals, forming a 2×2 matrix:

Cell 1 — Think Like Humans (Cognitive Simulation)

• Goal: Make machines whose internal decision-making process mirrors human cognition.
• Requires introspection or psychological experiments to know “how humans think.”
• Field: Cognitive science — the intersection of AI and cognitive psychology.
• Key idea: if we understand human thought precisely enough, we can replicate it.

Cell 2 — Act Like Humans (The Turing Test Approach)

• Goal: Behavioral indistinguishability from humans.
• Turing Test (1950): A human interrogator sends text questions to an unknown entity. If the human cannot reliably distinguish the machine from a person, the machine passes.
• Skills required: NLP, knowledge representation, automated reasoning, machine learning, computer vision, robotics.
• Weakness: Passing the Turing Test ≠ actually intelligent — it’s a behavioral benchmark, not a cognitive one.

Cell 3 — Think Rationally (Laws of Thought)

• Goal: Follow the “right” rules of logic — Aristotle’s syllogisms, formal deductive reasoning.
• Approach: encode world knowledge as logical sentences; use inference to derive conclusions.
• Limitation: (a) hard to encode everything in formal logic; (b) reasoning from noisy or incomplete information is not handled well; (c) exponential complexity in the worst case.

Cell 4 — Act Rationally (The Rational Agent)

• Goal: Take the best possible action given beliefs about the world and a measure of performance.
• A rational agent acts to maximize its expected performance measure.
• This is AIMA’s chosen framework — it subsumes the “laws of thought” approach and is more general.
• Key advantage: it is mathematically precise, does not require modeling human thought, and handles uncertainty naturally.

Which Definition Does AIMA Use?

AIMA adopts the rational agent approach (act rationally).

“The standard model: AI is concerned with building agents that act rationally.”

Rational action is defined relative to a performance measure — a standard that defines what counts as success for an agent in its environment.

Key Terms