Rethinking Intelligence: What if Intelligence is Just Pattern Matching?

Egil Diau
1National Taiwan University
Paper Zenodo
Project teaser

TL/DR:   We propose a simple idea: most intelligent behavior is pattern matching—not symbolic logic, not Bayesian inference. The evidence comes from two sides. First, intelligence is difficult to observe directly—but its failures are not. These failures aren’t random—they follow clear, structured patterns of misalignment. Second, intuition and creativity come from contextual cues triggering stored structure, not explicit reasoning. We show that many core functions—language, reasoning, planning, and social inference—can be explained as pattern matching.

Failure Reveals the Structure of Intelligence

Causal Reasoning

Causal Reasoning
Post Hoc Fallacy, Illusory Causation

Planning

Planning
Planning Fallacy, Typicality Bias

Counterfactual Thinking

Counterfactual Thinking
Hindsight Bias, Outcome Bias

Social Inference

Social Inference
Fundamental Attribution Error, Stereotyping

Intelligence is hard to observe directly—but its failure modes are not. Across domains, we see systematic biases emerge not from randomness, but from structured pattern matching gone wrong. By tracing these consistent breakdowns, we can infer the underlying architecture of intelligence itself.

Abstract

Recent advances in large language models have challenged long-held assumptions about the nature of intelligence, showing that behaviours once thought to require symbolic reasoning or abstract planning can emerge from large-scale pattern matching.

We argue that this is not an artifact of artificial systems, but a reflection of how human intelligence itself operates. We support this view with two converging lines of evidence. First, cognitive biases—such as stereotyping and base-rate neglect—reflect systematic misalignments in pattern recognition, rather than logical errors. Second, intuition, creativity, and sudden insight often stem from rapid alignment with stored patterns, enabling flexible responses without explicit reasoning.

We revisit core cognitive functions—language, causal reasoning, planning, and social inference—and show that they can be unified under a single mechanism: memory-driven pattern matching. This reframing suggests that many forms of intelligent behavior can be explained as pattern matching over stored experience, rather than abstract computation.

Two Main Evidence

Core Cognitive Diagram

Why It Matters?

Intelligence is one of the most used—and least understood—concepts in science and technology. It’s invoked to sell products, justify theories, and predict the future. But what is it, exactly? The term is often overloaded, vague, or circular. Many use it as a placeholder for whatever humans or machines seem to do well. Others overhype it, assuming intelligence is a singular essence that will soon be replicated.

But real understanding begins with constraints. By focusing not on what intelligence can do, but where it consistently fails, we shift from speculation to structure. Biases and breakdowns aren't noise—they are the visible seams of an otherwise hidden system. This perspective reframes intelligence not as mystery or magic, but as a pattern-matching engine with traceable limits.