Inductive reasoning in minds and machines.

Area of Science:

• Cognitive Science
• Artificial Intelligence
• Psychology

Background:

• Human induction, the ability to generalize from knowledge, is crucial for intelligence.
• Existing cognitive models of induction are limited to simple problems and lack predictive power.
• Current large language models (LLMs) fail to replicate human inductive reasoning patterns.

Purpose of the Study:

• To develop a novel approach for modeling human inductive reasoning.
• To integrate LLM knowledge representations with established psychological theories of induction.
• To achieve human-like quantitative predictions for diverse induction arguments.

Main Methods:

• Combined rich knowledge representations from LLMs with cognitive psychology theories of induction.
• Utilized benchmark empirical findings on human induction for validation.
• Tested the model's ability to generate human-like responses to natural language arguments.

Main Results:

• The integrative approach successfully captured key empirical findings in human induction.
• The model demonstrated the ability to generate human-like responses for thousands of common categories and properties.
• Achieved quantitative predictions for complex induction arguments beyond toy problems.

Conclusions:

• Integrating AI (LLMs) with cognitive science theories offers a powerful framework for modeling high-level human cognition.
• This approach advances our understanding of the cognitive mechanisms underlying human induction.
• Demonstrates the potential of interdisciplinary methods in artificial intelligence and psychology to model complex cognitive functions.