Analogical Reasoning

Analogical reasoning is the study of how people and AI systems recognize shared relationships between different situations and use them to transfer knowledge or solve new problems. Approaches include symbolic structure mapping, geometric rules over vectors or probability distributions, and language-model methods that retrieve examples or abstract and align relations. Researchers evaluate these methods on word and proverb analogies, mathematical problem solving, narratives, visual transformations, video, and cross-lingual tasks. Models can perform well on simpler comparisons, but often struggle to compose multiple relations, match structure across different forms, and distinguish deep causal or relational similarity from surface resemblance.