Claire Stevenson

Claire Stevenson observes that, while analogical reasoning has been widely studied in human development and is only just beginning to emerge in large language models (LLMs), little is known about how AI acquires this skill.

Her research explores the development of analogical reasoning in both children and AI, as illustrated by puzzles such as “horse is to stable as chicken is to …?” She focuses particularly on the shift from associative thinking—for instance, “chicken–egg”—to relational reasoning, such as understanding “lives in.”

Whereas children typically make this transition between the ages of four and eight, AI models tend to exhibit early associative patterns but encounter difficulties with abstraction, generalisation, and compositional thinking.

By charting these developmental stages and comparing the trajectories of humans and machines, her study aims to both enhance AI reasoning and clarify whether models truly reason or merely reproduce associations and memorised patterns.