Psychology methods proposed to probe how LLMs reason

A new Review published in Science Advances argues that concepts and methods borrowed from psychology could help researchers evaluate whether and how large language models (LLMs) can employ reasoning. The paper, by Thilo Hagendorff and colleagues, is titled “Machine psychology for language models.”

The authors write that the ever-growing capabilities of these systems make them challenging but also increasingly important to understand and explain, especially since expanding capabilities also bring greater potential for unforeseen harm. Their proposed approach, called “machine psychology,” advocates seeking inspiration from the behavioral sciences.

Because psychology’s protocols already assess emotional, cognitive, and social development, the Review suggests they could be modified to examine LLMs’ intelligence, biases, decision-making, social skills, and language. Adapting established behavioral testing methods to artificial systems could, in principle, offer a more systematic way to quantify what these models can and cannot do.

The authors also urge caution. “Many psychological concepts are normatively laden and can foster mismatches in expectations between AI experts and the public regarding machine capability,” they write. In other words, applying human-centered psychological terms to machines risks creating misunderstandings about what LLMs actually achieve.

The Review was published on 28 October 2026 with the DOI 10.1126/sciadv.aec1912. As a peer-reviewed perspective piece, it presents a proposed research framework rather than new experimental results on LLM reasoning itself.

Subject of Research: Using psychological concepts and methods to evaluate reasoning and behavior in large language models

Article Title: Review: Perspectives from psychology could help quantify LLMs’ thinking capacity

Article References: Review: Perspectives from psychology could help quantify LLMs’ thinking capacity. Original publication (Bibliographic details unavailable.)

Image Credits: AI Generated

DOI: Not provided

Keywords: large language models, machine psychology, reasoning, psychology, artificial intelligence, Science Advances, behavioral science, AI evaluation, cognitive assessment, AI bias, Review, Perspectives