The Mind in the Machine - A Survey of Incorporating Psychological Theories in LLMs

Sep 17, 2026

Psychological insights have long shaped key NLP advances, including cognitive mechanisms behind attention, reinforcement learning, and Theory of Mind-inspired social modeling. As large language models (LLMs) scale in complexity, growing consensus holds that psychology is crucial for capturing human-like cognition, behavior, and interaction. This paper reviews how psychological theories inform LLM stages—data, pre-training, post-training, and evaluation and application—while surveying trends, gaps, tensions, and opportunities for integration.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages39

Reading Time1 h 38 min

AccessFree to download

Abstract

1 Introduction

2 Preprocessing

3 Pre-training

4 Post-training

5 Evaluation and application

6 Overarching questions

The Mind in the Machine: A Survey of Incorporating Psychological Theories in LLMs Zizhou Liu∗♠ , Ziwei Gong∗♠ , Lin Ai ∗♠ , Zheng Hui ∗†, Run Chen∗♠, Colin Wayne Leach⋄, Michelle R. Greene⋄, Julia Hirschberg♠ ♠ Department of Computer Science, Columbia University,⋄ Department of Psychology, Barnard College,†The Language Technology Lab, University of Cambridge ∗ Equal contributions. {sara.ziweigong, [lin.ai](lin.ai) , runchen, [julia](julia}@cs.columbia.edu)[}](julia}@cs.columbia.edu)[@cs.columbia.edu](julia}@cs.columbia.edu) ,{cleach, [mgreene](mgreene}@barnard.edu)[}](mgreene}@barnard.edu)[@barnard.edu](mgreene}@barnard.edu),{zl2889, [zh2483](zh2483}@columbia.edu)[}](zh2483}@columbia.edu)[@columbia.edu](zh2483}@columbia.edu) Abstract Psychological insights have long shaped pivotal NLP breakthroughs, including the cognitive underpinnings of attention mechanisms, formative reinforcement learning, and Theory of Mind-inspired social modeling. As Large Language Models (LLMs) continue to grow in scale and complexity, there is a rising consensus that psychology is essential for capturing human-like cognition, behavior, and interaction. This paper reviews how psychological theories can inform and enhance stages of LLM development, including data, pre-training, post-training, and evaluation&application. Our survey integrates insights from cognitive, developmental, behavioral, social, personality psychology, and psycholinguistics. Our analysis highlights current trends and gaps in how psychological theories are applied. By examining both cross-domain connections and points of tension, we aim to bridge disciplinary divides and promote more thoughtful integration of psychology into future NLP research. 1 Introduction As Large Language Models (LLMs) grow in scale and complexity, the Natural Language Processing (NLP) community is increasingly recognizing psychology as essential for capturing human-like cognition, behavior, and interaction (Qu et al., 2024; Lewis, 2025) . Psychology, grounded in empirically validated and computationally adaptable frameworks (Sartori & Orr, 2023; Ong, 2024), can address core LLM challenges such as reasoning fidelity, context retention, and user interaction. Reflecting these strengths, psychological insights have long shaped pivotal NLP advances, including the cognitive inspirations of attention mechanisms, formative reinforcement learning approaches, and Theory of Mind-inspired social modeling. Despite extensive multidisciplinary efforts, a holistic review systematically integrating psychology across the LLM lifecycle remains missing. Most surveys and position papers remain fragmented, typically falling into three broad categories: (1) Some investigate how LLMs can empower traditional psychology or cognitive science research, for instance by modeling human reasoning and behavior at scale (Demszky et al., 2023; Abdurahman et al., 2024; Ong, 2024; Ke et al., 2024) . (2) Others approach LLMs as subjects of psychological analysis, aiming to adapt or extend psychological theory – such as personality or cognition frameworks – to interpret and evaluate model behavior (Li et al., 2024b; Hagendorff et al., 2023; tse Huang et al., 2024; Pellert et al., 2024) . (3) Finally, a third group leverages a single or limited set of psychological constructs to enhance model alignment or multi-agent frameworks – improving system reliability, social interaction, and trustworthiness (Liu et al., 2023; Dong et al., 2024b) . This includes research on social influence for AI safety (Zeng et al., 2024a), moral reasoning in legal tasks (Almeida et al., 2024), and partial integrationsof social or developmental psychology (Sartori & Orr, 2023; Zhang et al., 2024c; SerapioGarca et al., 2025) . However, no existing work provides a unified map of how diverse psychological sub-areas can be harnessed, from data through application. Our survey fills this gap by offering a stage-wise view of how psychology can strengthen

FAQs

What problem does the survey address about psychology and LLMs?

A holistic, stage-wise review integrating psychology across the entire LLM lifecycle is missing; most existing work is fragmented across separate themes.

Which stages of LLM development does the paper review?

The paper discusses four key stages: preprocessing, pre-training, post-training, and evaluation and application.

What kinds of psychological theories are included?

The survey integrates theories from cognitive, developmental, behavioral, social, psycholinguistic, and personality psychology.

Sep 17, 2026

Uploaded by นรินทร์