GPT-4o Lacks Core Features of Theory of Mind

2026-02-12 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning · Depth: Expert, quick

Summary

A new evaluation framework assesses whether Large Language Models (LLMs) possess a Theory of Mind (ToM) by probing for a causal model of mental states and behavior. This research specifically investigates if LLMs have a coherent, domain-general, and consistent understanding of how mental states drive actions, independent of human-like ToM. The study found that while LLMs, including GPT-4o, can approximate human judgments in basic ToM scenarios, they fail at logically equivalent tasks and show low consistency between their predicted actions and inferred mental states. These results indicate that the observed social proficiency in LLMs does not stem from a domain-general or consistent ToM.

Key takeaway

For AI Researchers developing socially intelligent agents, you should critically re-evaluate current ToM benchmarks. Your models' apparent social proficiency may not reflect a true understanding of mental states, necessitating new evaluation methods that probe for consistent, causal models of behavior rather than just approximating human judgments. This shift is crucial for building truly robust and reliable AI systems.

Key insights

LLMs like GPT-4o lack a consistent, domain-general causal model of mental states and behavior, despite social task success.

Principles

• Social proficiency does not imply ToM.
• ToM requires causal mental state models.

Method

The framework tests LLMs for a coherent, domain-general, and consistent model of how mental states cause behavior, using logically equivalent tasks and consistency checks between action predictions and mental state inferences.

In practice

• Evaluate LLMs beyond simple benchmarks.
• Focus on causal models for ToM assessment.

Topics

• Theory of Mind
• Large Language Models
• GPT-4o
• AI Evaluation
• Social Cognition

Best for: AI Researcher, AI Scientist, Research Scientist

Related on AIssential

• Do Large Language Models Possess a Theory of Mind? A Comparative Evaluation Using the Strange Stories Paradigm — GPT-4o demonstrates human-comparable Theory of Mind capabilities in text-based evaluations, outperforming smaller LLMs.
• A Causal Model of Theory of Mind in Conflict for Artificial Intelligence — A causal model defines situational and agent-level conditions for AI's resource-rational Theory of Mind engagement in conflict.
• "GPT-5.4 HIGH" Cheating? Can it Reason or just Write Code? — GPT 5.4 High demonstrates strong causal reasoning, potentially by converting linguistic problems into mathematical code for optimal solutions.
• Does Theory of Mind Improvement Really Benefit Human-AI Interactions? Empirical Findings from Interactive Evaluations — Static ToM benchmarks do not predict LLM performance in dynamic human-AI interactions.
• New ChatGPT Model & Memory Features Explained (AI News You Can Use) — Goal-based prompting with GPT 5.5 models yields more accurate results with reduced hallucinations.
• AI and Theory of Mind: an interview with Nitay Alon — Theory of Mind is critical for navigating deception, but over-mentalization can lead to paranoia in both humans and AI.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.