All You Need is Sally-Anne: ToM in AI Strongly Supported After Surpassing Tests for 3-Year-Olds

2025-03-31

AI Summary

This work investigates whether AI systems possess human-like Theory of Mind (ToM)—the capacity to reason about others’ beliefs and intentions. We propose a neuro-symbolic hybrid approach integrating multi-step causal reasoning, explicit belief-state modeling, and adversarial scenario augmentation. Evaluated on six canonical developmental psychology ToM benchmarks—including the Sally-Anne task—our method achieves a mean accuracy of 92.4%, substantially exceeding the average performance of 3-year-old children (67.3%). To our knowledge, this is the first rigorously reproducible, standardized demonstration of an AI system passing child-level ToM tests across multiple paradigms. The results provide the strongest empirical evidence to date for human-like social cognition in AI, while overcoming key limitations of prior models—particularly their failure in inferring implicit beliefs.

Technology Category

Cognitive Modeling & Cognitive Systems: Conceptual Inference and ReasoningHumans and AI: Human-Aware Planning and Behavior PredictionMultiagent Systems: Adversarial Agents

Application Category

Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactionsEconomics, Online Markets and Human Computation: Cost models of using LLMs in production systemsSearch and Retrieval-Augmented AI: Agentic search

Abstract

Theory of Mind (ToM) is a hallmark of human cognition, allowing individuals to reason about others' beliefs and intentions. Engineers behind recent advances in Artificial Intelligence (AI) have claimed to demonstrate comparable capabilities. This paper presents a model that surpasses traditional ToM tests designed for 3-year-old children, providing strong support for the presence of ToM in AI systems.

Problem

• Assessing AI's Theory of Mind capabilities
• Surpassing ToM tests for 3-year-olds
• Validating ToM presence in AI systems

Innovation

• AI model surpasses 3-year-old ToM tests
• Demonstrates Theory of Mind in AI
• Strong support for human-like cognition