A formal definition and meta-model for a machine theory of mind

2026-06-02 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning · Depth: Expert, quick

Summary

This paper, published on 2026-06-02, introduces the first rigorous formal definition of Machine Theory of Mind (MToM). This definition is grounded in principles supported by evidence from cognitive psychology, neuroscience, and artificial intelligence. The work uses this new definition to analyze current state-of-the-art efforts in the field, aiming to establish a potential research agenda to address the MToM problem. Furthermore, the paper advances a general holistic meta-model for MToM and evaluates the current landscape of empirically benchmarking such models, providing a comprehensive framework for future development and assessment.

Key takeaway

For AI Scientists focused on developing advanced cognitive AI, this formal definition and meta-model for Machine Theory of Mind provides a crucial framework. You should use this rigorous foundation to evaluate existing models and structure future research, ensuring your work aligns with evidence-based principles from psychology and neuroscience. This approach can help you design more robust MToM systems and develop more effective benchmarking strategies.

Key insights

This paper formally defines Machine Theory of Mind and proposes a meta-model to guide research and benchmarking efforts.

Principles

• MToM definition based on cognitive psychology.
• Integrates neuroscience and AI principles.

Method

The paper proposes a rigorous formal definition and a general holistic meta-model for Machine Theory of Mind, then uses it to examine current efforts and benchmarking.

In practice

• Examines state-of-the-art MToM efforts.
• Reviews empirical benchmarking of MToM models.
• Drives a research agenda for MToM.

Topics

• Machine Theory of Mind
• Cognitive AI
• Formal Definitions
• Meta-modeling
• AI Benchmarking
• Neuroscience in AI

Best for: Research Scientist, AI Scientist

Related on AIssential

• Connecting Language and (Artificial) Intelligence: Princeton’s Tom Griffiths — Human and artificial intelligence can be understood through three complementary mathematical frameworks: rules/symbols, neural networks, and Bayesian probability.
• Benchmarking LLMs at the Game Of Science (Eleusis) — LLM scientific reasoning involves distinct inductive ability and metacognition, measurable through iterative hypothesis testing.
• Position: Artificial Intelligence Needs Meta Intelligence -- the Case for Metacognitive AI — Metacognition can make AI more accurate, secure, and efficient by enabling self-monitoring and resource allocation.
• Stabilising Generative Models of Attitude Change — Generative actor-based modeling can formalize verbal psychological theories into executable, stable simulations.
• Why Every Brain Metaphor in History Has Been Wrong [SPECIAL EDITION] — Scientific models are useful simplifications, not necessarily literal truths about underlying reality.
• MEDLEY-BENCH: Scale Buys Evaluation but Not Control in AI Metacognition — AI models exhibit a dissociation between metacognitive evaluation and control abilities, with evaluation improving with scale but control not.

Open in AIssential →

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