Commonsense Psychology in Minds and Machines

Commonsense Psychology in Minds and Machines

by Moira R. Dillon

Humanities, Social Sciences & Thought Science — General & History GTInterdisciplinary studiesGTKCognitive studies
🎙 Moira R. Dillon 👥 15K 📅 June 9, 2026 ⏱ 40 min 👁 64 📄 expert opinion 🧭 2026-08-15
Available in: English (current) Français

Keywords

commonsense psychologycore knowledgeinfant cognitionAItheory of mind

Summary

In this talk, Moira R. Dillon, Assistant Professor of Psychology at NYU, explores the differences between human and artificial intelligence, focusing on commonsense psychology. She argues that human intelligence is characterized by core knowledge—innate, universal, and pre-linguistic knowledge about people, places, and things—which is absent in current AI systems. She contrasts the scope and limits of human cognition with AI’s goal of unlimited scope, and highlights that AI, trained on vast language corpora, lacks the foundational commonsense that infants possess. Dillon presents experimental evidence from her lab showing that 11-month-old infants infer goals from simple geometric shapes, similar to adults in the classic Heider-Simmel study, demonstrating the abstractness of core social knowledge. She contrasts this with AI’s tendency to rely on observable behavioral predictions, which fails in novel scenarios. The talk concludes that human and machine intelligence are fundamentally different, and that building human-like AI requires incorporating core knowledge, a challenge for current AI research.

155 words

Critical Evaluation

Value of the Information & Strength of the Argument

The talk provides valuable insights into the distinction between human and artificial intelligence, emphasizing the role of core knowledge. Dillon’s argument is well-supported by references to established research (e.g., Heider & Simmel, core knowledge theory) and current AI limitations (e.g., Apple’s paper, quotes from Melanie Mitchell and Ernest Davis). She effectively uses examples and analogies (e.g., the excavator vs. human digging) to illustrate her points. The argumentation is coherent and persuasive, though it presents a specific perspective rather than a balanced debate.

Scientific Rigor, Source Quality, Title Accuracy

The talk demonstrates scientific rigor by grounding claims in empirical research and citing relevant literature. Dillon references the Heider-Simmel study, core knowledge research, and quotes from AI researchers like Melanie Mitchell and Ernest Davis. The title accurately reflects the content, which compares human commonsense psychology with AI. The talk is well-structured and the sources are credible, though some claims (e.g., the Apple paper) are presented without full context. No comments were provided for analysis.

171 words

Title / Content Match

The title accurately reflects the content, which compares human commonsense psychology with AI capabilities.

Quality & Reliability

8/10

The talk is delivered by a recognized developmental cognitive scientist, grounded in established research (Heider & Simmel, core knowledge theory) and includes references to current AI limitations (Apple paper, Mitchell, Davis & Marcus). The content is well-structured and scientifically informed, though it presents a particular viewpoint rather than a systematic review.

Key Moments

Markers derived by PSI from the transcript: the creator did not define chapters.

Cited Sources

Concurring Sources

  • Core knowledge theory — Supports the concept of core knowledge as innate and universal.
  • Heider-Simmel demonstration — The study referenced in the talk showing adults attribute social meaning to shapes.

Dissenting Sources

  • Apple's paper on LLM limitations — The talk cites Apple's paper claiming AI gives the illusion of thinking, but this is a specific finding that may be debated in the AI community.

Contribution & Novelties

The talk offers a clear synthesis of research on core knowledge and its implications for AI, highlighting the fundamental differences between human and machine intelligence. It provides a compelling argument that current AI lacks commonsense psychology, which is essential for human-like understanding. The talk bridges developmental psychology and AI research, suggesting that incorporating core knowledge into AI is a key challenge.

Pour aller plus loin :

  • Core knowledge theory — Overview of the theory proposed by Elizabeth Spelke.
  • Heider-Simmel demonstration — The classic 1944 study on attribution of social meaning to geometric shapes.
  • Theory of mind — The ability to attribute mental states to others, relevant to commonsense psychology.

109 words

Radar Profile

The radar profile shows high scores in information quantity, quality, and reliability, with a slightly lower technical level, indicating a well-balanced and accessible scientific talk.

Reliability 8/10