Human Centered Can AI Take Common Sense from a Baby?
Apr 30, 2025
In this engaging discussion, David Moore, a developmental scientist and psychology professor emeritus at Pitzer College, dives into the intriguing realm of common sense in artificial intelligence. He examines how current AI models fall short compared to infant cognition in generalizing from limited experiences. The conversation highlights DARPA's Machine Common Sense program, explores the importance of early language development, and contrasts human learning with machine learning. Moore also touches on the prospects of merging developmental science with AI to enhance its capabilities.
AI Snips
Chapters
Books
Transcript
Episode notes
Bringing Developmental Science To AI
- DARPA convened developmental scientists with AI researchers to emulate how children acquire common sense.
- The evaluation team created rigorously controlled tests inspired by infant experiments to measure machine common sense.
Psychologist Designing AI Baby Tests
- Moore's evaluation role was to design artificial worlds and infant-style experiments for the AI agents.
- He guided coders on test scenarios but did not build the agents himself.
Generalization Is The Core Test
- The program converged on one core requirement: common sense must generalize to novel situations without extra training.
- Teams agreed generalization was central even though 'common sense' remained fuzzy and debated.
