On Games -- and Play | ARC Prize @ MIT
Summary
Laura Schulz, a professor at MIT's Brain and Cognitive Sciences Department, challenges the prevailing notion that efficiently acquiring new skills, often studied through games, is the most distinctive feature of human intelligence. Drawing on 20 years of research into play, Schulz highlights the fundamental difference between goal-directed, efficient actions in games and the often inefficient, self-generated, and exploratory nature of play in children. She presents examples where children deliberately engage in costly, non-goal-oriented actions, even inventing new tools or re-defining objectives, which computational models of learning struggle to capture. Schulz argues that while games are valuable for understanding learning, they risk overlooking the uniquely human capacity to invent arbitrary problems and objective functions for no immediate external reward, suggesting this creative problem-generation is the true core of human learning.
Key takeaway
For AI scientists and research scientists developing models of intelligence, you should consider shifting focus from optimizing for predefined objective functions to exploring mechanisms for generating novel problems and goals. Your models could benefit from incorporating the capacity for self-initiated, seemingly inefficient exploration, as this creative problem-setting, rather than just efficient problem-solving, may be central to advanced cognitive capabilities and robust learning systems.
Key insights
Human intelligence is uniquely defined by the capacity to invent arbitrary problems and objective functions, not just efficient skill acquisition.
Principles
- Play often involves inefficient, self-generated actions.
- Human learning includes creating new objective functions.
- Games simplify intelligence, risking "looking under the lamp post".
Method
Schulz's research method involves observing children's spontaneous play behaviors in open-ended environments, contrasting them with goal-directed tasks to highlight the distinction between efficient action and exploratory, self-motivated activity.
In practice
- Design AI systems that invent novel problems.
- Study learning in open-ended, unconstrained environments.
Topics
- Human Intelligence
- Early Childhood Cognition
- Games vs. Play
- Objective Functions
- Cognitive Development
Best for: AI Scientist, Research Scientist
Related on AIssential
Editorial summary, takeaway, and curation by AIssential. Original article published by ARC Prize.