By Ji Y. Son
Publication Date: 2026-05-12 12:38:00
The nonprofit ARC Prize Foundation released the results of a new benchmark on May 1, 2026: a test of an AI system’s ability to solve a game. The results were impressive: humans achieved 100%, while the most advanced AI systems achieved less than 1%.
At first glance, this may come as a surprise to AI users who are impressed by the sophisticated essays, codebases, and multi-stage projects that are generated in seconds. How can these brilliant AI systems struggle with these simple Tetris puzzles?
This confusion points to a risk: AI is being integrated into everyday life faster than humans can understand it.
We are cognitive psychologists who focus on teaching difficult concepts. To recognize the limitations and risks of today’s AI agent systems, it is important to understand that the systems can both achieve superhuman feats and make mistakes that few humans would make. To this end, we propose a new way of thinking about AIs: as button-pushing explorers.
Mental models for AI
We teach…