Exploring Human Behavior During Abstract Rule Inference and Problem Solving with the Cognitive Abstraction and Reasoning Corpus

Researchers introduced CogARC, a human-centered dataset derived from the Abstraction and Reasoning Corpus, to study cognitive strategies in abstract reasoning. In two experiments with 260 participants solving 75 visual puzzles, humans achieved ~90% and ~80% accuracy respectively. The study revealed that people successfully extract input-output rules from limited examples but employ diverse solution strategies on harder problems.