![[2602.22408] Exploring Human Behavior During Abstract Rule Inference and Problem Solving with the Cognitive Abstraction and Reasoning Corpus](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.22408] Exploring Human Behavior During Abstract Rule Inference and Problem Solving with the Cognitive Abstraction and Reasoning Corpus
Summary
This article presents the Cognitive Abstraction and Reasoning Corpus (CogARC), a study exploring human abstract reasoning through problem-solving tasks, revealing insights into cognitive strategies and performance variations.
Why It Matters
Understanding human behavior in abstract reasoning is crucial for advancing artificial intelligence. The findings from this study provide valuable insights into cognitive strategies, which can inform the development of AI systems that mimic human-like reasoning and adaptability.
Key Takeaways
- CogARC is designed to study human abstract reasoning and problem-solving.
- Participants demonstrated high accuracy but varied significantly in performance across tasks.
- Longer deliberation times were observed for more complex problems, indicating cognitive effort.
- Familiarity with task structure led to quicker responses but slightly decreased accuracy over time.
- The study highlights the diverse strategies humans use to generalize and adapt under uncertainty.
Computer Science > Artificial Intelligence arXiv:2602.22408 (cs) [Submitted on 25 Feb 2026] Title:Exploring Human Behavior During Abstract Rule Inference and Problem Solving with the Cognitive Abstraction and Reasoning Corpus Authors:Caroline Ahn, Quan Do, Leah Bakst, Michael P. Pascale, Joseph T. McGuire, Michael E. Hasselmo, Chantal E. Stern View a PDF of the paper titled Exploring Human Behavior During Abstract Rule Inference and Problem Solving with the Cognitive Abstraction and Reasoning Corpus, by Caroline Ahn and 6 other authors View PDF HTML (experimental) Abstract:Humans exhibit remarkable flexibility in abstract reasoning, and can rapidly learn and apply rules from sparse examples. To investigate the cognitive strategies underlying this ability, we introduce the Cognitive Abstraction and Reasoning Corpus (CogARC), a diverse human-adapted subset of the Abstraction and Reasoning Corpus (ARC) which was originally developed to benchmark abstract reasoning in artificial intelligence. Across two experiments, CogARC was administered to a total of 260 human participants who freely generated solutions to 75 abstract visual reasoning problems. Success required inferring input-output rules from a small number of examples to transform the test input into one correct test output. Participants' behavior was recorded at high temporal resolution, including example viewing, edit sequences, and multi-attempt submissions. Participants were generally successful (mean accuracy ~90% f...
