An artificial intelligence system can now autonomously propose and test psychological theories, closing the loop on theory-building in cognitive science. The system, called the Automated Cognitive Scientist (AutoCog), uses large language models to generate competing executable models of decision-making, design experiments that distinguish them, collect data from human participants, and synthesize improved theories.
The Research
Developed by researchers at Princeton University and the Max Planck Institute for Biological Cybernetics, AutoCog was tested in the domain of multi-cue decision-making. The AI began with established theories (e.g., equal-weight, take-the-best) and then autonomously ran cycles of theory generation, experiment design, data collection (from online participants), and theory revision. When seeded with simulated data, AutoCog recovered known decision-making strategies, including unconventional ones, showing its discoveries are data-driven. In two experiments with human participants (total N = 340), AutoCog produced theories that outperformed the seeding theories and generalized to held-out experimental settings. It also discovered a novel theory: that choices show diminishing sensitivity to feature values. This prediction was confirmed in a preregistered study with 150 new participants.
Why It Matters
This breakthrough could accelerate cognitive science by automating the slow, manual process of theory-building. Instead of researchers spending years refining models, AutoCog can rapidly iterate through many hypotheses. For the public, this means faster progress in understanding how we make decisions—with potential applications in education, AI alignment, and personalized brain training. The system’s success also demonstrates that AI can contribute creatively to scientific discovery, not just crunch numbers.
What You Can Do
Stay curious about how your mind works. Engage with cognitive science findings by taking evidence-based brain tests like IQ assessments and training tasks that challenge decision-making. The more we learn about decision strategies, the better we can design tools to sharpen your thinking.
Source: arXiv q-bio.NC
Curious about your own brain? Take our free adaptive IQ test or try 306 brain training levels.
