CogGym is a large-scale, collaborative platform where artificial intelligence meets cognitive science. Our mission is to provide a unified benchmark for comparing human and machine intelligence, built from the most robust and insightful experiments in the history of cognitive science. We partner with leading research labs to curate and standardize hundreds of their foundational studies into a living, interactive library.
Choose an experiment to participate in
Alanqary et al. (2021)
Participants watch an agent navigate doors, collect keys, and pursue colored gems while deciding which gem is the intended target.
Bates et al. (2019)
Participants viewed images of virtual scenes with a cylindrical volume of liquid positioned above randomly generated obstacle courses and predicted wh...
Bates et al. (2019)
Participants viewed images of virtual scenes with liquid above obstacle courses (3 simple shapes: triangles, squares, circles) and predicted what perc...
Balaban & Ullman (2025)
Single-ball imagined-object tracking. Participants watch a ball move briefly, then imagine its motion and respond when it would hit the ground.
Balaban & Ullman (2025)
Two-ball imagined-object tracking. Participants imagine both balls' trajectories and respond when each would hit the ground.
Balaban & Ullman (2025)
Perceptual two-ball tracking. Participants respond when they see each ball hit the ground (no imagination phase).
Balaban & Ullman (2025)
Two-ball imagined tracking with strong grouping cues (both balls move in the same direction with hyperbolic paths).
Balaban & Ullman (2025)
Occlusion condition: balls pass behind a gray screen; participants imagine motion and respond when each ball hits the ground behind the occluder.
Balaban & Ullman (2025)
Minimal-physics condition: disks move in straight lines toward black side walls; participants imagine and respond when each disk hits its wall.
Balaban & Ullman (2025)
Noise/control task: participants click where they last saw each ball before it disappeared.
Balaban & Ullman (2025)
Motivation manipulation: same as Experiment 1b, with bonus incentives for accuracy.
Balaban & Ullman (2025)
Extended version of Experiment 1b with many more trials to probe fine-grained serial predictions.