These smart bricks know what object they make up

Researchers at IT University of Copenhagen and Sakana AI created simulated cubes with independent neural networks that, by exchanging messages with neighbors and using only local information, achieved 85 to 100 percent accuracy in identifying their assembled object category across seven types—boats, cars, chairs, guitars, houses, planes, and tables—and could even regenerate missing cubes and identify objects with up to 15 percent of components removed.

This research is simulation-based; physical implementation remains experimental.