What algorithms power animal intelligence? How can we reproduce these algorithms in machines? To address these questions, I developed an interdisciplinary approach at the intersection of psychology, artificial intelligence, and video games. Using a video game engine, we raise newborn animals and ‘newborn’ artificial agents in the same virtual worlds, then test their behavior with the same tasks. Since the animals and agents are tested in the same video game environments, we can directly compare the learning abilities of brains and machines. In this talk, I will describe how my lab is using this approach to reverse engineer object perception and collective behavior. By modeling animals as “runnable” artificial agents raised in realistic virtual worlds, we can build engineering-level models of the brain’s learning mechanisms, capturing enough causal structure in the mechanisms to reproduce them in machines. These learning algorithms are an untapped goldmine for next-generation artificial intelligence systems and can serve as rigorous computational models for studying the origins and computational foundations of intelligence.