In comparing brain scans of humans playing classic Atari video games to artificial intelligence (AI) networks trained to play the same games, California Institute of Technology (Caltech) researchers found the AI's “neurons” behaved similarly to neurons in the human brain.
The Caltech team trained DeepMind's Deep Q Network (DQN) agent to play "Pong," "Space Invaders," and "Enduro" (a racing game), then used its artificial neurons to predict behavior and brain activity in functional magnetic resonance imaging brain scans of human players.
Caltech's John O'Doherty said, "If we can understand why and how the brain can solve these games much more efficiently compared to an AI, this may help guide the development of smarter and more humanlike AI algorithms in the future."
From California Institute of Technology
View Full Article
Abstracts Copyright © 2021 SmithBucklin, Washington, DC, USA