Dreamer 4: the Artificial Intelligence Agent Who Learns without Practicing in the Real World

WWWhat's new

Factuality

Ownership

For years, artificial intelligence has proven to be able to overcome humans in games such as Go, chess or Atari. However, these achievements have been conditioned by a reinforcement learning model based on trial and intensive error, which requires millions of interactions for an agent to acquire acceptable skills. This approach, although successful in digital environments, is unworkable in the physical world, where experiencing can be costly, sl…