DeepMind’s latest AI program achieves “superhuman performance” in tasks without following rules.
Like the research center’s earlier AI agents, MuZero mastered dozens of old Atari video games, chess, and Asian board games like Go and Shogi.
But unlike its predecessors, it had to make up its own rules.
It is currently being put to practical use to find a new way to encode videos that could drastically cut YouTube’s costs.
“The real world is messy and complex, and no one can give us rules for how it works,” DeepMind lead researcher David Silver told the BBC.
“And yet humans are able to come up with plans and strategies for what to do next.
“For the first time we really have a system that can understand on its own how the world works, and use that understanding to do the kind of complex forward planning that we’ve seen before in games like chess.
“It can create something from nothing, discover the rules of the world through trial and error, and use those rules to achieve superhuman performance.”
