In a blog post announcing its latest innovations, DeepMind showed off its MuZero machine learning AI that can play multiple different games and score record-breaking scores without being told the rules. Learning from his previous moves, MuZero can strategize while playing in a completely unknown environment.
Systems using look-ahead search, such as AlphaZero, have had remarkable success in classic games such as checkers, chess, and poker, but they rely on knowing the game rules and being informed about the dynamics of their environment. Therefore, they have difficulty in solving problems that they encounter in real life, which are often complex.
MuZero currently plays Ms. Pac-Man, Go, chess, shogi, but such advances in AI could have a positive impact in enabling people to solve adaptive algorithms without rule sets, a daily challenge.
Artificial intelligence uses 3 different parameters when creating the game strategy:
How good is the current location?
What is the best action to take next?
How successful was the last action?
Essentially, artificial intelligence simplifies the whole game into different questions and then determines how to proceed. He is constantly learning throughout the game to make these decisions, and the results can be extremely impressive.
Such algorithms are likely to become integral to building robots that can deal with the real world rather than playing predefined roles with limited flexibility.
15027
