Cake
  • Log In
  • Sign Up
    • The frustrating and scary thing is AlphaZero doesn't tell us how it came up with these insights.

      I wonder, is it because AZ has the opportunity to test (almost) every possible scenario that a human could never have the ability to test? Did Alpha have the opportunity to essentially run the largest trial and error test in history over the course of billions or trillions (?) of moves.

      I don't really understand the scale of which Alpha is able to simulate chess. " it examined only 60 thousand positions a second"...

    • There is the math you learn in grade school and high school and then there is THE MATH that spawns Alpha Zero.

      Barry Longyear, the author of the novella Enemy Mine, wrote about a future where a supercomputer has a world leader executed.

      When the United Nations demands a justification, printouts ten feet high and covering a football field are generated by the machine to explain all the factors that went into its decision.

      I often think of algorithms in terms of shortcuts. Where Stockfish calculates and rates 60 million possible moves, AlphaZero has realized that only about 1/100th of those possibilities is worth considering.

      How much time would it take humans to even review the data supporting these decisions?

      It would be interesting, nonetheless, to see the shortcuts that AlphaZero has come up with, even if we lack the time to manually validate them.