What is a matchbox computer?
See here for more information.
1. Play a game with a human using the steps below:
a. The human plays a move and takes the matchbox that has his move printed on it.
b. There will be several beads in the box. The human will take a random bead and move the pawn according
to the corresponding move.
c. Repeat with the matchboxes that corresponds to every move
-If the computer wins then put the beads back in the box
-If the computer loses then take the beads out.
2. Keep repeating this, and after some time, the computer will be able to play a perfect game.
The theory behind this is that the computer will "learn" to play a perfect game by eliminating all the possibilities that lead to a loss. This algorithm would not work for things like categorizing images or identifying human faces, but it did work for AlphaZero and AlphaGo, which learned to play the complex games of chess and go by just playing games against itself and learning from its mistakes.
]]>