A Faster Minimax
Minimax has been the dominant paradigm for many board games for so long, able to deliver either good or perfect solutions to many game problems. Such is this success that in many primary areas of human game intellect the best play is based on minimax-based AI. However, minimax has its limitations, and in some domains with great combinatorial complexity, minimax has been replaced by Monte Carlo tree search (MCTS) as the preferred solution. Where unscripted MCTS is rolling out to terminal nodes (the expected case), it sadly depends on spending most of its time assessing junk variations in the simulation phase in order to finally pick a “best” move. Note that in this case, ...