All three opponents look at the same board. What changes is how far ahead
they look before they move.
01 · Easy
Mostly random
Easy picks a random empty square. About half the time it first checks
whether any square wins on the spot, and takes it if one does, so it can
still pounce. It never blocks you.
02 · Medium
Lily’s original
This is the rule-based AI from Lily’s Unit 10 “Play with AI” assignment,
kept rule for rule as she wrote it. It runs down five rules in order: win if it
can, block your winning move, take the center, take a random corner, take
a side. It only ever looks one move ahead, so it can’t see a fork coming:
start in a corner, answer its center with the opposite corner, then block
the corner it takes. That block gives you two ways to win, and it can only
stop one.
03 · Hard
Minimax · unbeatable
Hard searches ahead to the end of the game with minimax, assuming you
always find your best reply. Each finished game scores +10 for an AI
win, −10 for a loss and 0 for a draw, one point closer to zero for every
move it took, so quicker wins and slower losses rank higher.
Alpha-beta pruning skips branches that can’t change its decision, so the
search takes milliseconds. Perfect play from both sides is a draw, so
against Hard the best you can do is draw.
Reading the scores
With Show AI thinking on, every empty square shows the
minimax score of the next move going there, from the AI’s side: positive
means the AI can force a win, 0 a draw, negative a loss. On its turn the AI
is the maximizer, and Hard always plays a top score while Easy and Medium
may not. On your turn you are the minimizer: the lowest number is your best
reply.