There's a really fun toy example of genetic algorithms that uses this compact representation of tictactoe gamestate (encoding each game into a concatenated list of ternary states, of which there are 19,682 possible values if completely unoptimized):

Consider a "genome" for a tictactoe player to be a 19,682-long array that holds "next moves" at each slot in the array. For any given gamestate, look up the player's next move via their genome.

Randomize genomes originally. Make tons of them. Compete w/ each other, then play around with all sorts of fun mating/mutation strategies for swapping different moves between genomes.

Fun to see a perfect tictactoe player evolve, super approachable toy example.