At the moment, the AI evaluates the position based on how many moves each side can make. This makes it have positional awareness, but it sometimes neglects material gains/losses, because they seem unimportant in terms of possible current moves. This problem is caused by the limited look-ahead depth of around 4 currently possible - a rook stuck in a corner might be worth picking up, even though it may only be fully active in 5 or 6 moves - but the AI does not see far enough to understand this.
As the move evaluation strategy works better once less pieces are on the board and there are more spaces available for your pieces, this strategy is fine for the later stages of the game.
A different strategy could be used for the early game, however, maybe based on a material evaluation function with weighted squares, to mitigate the problems described above.
At the moment, the AI evaluates the position based on how many moves each side can make. This makes it have positional awareness, but it sometimes neglects material gains/losses, because they seem unimportant in terms of possible current moves. This problem is caused by the limited look-ahead depth of around 4 currently possible - a rook stuck in a corner might be worth picking up, even though it may only be fully active in 5 or 6 moves - but the AI does not see far enough to understand this.
As the move evaluation strategy works better once less pieces are on the board and there are more spaces available for your pieces, this strategy is fine for the later stages of the game.
A different strategy could be used for the early game, however, maybe based on a material evaluation function with weighted squares, to mitigate the problems described above.