Appendix F: Ideas for Future Exploration

Appendix F: Ideas for Future Exploration

For various reasons, the following ideas did not make the "first cut" for inclusion in the proposed evaluation function. Perhaps these ideas need more refinement. Maybe someone will find a clever way to implement them in the future.

Idea number 1: Pieces can be rewarded for their ability to perform tactical tricks like pins, forks and skewers 1, 2 or 3 moves into the future. This can be determined from information extracted from the "future mobility" database for each position examined. Testing will be required to see if a particular tactical trick is worth including in the evaluation function. [Upon reflection, this concept might need further work, as the performance benefits obtained might not outweigh the loss of search depth required to implement.]

Consider the following position:

A traditional chess engine, when asked to evaluate this position, will give pieces bonuses based on the square it occupies and possibly for the number of moves it can make.

Using the proposed heuristic, our computer will "know" what each piece can do 3 moves into the future. Therefore, it will "know" that the Bishop can move Be3-d4-a7 and therefore is 1 move away from pinning the Knight to the King. It will give the Bishop a bonus for this, and an even larger bonus once it moves to e3, since it will then be performing the pin.

A traditional chess engine will figure this out, but only after a few additional "plys" of search. This will include perhaps many other positions that should not have to be looked at.

A computer using the heuristic proposed in this paper uses "knowledge" of future moves to focus its search. The positions that are looked at are fewer, but are more likely to be related to the main line of play. This also happens to be similar to how a human plays chess.

Idea number 2: We can look at these maps of potential moves from our mobility database and decide on a general strategy of Kingside attack, Queenside attack, or whole board attack. The move maps we have generated will tell us if we have an exposed king, weak pawns, or other “triggers” that might indicate such an attack will be to our advantage.

We can reward pieces for having potential moves that fit this plan - for example, if a Kingside attack is decided on, a piece is given a bonus if it can make moves to reach squares on that side of the board and deep in the enemy’s territory.

Idea number 3: In our evaluation function we can define certain squares in the position being examined to be "of interest" and we can reward our pieces extra if somehow they can reach this square. For example, our opponent's queen and rook might be positioned so that it is possible to "fork" them (perform a simultaneous attack on both pieces) with our knight from a certain square. We would first locate these squares, if they exist on the board, then give our knight a bonus based on it's distance (in moves) from this specific square.

Idea Number 4: Michael Stean (Simple Chess) and Carsten Hansen (Improve Your Positional Chess) have written about outpost squares and their importance in a chess game. Pieces placed on outpost squares cannot easily be dislodged and can form a springboard for attack or for creating positional advantage.

Perhaps we can locate outpost squares in the current position we are evaluating by looking at our opponent's pawn mobility 3 moves into the future. If we sum together all squares that the pawns can attack (and eliminate these from consideration), we can quickly identify (from the squares that remain) the 'outpost squares' (optimal squares for our pieces that cannot easily be dislodged) and reward our pieces if somehow they can reach these squares.

Idea Number 5: Improving your worst placed piece

Trainer Mark Dvoretsky has said, "In positions of strategic manoeuvering (where time is not of decisive importance) seek the worst placed piece. Activating that piece is often the most reliable way of improving your position as a whole."

A great idea. In certain positions we might score the individual pieces and find the worst placed piece. We might spend more time in our searching, analyzing ways to better place this piece on the board and so improve our position.