Draughts Against the Computer
A game of draughts against computer opposition is only as interesting as what the program does when it has to choose, so here is what this one does. It has no opening book, no database of past games and nothing memorised at all. From the position in front of it, it plays out every legal continuation to a fixed depth, scores what it finds and picks the best line, and the three buttons under the board set how deep that goes.
How far ahead each strength looks
Gentle looks two half-moves ahead: its own move, and your answer to it. Steady looks four, so two moves each. Tough looks six. Those are small numbers and saying them out loud is the point, because they tell you exactly what beats each setting. A trap that only pays off on your third move is invisible to Gentle and to Steady, and Tough will see it coming. Nothing here plays a strength it does not have. The buttons change the depth of the search and nothing else, and you can press a different one mid-game without disturbing the position.
It keeps looking while the captures are forced
A program that stops searching at a fixed depth has a famous weakness. If the depth runs out in the middle of an exchange it counts the pieces standing at that instant and believes a total that is about to change, so it walks into a trade that looks level on the fourth move and is a catastrophe on the fifth. The repair here rides on the compulsory-capture rule. When the search reaches its limit in a position where a capture is forced, it does not stop; it carries on until the forcing has run out, up to six extra half-moves. The continuation is compulsory, so there is barely anything to choose between and very little to compute, and the opponent stops giving away pieces one move past the end of its own vision.
What it thinks a piece is worth
Its scoring is deliberately plain, and knowing it tells you how to mislead it. A man is worth a hundred and a king a hundred and seventy-five, so about one and three-quarter men: it will spend a man and a bit on a crowning and count that a good afternoon. Every row a man has advanced adds five, so it pushes rather than shuffles. A man still sitting on its own back row is worth eight more than one that has left, because a hole in your back row is where the other side gets its crown. And it scores the two outside files lowest of the lot. A piece on the edge can never be jumped, which sounds like safety, but it is also a piece with one diagonal instead of two, and this opponent would rather be in the middle.
Perfect play is a draw, and that is not an opinion
The eight-by-eight game with these rules has been solved. In 2007 a research team published a proof that neither side can force a win from the standard opening position: with best play on both sides it is a draw, established over a search space of five hundred million million million positions. Nothing on this page is that program, and it is not close. The result is still the useful thing to carry around, because it means a draw against a strong opponent is the correct outcome rather than a failure of nerve, and because it settles the question of whether the player who moves first has been handed something.
Where a person still wins
The opponent cannot see past its own depth and has no plan beyond it, so anything longer than the search beats it. A build-up that costs you material for four moves and returns three pieces on the fifth is invisible while you are assembling it, and by the time it has come inside the horizon it is already too late to prevent. Long forcing sequences are your best weapon for exactly the reason they help the machine: a forced reply is a reply you can calculate precisely, and you are allowed to count further ahead than six half-moves.