To really put the chess analogy in perspective, consider this:
Background - most AI algorithyms basically come down to selective brute force. You try a move, see what the result of it is, then pretend to be the opponent and see what move they should make, and so on, hopefully taking the shortcut of not continuing too far down paths that don't look likely to be beneficial. This all relies on a few things - a good heuristic to score a game position, and a way to quickly try out all possible future moves from a position (usually this boils down to the heuristic being fast to compute, and the number of moves being small). WIth this in mind, consider the following games:
In checkers, there is a very low branching factor, as there are generally under 10 valid moves in any board position. A prograsm has been written that plays a "perfect" end game of checkers - ie, always makes the best move. The end game of checkers is thus considered a solved problem (in the same way as to most people tic-tac-toe is a solved problem ... two competent players should always draw against each other). It's basically jsut database lookups, nothing exciting.
In a typical chess game, the branching factor is relatively "low" - in any given position, there are usually only 10 - 20 valid moves. At this sort of level, a very good computer (eg deep blue) can compete against and beat the best humans, but it's pretty borderline. The heuristics involved here usually involve valuing each piece, and also to an extent board position.
In a typical game of go (which uses a 19x19 board unless i'm mistaken), there is a large branching factor, although I'm not familiar enough with the game to hazard a guess at what it might be. Suffice to say it's much higer than chess. Computer go players
suck versus good huamn players ... they dont even come close to winning.
In a huge map game of civ 3, the board is 180x180, although this is slightly misleading, as the tiles are marked such that only (even, even) or (odd, odd) pairs can exist, thus halving the number of tiles. Even so, the board is huge in comparison to the other games. Added to this, there are up to 15 AI players, each of whom might control hundreds of pieces (units). Unlike other games under discussion, the AI has the oppourtunity to move most or all of these each turn, not just one, further increasing the branching factor. Additionally, with railroads, some moves are ZERO cost, making the potential moves for each unit pretty huge! The AI is expected to move all these units as part of a coherant strategy, as well as managing cties in such a way that enhance their unit strategy, as well as trading effectively, as well as potentially persuing any number of other short, mediam and long term goals (culture, tech research, etc). Even coming up with a good heuristic to rate a board position is hard.Let alone making it something that is simply computed. Plus the huge number of possible combinations (ie high branching factor) makes normal brute force methods impractical.
To write an AI that is even remotely optimal at such a game would not only take years to write, it would take years to run per turn in the mid to late game.
Firaxis have done a pretty damn good job making the AI seem as smart as it does, but basically it's very linear and reacts pretty simply to situations. It doesn't figure out what would be "best", it simply sees one thing happen and performs whatever set response(s) it has been programmed to do. Hence you see silly things like being able to move your units back and forth and having it chase them from city to city, never getting in range to attack (which of course is supposed to be fixed in the patch). Until we made some pretty fundamental breakthroughs in AI research (quantum computing maybe? I can't see how to make it managable on traditional computing systems), we can expect the AI will always be more artificial than intelligent.
All this is a good argument for why we need MP, imo.
