ericbabe
Posts: 11927
Joined: 3/23/2005 Status: offline
|
quote:
ORIGINAL: Arinvald I guess that is the breakdown; those of us that got started as boardgamers and these youngsters who have been spoiled with fancy graphics. I think that graphics have progressed to the point that they are no longer an issue. Hell, I was very happy with V4V/W@W graphic levels and those games are almost fifteen years old now. So I agree, the next holy grail for computer wargaming is in the direction of AI. The more options a player has, the more difficult it is to program good AI. The difficulty actually becomes exponential -- literally -- with the number of choices a player has. Would you and other Matrix players prefer games with simpler rule-sets if they had more challenging AI? Let me provide some benchmarks... Consider chess: on a player's turn he can move one piece of the 16 pieces that he has, and each piece can move to one of maybe, by way of estimation, 10 sensible locations. That's only 160 options per turn. It's possible on a PC to make decent AI for chess -- nothing that a really skilled chess player couldn't beat -- but decent enough to give most chess players a good workout. This is due in large part to the fact that decent AI routines for chess use libraries of opening moves that have been worked out by human chess grandmasters -- if you've ever played chess against an AI that doesn't use libraries you'll know what I mean. A good algorithm to beat a chess *grandmaster* requires networked supercomputers, but most players aren't grand masters. Now consider Go: Go is played on a 19x19 grid and players place one stone each turn. That's about 200 choices per turn -- more choices at the beginning of a game when choices are more crucial, fewer as the game progresses. Yet Go, it turns out, has so many options that it is difficult to write an algorithm to run on a PC or even a supercomputer that will be of a challenge to any player who is not a complete novice. Go doesn't have that many more sensible options per turn than chess, yet because of the exponential relationship between choices/complexity the small increase in options results in a game system for which it is very, very difficult to write good AI. There are computer scientists who have spent 30 years of their career developing algorithms for Go, yet the best algorithm for Go as of 2001 only ranks a 10-kyu in the Go ranking system (the ranking given to weak amateurs.) Some good Go AI has been written ... for a variation of the game played on a 9x9 grid, which brings the complexity down to an average of 40 choices per turn (or about 80 choices during the crucial opening game.) Now imagine a game rather like chess but with a random starting configuration of perhaps 30 pieces per side played not on an 8x8 featureless grid but on a 40x40 map with forests, heights, roads and rivers. In addition to moving, allow pieces to change formation (which effectively changes the entire map for them), fire at any hex in range, or charge at many points along their movement paths. Instead of moving one piece per turn, you can move all the members of your corps with certainty, and may be able to move more before the other player has a chance at a move. Throw in fog and rain and resupply, proximity morale rules and reinforcement. I don't even know how to estimate the number of choices available to players per turn but it's a lot higher than there are per turn in a game of Go. The absence of a really killer AI for wargames is not a matter of developers' deciding AI isn't important; there are mathematical impediments fundamental to the reality of the situation. One reason that RTS games are ubiquitous, I believe, is that it is not necessary to have a good AI in RTS games. The computer can process information faster and can "click" on its units at millisecond intervals. Playing a good game in RTS often comes down to having good mouse/keyboard skills and being able to click on things very fast -- learning all the keyboard shortcuts to maximize the orders/second that one can issue. To make RTS more challenging, the program just has to give players more things to click on, more enemies to consider, in the same constraint of time. Even a mediocre chess algorithm can be pretty good when the human player is playing with a 2-second time limit. Really in an RTS game model the computer just has a "material bonus" -- the bonus is in calculations per second. It may make very simplistic calculations, but it can do millions of them in the time it takes a human to add two numbers together. Turn based games nearly all rely on material bonuses for the AI built into the rules -- more money for the AI, more units, and so forth: the Civilization games are enjoyable but aren't really challenging until the AI is given a huge production bonus and can produce and support several times the number of units as the human players. To make a game that has very challenging AI without AI cheats, my estimate is that I'd need to reduce the effective choices per turn to between 50 and 100 sensible choices. By compiling huge libraries of moves (as good chess algorithms do) I could possibly increase this to 150 choices, more depending on how the libraries/rule-systems relate, but the opening move libraries for chess are taken from books written by chess grandmasters and so it seems unlikely that any game I might design would have such a readily available resource. I'm asking the question of whether or not you would enjoy a game with a much simpler rule-set if it had very challenging AI because I'm genuinely interested in such games. This is one design model we have considered pursuing in the future. The games would have to be decidedly much "gamier" and less like a simulation. My guess is that games with a very simple set of rules and an elegant but simple UI would probably have a very small niche market -- as wargamers in general seem to prefer more rules, more pieces, bigger maps, longer games -- but such games might be easier enough to produce (fewer rules means less overhead all around) that even a very small niche could still support them. (And, frankly, I'd personally enjoy developing such games.)
|