When players played Doom a few years back, the behaviour of the enemies was pretty easy to predict (although I must say, I stink at Doom and games like that, so for me it actually wasn''t all that easy, but still, it was no Herculian -sp?- task). A few weeks back I was playing a demo game of Noone Lives Forever (a James Bond type FPS). And I played it for quite a few days (it''s a really nice demo). I must say that the difference in intelligent behaviour was obvious. Enemies duck for cover, throw over tables to hide behind, jump out from behind walls, run towards you in teams trying to overpower you. A nice step forward (though maybe not big enough for some, as it''s still not too hard to predict enemy moves). Alright, so we can give some ''easy'' AI to units (the AI only has to respond to one player and only to the actions of one character, and the AI units only have a few types of actions to choose from: close in on enemy, find cover, shoot at enemy, perform special moves). Can we take another step forward? Can we step towards a bigger, better, bolder AI? I think one of the classic examples of ''AI'' (is it really artificial intelligence? I still think it''s more of a ''smart scripting'') is the chess computer. Although I''m not really aware of exactly how a chess program is created, it seems to me that the designers have to input a reaction to every player move. It pretty much just processes every move possible (and thinks ''ahead'') and based on the programmed software ''makes a choice'' of which move to perform. Each and every year, these chess software programs seem to become smarter, even beating the human world champion every now and then (what are the latest results?). Will it ever become good enough to claim the title of world champion? And is it the software or the human beings that created the software that should take pride in the victories? Still, for now humanity can still beat those chess programs(Right? Although I''m sure that the average Joe would probably get defeated by the top chess software time and time again, I think that the really good human chess players can still win against a computer program). So, what exactly do we expect from games that use AI? Do we expect the computer opponent to be able to defeat us without cheating? Without putting the player at almost impossible odds of winning? Because without a challenge... there''s no fun (except to the few people who as little kids took pleasure in beating up little kids and tapping themselves on the shoulder for beating up yet another much weaker kid). In Noone Lives Forever the ''impossible odds'' come in the form of you, one character, having to defeat them, many, many enemies. In RTS games ala C&C the ''impossible odds'' come in the form of the computer being able to process orders faster, or ''cheat'' a little. In most strategy games (the last one I played was Fantasy General) the ''impossible odds'' come much like the one in Noone Lives Forever: the enemy simply starts with more resources, more units, stronger units, etc. Is there ANY game where the player is supposed to combat a computer opponent, where the computer opponent is NOT much stronger, much richer, much faster or much better at cheating? Command & Conquer: Will there ever come a time when the computer can play fair, instruct its own troops as fast as the player instructs his (or hers)? Fantasy General: Will there ever come a time when the computer can play a fair game without giving itself many more and much stronger units? Will AI ever match Human Intelligence enough to warrant a fair battle of wits?(and I don''t believe the average human intelligence is all that great, so we don''t even have to set our AI goals to high) If yes: When? How? If no: Why not? What are our alternatives?

Hey, Silvermyst!

Actually, chess programs aren''t able to contain comprehensive databases of all possible moves. This is because the number of possible games grows exponentially beyond the first moves.

On the first move, each player may choose from 20 possible moves. So after each player has made just the first move, there are 400 possible positions. And the possibilities grow from there.

For this reason, chess programs use "pattern matching". The programs are taught how to recognize certain basic features of a given position. For instance, the program will recognize the value inherent in a useful open file for a rook.

The chess program is given extra help by having access to a limited database which contains common openings, and probably a comprehensive database of all possible moves out to a certain number of moves.

As for AI in general, I''ve been reading an interesting book on game design which addresses that subject.

In "Game Design: Theory and Practice", the author points out that the AI for a game really only needs to be good enough to provide a good challenge for the player. For instance, Tetris didn''t feature any real AI. The only AI-like programming used in that game was a random number generator which chose which block would appear next.

Of course, some games require very advanced AI. FPSs are expected to feature top notch AI. But for other games, AI may not really need to be impressive, as long as the player enjoys a good challenge.


Jonathon

quote: "Mathematics are one of the fundamentaries of educationalizing our youths." -George W. Bush

I really don't think chess programming has much to offer in terms of a generally intelligent agent. And I also think there is no limit to what could be usable in games given improvements in AI technology. Some key areas which would greatly expand game characters include belief systems, knowledge integration and acquisition, and common sense reasoning.

If you're really interested in AI that is different from that which you find in today's games, you might want to do some searches with the keyword 'AI' and one of the following: belief systems, defeasible reasoning, monotonic reasoning, nonmonotonic reasoning, axioms, knowledge bases, universal quantification, existential quantification, situation calculus, predicate calculus, modeling discourse, Cyc, OSCAR, Soar, Wordnet, Loom, Prolog, Otter, automatic deduction, Prolog, LISP, Scheme, common sense reasoning, ontologies, truth maintenance, default reasoning, resolution refutation, natural deduction, semantic nets, the frame problem, symbolic processing, metalevel reasoning, etc.



Edited by - bishop_pass on May 7, 2001 12:11:08 AM

The bots in some recent FPS games (UT comes to mind) don''t cheat much or at all, and are pretty difficult one on one. They play the same game that the human players are playing and you don''t fight millions and millions of them at a time.

''bots in FPS do cheat - they''re perfect shots. If a ''bot shoots at you and misses, it''s because he was programmed to miss, not because he''s a bad shot or anything.

Gosh, how long will it be before someone invents an AI system which can help me to understand what Bishop Pass said?


Jonathon

quote: "Mathematics are one of the fundamentaries of educationalizing our youths." -George W. Bush

fuck you bitch ...DOOM rulles!!!!!!!!!!!!!!!!

fuck you some more !! dooms AI is what made the game,maybe you should have not playd it on easy?

Develop artificial neural networks!


They should offer "human-like" challenge (AI).

quote: Original post by Silvermyst

Is there ANY game where the player is supposed to combat a computer opponent, where the computer opponent is NOT much stronger, much richer, much faster or much better at cheating?

I think there''s an ugly tradeoff here: Right now, any game the computer could reliably beat you at is probably a game you don''t want to play. This is because the paths to victory (patterns like Jonathon was talking about) are known or relatively easily knowable. There''s a set of strategies to win, and this bores humans very quickly.

The bigger, more variable, and more open ended a game, the more of a nightmare it is to program the AI using conventional methods. That''s why for a big, complex game like Alpha Centauri or Galactic Civilizations, the AI cheats with information or resources; that''s why in an RTS, AI is best at uncreative frontal attacks and little else. Too many factors.

There''s another trade-off as well: Using some sophistocated techniques (ask bishop_pass for details, I''m sure he''d be happy to supply them ) you may be able to get VERY complicated AI that can kick your butt. But your framerate and polygon count may suffer massively in the process.


--------------------
Just waiting for the mothership...