Game AI been neglected because of graphics?
Two companies are selling two seperate games: one with incredible graphics but mediocre AI, the other with mediocre graphics but incredible AI. Which do you think will sell better?
Im sorry to nag about this, but this is the old "realistic vs. fun" argue, that can never be solved (bacause some think that reaslistic IS fun, and some dont).
But we can agree that games are suposed to be fun ,right?
So, if the game isnt fun, we can complain that its not realistic enough,
and that must be because the AI isnt realistic enough! right? Ok then
were clear.
My point is that i think alot of games with unrealistic AI is realy realy fun, and so, the "game company" did its part, and should not be blamed for its priorities. We must agree that games always can be better, and i think they will be and i hope that the ultimate game hasnt been developed yet.
My last nagging argument:
Think about the AI of the Super Mario Broz enemies... Now thats not alot of brain, but nobody blames them, because of all the fun!
dont say: bad AI / good AI,
say: bad game / good game,
THEN blame the company!
p.s: i totaly agrees with "WE NEED MORE AND BETTER GAMES" witch would be the result of better AI,GFX,SOUND,GAMEPLAY....
But we can agree that games are suposed to be fun ,right?
So, if the game isnt fun, we can complain that its not realistic enough,
and that must be because the AI isnt realistic enough! right? Ok then
were clear.
My point is that i think alot of games with unrealistic AI is realy realy fun, and so, the "game company" did its part, and should not be blamed for its priorities. We must agree that games always can be better, and i think they will be and i hope that the ultimate game hasnt been developed yet.
My last nagging argument:
Think about the AI of the Super Mario Broz enemies... Now thats not alot of brain, but nobody blames them, because of all the fun!
dont say: bad AI / good AI,
say: bad game / good game,
THEN blame the company!
p.s: i totaly agrees with "WE NEED MORE AND BETTER GAMES" witch would be the result of better AI,GFX,SOUND,GAMEPLAY....
I don't think that AI is being neglected because of graphics or physics. I think people have an unrealistically high expectation of progress in the field of AI. AI is ten times harder than physics (rigid body dynamics, fluid dynamics, etc) and a hundred times harder than graphics. What do I mean?
First, graphics and physics are just physics. "Physics" in games means Newtonian mechanics and maybe fluid dynamics. These topics have been studied rigorously for about 400 years. Computationally, you are just solving a geometry problem with ODEs (for rigid body dynamics) or a PDE (for fluid dynamics). Sure, it takes some work to design efficient data structures and algorithms, but there are really only a handful of possible computational approaches, e.g. separating planes, constraints, verlets. Graphics is just an approximate simulation of the behavior of light and its interactions with different materials. This is even easier than physics. The basic graphics pipeline is highly optimized and the pipeline algorithm is shockingly parallel and easily accelerated in hardware. Graphics is really nothing more than linear algebra and maybe a little calculus for lighting equations. Graphics has advanced so much in recent years because it is so easy. Yes, there is still a lot of room for improvement, but the problems are well understood.
Furthermore, the same graphics and physics methodologies apply to virtually all game genres. Rendering is rendering, regardless of whether you are rendering a shooter, a racing game, or an RTS.
On the other hand, AI is a very diverse field that is attempting to solve an extremely hard problem. Depending on your point of view, it either took a billion years of evolution or an act of God to produce intelligent agents here on earth. Are we to suppose that humans could have solved the problem in a mere 60 years? Of course, no one is suggesting that games implement conscious, sentient intelligent agents. Rather, games must produce the illusion that opponents are behaving in an intelligent, possibly coordinated, manner. The available AI methods depend strongly on the type of game being developed. For instance, FPS and RTS games need sophisticated path finding capabilities, but the underlying geometric data structures place very different constraints on the AI solution. The path finding methods used in a football game are likely to differ considerably from those in an RTS or FPS. Group-level tactical and strategic reasoning pose their own set of problems and potential solutions. Other seemingly easy games can actually be a big challenge for an AI programmer (for instance, do a google search for poker AI and you'll see what I mean).
None of this is meant to be a discouragement to AI programmers. I think that AI is an extremely important component of any good game, but I also feel that AI is significantly harder than many people realize, especially compared to graphics or physics.
First, graphics and physics are just physics. "Physics" in games means Newtonian mechanics and maybe fluid dynamics. These topics have been studied rigorously for about 400 years. Computationally, you are just solving a geometry problem with ODEs (for rigid body dynamics) or a PDE (for fluid dynamics). Sure, it takes some work to design efficient data structures and algorithms, but there are really only a handful of possible computational approaches, e.g. separating planes, constraints, verlets. Graphics is just an approximate simulation of the behavior of light and its interactions with different materials. This is even easier than physics. The basic graphics pipeline is highly optimized and the pipeline algorithm is shockingly parallel and easily accelerated in hardware. Graphics is really nothing more than linear algebra and maybe a little calculus for lighting equations. Graphics has advanced so much in recent years because it is so easy. Yes, there is still a lot of room for improvement, but the problems are well understood.
Furthermore, the same graphics and physics methodologies apply to virtually all game genres. Rendering is rendering, regardless of whether you are rendering a shooter, a racing game, or an RTS.
On the other hand, AI is a very diverse field that is attempting to solve an extremely hard problem. Depending on your point of view, it either took a billion years of evolution or an act of God to produce intelligent agents here on earth. Are we to suppose that humans could have solved the problem in a mere 60 years? Of course, no one is suggesting that games implement conscious, sentient intelligent agents. Rather, games must produce the illusion that opponents are behaving in an intelligent, possibly coordinated, manner. The available AI methods depend strongly on the type of game being developed. For instance, FPS and RTS games need sophisticated path finding capabilities, but the underlying geometric data structures place very different constraints on the AI solution. The path finding methods used in a football game are likely to differ considerably from those in an RTS or FPS. Group-level tactical and strategic reasoning pose their own set of problems and potential solutions. Other seemingly easy games can actually be a big challenge for an AI programmer (for instance, do a google search for poker AI and you'll see what I mean).
None of this is meant to be a discouragement to AI programmers. I think that AI is an extremely important component of any good game, but I also feel that AI is significantly harder than many people realize, especially compared to graphics or physics.
I think one primary reason as to why graphics and physics has (and still is) advancing more quickly than AI is because it's a hell of a lot easier to debug. Graphics and physics are instant... setup a scene, watch it for 2 seconds, and then put a bandaid on whatever is obviously not working (blending setting, texture, color, gravity, etc).
If we approach 99.9% humanlike AI, then I wouldn't really care as long as there's multiplayer to be honest. Unless it's a game like Black and White, now THAT was some very remarkable AI.
If we approach 99.9% humanlike AI, then I wouldn't really care as long as there's multiplayer to be honest. Unless it's a game like Black and White, now THAT was some very remarkable AI.
SlayerDave: I think you far underestimate what has already been done in researching game-applicable AI. It isn't as if I want developers to have their design team researching AI for a year before starting the game, I just want them to buy an 'Intro to Game AI in 24 hrs' book and look at what is common knowledge these days.
Quote:Original post by Extrarius
SlayerDave: I think you far underestimate what has already been done in researching game-applicable AI. It isn't as if I want developers to have their design team researching AI for a year before starting the game, I just want them to buy an 'Intro to Game AI in 24 hrs' book and look at what is common knowledge these days.
Sure, there are plenty of commonly used AI techniques in game development. But the whole point of this thread is that game AI sucks, is progressing much slower than game graphics, and is making less of a positive impact on the gameplay epxerience than graphics and physics. I'm not suggesting that game developers become AI researchers - that's what CS departments at universities are for.
My point was just that AI doesn't appear to be progressing as fast as graphics or physics because AI is a lot harder. Players want a more realistic game experience and it's a lot easier to crank up the graphics and physics than it is to implement high-quality AI.
Quote:Original post by SlayerDavePerhaps I'm misreading this thread, but to me it feels like it is about the lack of AI in games, rather than the lack of game AI techniques. The problem is poor or no implementations in real games (as opposed to indie games, tech demos, and proof-of-concept projects). We know how to make awesome AI, but nobody is using it in commercial game development.Quote:Original post by Extrarius
SlayerDave: I think you far underestimate what has already been done in researching game-applicable AI. It isn't as if I want developers to have their design team researching AI for a year before starting the game, I just want them to buy an 'Intro to Game AI in 24 hrs' book and look at what is common knowledge these days.
Sure, there are plenty of commonly used AI techniques in game development. But the whole point of this thread is that game AI sucks, is progressing much slower than game graphics, and is making less of a positive impact on the gameplay epxerience than graphics and physics. I'm not suggesting that game developers become AI researchers - that's what CS departments at universities are for.
My point was just that AI doesn't appear to be progressing as fast as graphics or physics because AI is a lot harder. Players want a more realistic game experience and it's a lot easier to crank up the graphics and physics than it is to implement high-quality AI.
We know how to make awesome AI, but nobody is using it in commercial game development.[\quote]
As you sugested, I am goign to respond separating techniques from features. In the end, techniques are just a particular implementation. Yes, they can be critical for time savings during dev, etc, but the player won't necessarily be able to tell the difference between a Fuzzy Neural Net Probability Wavelet AI and an massive, huge set of if statements. ;)
Features people have mentioned in this thread so far are:
1) Strong tactical AI
2) AI that recognizes failures and changes tactics
3) Presenting the player with more challenges and diverse situations.
4) Replayability
(I apologize for my FPS-centric response, it is the area I am the most familiar with).
Strong tactical AI typically is applied in tactical games. ;) This is an area that is expanding, but you are right, perhaps devs aren't taking advantage of it enough. On the other hand, there are many games out/coming out which do. Band of Brothers and its squad/flanking/etc. Killzone and ShellShock (which William was involved with). I can't think of other names off hand, but there are several games exploring these concepts.
AI that recognizes failure and changes tactics is another good item. There are two obstacles here to more of this in games. The first is recognizing failures, attributing them to an action, and coming up with an alternate behavior is a decent amount of work. The other issue is that alternate tactics means more behavior permutations (and frequently more animation/audio content). Of course, there has to be room in game play space for the AIs to do something else. If, due to game design, an AI is constrained to a room with a single weapon and told to charge the player and kill them, the alternate behavior options are minimized.
Point 3 I would argue is a larger issue. This is more of an 'application of AI by content team' issue. Generally, level designers (or whoever is wiring a map) determine what routes there are, what AIs are in the space, when the AIs are created and destroyed, what areas the AIs can traverse (particularly important in console games with streaming; you need to know what textures are going to be in what area). Yes, AI programmers provide tools for level designers, but as they are the 'clients', it is frequently up to them to determine what is needed in the bigger picture.
Point 4 is closely tied to point 3. As the work of level designers generally dictates what happens and where, AI programmers ability to create diverse situations is constrained by the gameplay environment and embedded logic. Do you have any examples of replayability you would like to see?
And just to show I am not trying to globally defend AI programmers everywhere, here is a paper with some issues I agree with. :)
http://www.simbionic.com/gameai/papers/AAAI04-WS4-Wetzel.ppt
http://www.simbionic.com/gameai/papers/AAAI04-WS4-Wetzel.ppt
Quote:Original post by John Reynolds
I agree Physics has become an important part of certain genres of gaming. However, I don't think it has done so at the cost of AI. Every project I know of has dedicated AI programmers, and indeed more AI programmers than Physics, and if you look around at who is hiring it is clear how in-demand AI programmers currently are.
The Graphics vs Gameplay balancing-act has eased somewhat thanks the the GPU. Now the graphics has its own processor it freed up a lot of CPU time to Gameplay, Physics, AI, etc.
At the 2005 GDC (and indeed at many of the past few GDCs) virtually every serious project had a minimum of one dedicated AI programer. Several had two, and a handful had three (which just astounded me). They tended to break things up into handling different layers at that point, which I thought was interesting...this guy did the top-level strategic decision making AI and handed orders off to the guy who handled individual unit AI. The third usually did something like animation-related stuff, or companions, or perhaps NPCs (who usually just need to follow orders and not look completely stupid otherwise). It was fascinating.
I think AI will continue to be a major factor in any larger game design, with physics becoming more and more important where it makes sense (like FPS games).
This topic is closed to new replies.
Advertisement
Popular Topics
Advertisement