do you think a dedicated AI processor should be in the next gen systems?
- Thread starter pukemon
- Start date
You'll need to be more clear about the details here. You're either asking for something that isn't available or isn't necessarily going to be applied by developers in the way you desire.
The thing to keep in mind about AI is that there isn't any. The best stuff out there couldn't outperform a lab rat, provided you can figure out how to program a rat. The stuff we have now is advantageous for its programmability, not its performance.
If you had in mind a specialized piece of processing hardware for AI, nobody makes anything even remotely suitable for the task of a game console. The division between software AI and hardware AI has been an ongoing battle since the 60's. The software guys say any structure needed can be produced entirely in simuation. The hardware guys believe that if your base equipment doesn't bear a strong structural resmblance to a biological brain you'll never produce something that can be described as thinking the way an animal does.
There is some neural net hardware marketed for research and even a few commercial apps but nothing in the price range or on the fly reusability needed for a game system.
If, OTOH, you just mean a genral purpose processor within a system designated as being only for AI use within games, forget it. Would you really expect the platform companies to review third party source code just to make sure they only used that processor for AI functions? Not only would this be a silly and burdensome enforcement duty it also presumes that a dedicated processor is going to be one size fits all when it would be overkill or inadequate for many games.
Most bad AI is not for lack of processing capacity. It's just bad design and coding. Developers who don't know how to achieve convincing behavior in NPCs are going to be equally incompetent regardless of amount of silicon set aside for the task. To a certain extent gamers are asking for something that nobody has ever achieved even when vast sums of money and recognition is offered.
The thing to keep in mind about AI is that there isn't any. The best stuff out there couldn't outperform a lab rat, provided you can figure out how to program a rat. The stuff we have now is advantageous for its programmability, not its performance.
If you had in mind a specialized piece of processing hardware for AI, nobody makes anything even remotely suitable for the task of a game console. The division between software AI and hardware AI has been an ongoing battle since the 60's. The software guys say any structure needed can be produced entirely in simuation. The hardware guys believe that if your base equipment doesn't bear a strong structural resmblance to a biological brain you'll never produce something that can be described as thinking the way an animal does.
There is some neural net hardware marketed for research and even a few commercial apps but nothing in the price range or on the fly reusability needed for a game system.
If, OTOH, you just mean a genral purpose processor within a system designated as being only for AI use within games, forget it. Would you really expect the platform companies to review third party source code just to make sure they only used that processor for AI functions? Not only would this be a silly and burdensome enforcement duty it also presumes that a dedicated processor is going to be one size fits all when it would be overkill or inadequate for many games.
Most bad AI is not for lack of processing capacity. It's just bad design and coding. Developers who don't know how to achieve convincing behavior in NPCs are going to be equally incompetent regardless of amount of silicon set aside for the task. To a certain extent gamers are asking for something that nobody has ever achieved even when vast sums of money and recognition is offered.
BlueStorm781
CAGiversary!
I don't think the lack of resources has to do with a crappy AI, it's the developers who don't take the time "to flesh it out". Didn't Sony create the "Emotion Engine" for the purpose of an advanced AI*? And if that was so, it really hasn't changed the experience of playing against computer opponents much.
I think the AI problem is mostly due to the developers who really don't wan to take the time to make it as intricate and realistic as possible. Don't get me wrong, many developers do take the time to make good AI opponents, but when you look at games that are cash-ins on licenses, or just didn't have a lot of development time, they have pretty weak AI opponents.
But there's one thing that bugs me more than a weak AI opponent, and that's the AI who is a friggin cheating bastard. Like the opponents in Need For Speed Underground who run you off the road or how every CPU controlled car in Test Drive
practically tosses cars right in front of you to slow you down. It's one thing to have a weak AI, it's another to have what I call an "A-Hole AI", which almost ruins the fun of a game with frustration.
So, I don't think more resources in the hardware could make an improvement in a game's AI, it's really up to the developer and how much work they put into the code.
*If I'm wrong on the Emotion Engine thing, can someone tell me what it WAS originally for? I think it was nothing more than "blast processing", because I haven't seen the Emotion Engine in work since I've had the PS2.
I think the AI problem is mostly due to the developers who really don't wan to take the time to make it as intricate and realistic as possible. Don't get me wrong, many developers do take the time to make good AI opponents, but when you look at games that are cash-ins on licenses, or just didn't have a lot of development time, they have pretty weak AI opponents.
But there's one thing that bugs me more than a weak AI opponent, and that's the AI who is a friggin cheating bastard. Like the opponents in Need For Speed Underground who run you off the road or how every CPU controlled car in Test Drive
practically tosses cars right in front of you to slow you down. It's one thing to have a weak AI, it's another to have what I call an "A-Hole AI", which almost ruins the fun of a game with frustration.
So, I don't think more resources in the hardware could make an improvement in a game's AI, it's really up to the developer and how much work they put into the code.
*If I'm wrong on the Emotion Engine thing, can someone tell me what it WAS originally for? I think it was nothing more than "blast processing", because I haven't seen the Emotion Engine in work since I've had the PS2.
The name 'Emotion Engine' was largely hype but not in the way suggested. It was hyped as allowing for sufficient detail for characters to show recognizable emotions as in facial expressions in real-time animation. This is a different animal altogether from AI. After all, without any interaction it's impossible to say how much intellectual depth lies behind the most mobile of faces.
So far very few games have really offered this sort of thing. Final Fantasy X is probably the best known and it certainly did set some new standards for real-time animation in console gaming but there was nothing shown that couldn't be done on either of the competing systems. Some of the cooler aspects like Lulu's hair and clothing that acted like separate objects rather than a layer of paint on the characters are just a combination of high polygon counts and increased physics calculating under the hood. Some of that stuff can be improved in hardware. Many of NVIDIA and ATI's most recent demos for DirectX 9 level GPUs offer a lot of emotional simulation that is greatly enhanced by the highly refined simulation of human flesh.
In demos like 'Dawn' the light doesn't just bounce off her skin, it goes in a few layers before reflecting, just like a real animal. This reveals details under the suface, especially the circulatory system. When the character shows anger or embarrassment the emotion is much more believable because the flow of blood concentrates in certain areas like the cheeks. There is also a musculature simulation involved. The different emotions aren't just textures, they are deformations of the suface polygons in accordance with the simulated muscles to create a smile or other reaction.
This is pretty powerful since it allows for the whole spectrum of facial movement and coloring. It also eats up processing time like crazy, which is part of why the transistor counts on DX9 chips are so high. To do something useful with this capability in a game you need a substantial amount of processing to run a goodly chunk of code to make these emotional states arise in response to the player's actions in a way that doesn't make the NPC seem stupid or insane. (Assuming the character isn't suppose to have either of those traits.)
So when people wonder what the next generation of console will do that the current one cannot, the above is a prime example. The games won't just be prettier. The interaction will be much more complex. This also means that the development time for game that really exploit this capability could run very long and call for new types of tools.
So far very few games have really offered this sort of thing. Final Fantasy X is probably the best known and it certainly did set some new standards for real-time animation in console gaming but there was nothing shown that couldn't be done on either of the competing systems. Some of the cooler aspects like Lulu's hair and clothing that acted like separate objects rather than a layer of paint on the characters are just a combination of high polygon counts and increased physics calculating under the hood. Some of that stuff can be improved in hardware. Many of NVIDIA and ATI's most recent demos for DirectX 9 level GPUs offer a lot of emotional simulation that is greatly enhanced by the highly refined simulation of human flesh.
In demos like 'Dawn' the light doesn't just bounce off her skin, it goes in a few layers before reflecting, just like a real animal. This reveals details under the suface, especially the circulatory system. When the character shows anger or embarrassment the emotion is much more believable because the flow of blood concentrates in certain areas like the cheeks. There is also a musculature simulation involved. The different emotions aren't just textures, they are deformations of the suface polygons in accordance with the simulated muscles to create a smile or other reaction.
This is pretty powerful since it allows for the whole spectrum of facial movement and coloring. It also eats up processing time like crazy, which is part of why the transistor counts on DX9 chips are so high. To do something useful with this capability in a game you need a substantial amount of processing to run a goodly chunk of code to make these emotional states arise in response to the player's actions in a way that doesn't make the NPC seem stupid or insane. (Assuming the character isn't suppose to have either of those traits.)
So when people wonder what the next generation of console will do that the current one cannot, the above is a prime example. The games won't just be prettier. The interaction will be much more complex. This also means that the development time for game that really exploit this capability could run very long and call for new types of tools.
JSweeney
CAGiversary!
The problem is not a lack of hardware capability, it's a dearth of properly defined hueristics and insufficent knowldege bases. It's not that advanced AI isn't possible, it's that it's difficult and time consuming to properly define heuristics, and rather difficult to frame your knowledge base correctly.
- Feedback
- 131 (100%)
[quote name='pukemon']advanced? how so? i think it's time. fps, and sports games have stupid opponents. even halo. halo by far has the best ai, but once you beat it once, you know what to expect. even on legendary level.[/quote]
there are many games with better AI than Halo
there are many games with better AI than Halo
[quote name='Kaijufan']The Xbox Next is rumored to have 2 or 3 processors, so it is possible that developers for it will use one for AI if it does have that many processors.[/quote]
That is entirely dependent on the game. Considering the sheer power of the supposed class of CPU involved it's unlikely any one of those would be given entirely to the task of AI unless you had a very large number of characters who had to display convincingly individual behavior that couldn't be handled as a swarm. (Examples of swarms are the antelope stampede in Lion King or the watchmacallitsaurs in Jurassic Park. This kind of code translate pretty well to interactive stuff if the rendering speed is up to snuff for the desired quality.) You might have a game like Jak II where the city dwellers have more individual identity and aren't just half a dozen generic generic NPCs. This could lead to games where scenarios aren't preset and predictable. For instance, you have to track down a particular NPC who is wandering around with their own agenda which may be subject to change depending on other events you may have influenced.
This has great potential but it's also a massive task for designers. Games often credit themselves with being non-linear but if you make things too freeform you lose any narrative control and the player may find themselves wandering around with no idea of how to achieve anything more useful than mere survival. This is a problem that has been a critical element in finding the balance for MMORPGs. A lot of players actually get involved in rather mundane careers and never do anything of a quest-like nature. Most folks find this odd since they have real life for that sort of thing and play the game to take part in extraordinary events rather than moonlight in a virtual economy.
OTOH, these kinds of people can be a valuable asset to the company running the game. Who needs AI when you have real people doing the work of NPCs?
That is entirely dependent on the game. Considering the sheer power of the supposed class of CPU involved it's unlikely any one of those would be given entirely to the task of AI unless you had a very large number of characters who had to display convincingly individual behavior that couldn't be handled as a swarm. (Examples of swarms are the antelope stampede in Lion King or the watchmacallitsaurs in Jurassic Park. This kind of code translate pretty well to interactive stuff if the rendering speed is up to snuff for the desired quality.) You might have a game like Jak II where the city dwellers have more individual identity and aren't just half a dozen generic generic NPCs. This could lead to games where scenarios aren't preset and predictable. For instance, you have to track down a particular NPC who is wandering around with their own agenda which may be subject to change depending on other events you may have influenced.
This has great potential but it's also a massive task for designers. Games often credit themselves with being non-linear but if you make things too freeform you lose any narrative control and the player may find themselves wandering around with no idea of how to achieve anything more useful than mere survival. This is a problem that has been a critical element in finding the balance for MMORPGs. A lot of players actually get involved in rather mundane careers and never do anything of a quest-like nature. Most folks find this odd since they have real life for that sort of thing and play the game to take part in extraordinary events rather than moonlight in a virtual economy.
OTOH, these kinds of people can be a valuable asset to the company running the game. Who needs AI when you have real people doing the work of NPCs?
I can see it now. One chip to rule them all...
And when that one gamer finds the flaw in the logic, every game that uses it will be easily defeated. Sorry, but AI chips are not a good idea.
If you are really interested in AI, Game Developer magazine did an article on AI and why the game industry hasn't embraced it yet. The bottom line was, "AI is too advanced for gaming." Check it out if you can find it.
And when that one gamer finds the flaw in the logic, every game that uses it will be easily defeated. Sorry, but AI chips are not a good idea.
If you are really interested in AI, Game Developer magazine did an article on AI and why the game industry hasn't embraced it yet. The bottom line was, "AI is too advanced for gaming." Check it out if you can find it.
bread's done