Will AI really benefit from the extra processor power?

[nameless_one]

We keep hearing that we'll get better AI with the new consoles due to increase in processor power. I'm not so experienced in AI, but most of the game AI seems to be that it's bounded by poor algorithms then not enough computational power. It doesn't scale like graphics or physics calculations. Most of the game AI is very simplistic and wouldn't benefit much from more horsepower. Which means AI has to be rewritten completely which equals a lot of money, something most publishers will not be willing to spend money on.
Surely we'll be able to have the same AI we have now for many more characters at once, but will we able to have more advanced AI for a few characters?

 

[Titanio]

Surely we'll be able to have the same AI we have now for many more characters at once, but will we able to have more advanced AI for a few characters?

I think so, but it is true that AI is still a developing field generally. It is not a solved problem in the same way graphics might be considered to be (or not nearly as solved, if you don't consider graphics to be all worked out ). So yes, we are bound by an imperfect state of the art, but I don't think games are pushing up on that cutting edge at all.

 

[Wunderchu]

http://www.beyond3d.com/forum/viewtopic.php?t=23783

 

[jvd]

Of course it will .

There is one problem that plagues sound , a.i , physics and graphics .

And that is processing power . As processing power grows all the above will improve .

 

[BlueTsunami]

Surely we'll be able to have the same AI we have now for many more characters at once, but will we able to have more advanced AI for a few characters?

I think so, but it is true that AI is still a developing field generally. It is not a solved problem in the same way graphics might be considered to be (or not nearly as solved, if you don't consider graphics to be all worked out ). So yes, we are bound by an imperfect state of the art, but I don't think games are pushing up on that cutting edge at all.

Well, I believe AI can only go so far as far as algorithms go.....there is no such thing as a perfect AI (so far) in software form. I guess the more random the movement of the characters and the more reactions they get on certain actions would be a step up to the AI we have now but its my understading that AI intensive games causes the CPU to be heavily utilized. IF thats the case...better processing power is always welcomed but in the end it could be that better code and algorithms are needed..but thats the case for every other type of code.

 

[BOOMEXPLODE]

I think one day you will see AI systems licensed as middleware just like we're starting to see physics systems licensed as middleware. Developing really "smart" AI is just too huge of a project for a games developer.

 

[Wunderchu]

I think one day you will see AI systems licensed as middleware just like we're starting to see physics systems licensed as middleware.

already happening (the link to the other thread I posted above)

 

[ERP]

I think one day you will see AI systems licensed as middleware just like we're starting to see physics systems licensed as middleware. Developing really "smart" AI is just too huge of a project for a games developer.

Depends what you mean by AI.

I can see packaged solutions to common problems like path finding.

But in general it's a very specialised area, AI for a game like Madden is very different to say the AI for an FPS.

I'm still not sure I understand what people really want when they talk about better AI. do you really want to be challenged by a single character in an FPS, to me it seems that the crux of most of the single player FPS's is you against an army of enemies. Make them smart enough to be competitive one on one and the game is going to be pretty frustrating.

 

[Laa-Yosh]

Halo's AI was good, and it was used even better. Versatile and complex enough to create many different kinds of emergent gameplay, depending on the surrounding area, weapons, vehicles etc. Good enough to provide challange even in one-on-one fights in high difficulties, yet fast enough to build up small scale battles.
I'm quite curious to see what those Bungie coders can do with more processing power. One way would be to vastly increase the number of AI characters on both sides, and put the player into a fully dynamic battlefield... but I guess they'll go with a smaller increase and smarter AI.

About the frustration... deathmatching against better Quake players is IMHO a lot more frustrating then a tough AI - you can always lower the difficulty level

 

[Josh378]

Ok, so all of this "PS3/360 is weak at branch coding, so they will suck at A.I." comments are just plain bull?

So A.I. will improve this gen?

-Josh378

 

[Laa-Yosh]

The Bungie presentation on their AI indicated that most of their processing time is spent on raycasting to figure out what an AI can see, hear, shoot etc. and that each AI only tracks surrounding actors because it's such a huge speed hit to track everyone else. Running decision code for their finite state machine doesn't take much time.
So I suppose more processing power could be easily exploited in most cases. Increase number of actors, AI awareness, add more AI memory etc. etc.

 

[Monty]

well i would like a ai which plans ahead, instead of shoot when it sees someone in fireing range. such as on halo, that ice multiplayer level, human players who see someone coming through the teleporter would hide behind the teleporter and pop the person in the back with a shotgun (well, clever people anyway), ai there would just run through the teleporter to shoot u, no tactics invold, and since u would know the com would do that u r already prepared to blow them to hell.

 

[jvd]

Josh i can't see how a tri core 3.2 ghz cpu can be anything but a few times better at a.i than a 733mhz celron

 

[Josh378]

Josh i can't see how a tri core 3.2 ghz cpu can be anything but a few times better at a.i than a 733mhz celron

I'm reading the archtechnica review on the 360's tech....So, don't shoot me...Just reading what i find

-Josh378

 

[ERP]

The Bungie presentation on their AI indicated that most of their processing time is spent on raycasting to figure out what an AI can see, hear, shoot etc. and that each AI only tracks surrounding actors because it's such a huge speed hit to track everyone else. Running decision code for their finite state machine doesn't take much time.
So I suppose more processing power could be easily exploited in most cases. Increase number of actors, AI awareness, add more AI memory etc. etc.

Yup the actual percentage of time your average game spends in AI code is minimal unless you're counting things like ray casts, or other perception tests.

Path finding can get expensive aswell.

 

[ERP]

well i would like a ai which plans ahead, instead of shoot when it sees someone in fireing range. such as on halo, that ice multiplayer level, human players who see someone coming through the teleporter would hide behind the teleporter and pop the person in the back with a shotgun (well, clever people anyway), ai there would just run through the teleporter to shoot u, no tactics invold, and since u would know the com would do that u r already prepared to blow them to hell.

Think about this from a gameplay stanpoint for the moment.

I step into the teleporter an bang I'm dead, no skill involved, just dead. In what way is that fun gameplay. Realistic possibly fun no.

 

[aaaaa00]

To echo ERP's comments:

When people say they want "better" AI, they don't really mean they want "better" AI.

They do not want an enemy that is harder to defeat.
They do not want a bot with inhuman aim.
They do not want an AI that always knows where you are and perfectly anticipates where you're going to be next and what you're going to do.

No, what they really want is:

1. An AI that doesn't look stupid.

Running into walls, being unable to find their way around an object, stuck in loops, getting stuck on geometry, saying the wrong thing at the wrong time, ignoring player actions, all of this makes your AI look stupid.

2. An AI that doesn't do impossible things.

No clipping through geometry, no being able to see you when it shouldn't, no hits coming from impossible angles, no visible cheating.

3. An AI that is mostly predictable (but occasionally surprises).

If the user can't figure out the internal state of the AI, then he's just going to assume it's actions are random (and thus stupid). This means your state machines need to be simple enough for the user to understand them through the feedback systems available to them in the game (sound effects, animations, voice samples, etc).

This is actually kind of counter-intuitive -- you'd think more complex AI would give you more believable behavior, but this isn't always true.

4. An AI that is plausibly fallible.

In the end, the whole point of the game's AI is NOT to beat the player. The goal for game AI is to give the player a good fair challenge, but in the end, the AI must LOSE in the most entertaining way possible.

If the AI never loses, the player will decide that he's not having fun, and call the AI unfair.

If the AI loses too easily or in an implausible fashion, the player will decide he's not having fun, and think the AI is stupid.

 

[AlNom]

Not all elites in Halo 2 will follow you to their deaths. Although they can fire like a wildman while going to find their cover. Meanwhile another elite wacks you from behind. (thinking of Cairo Station on legendary)

 

[pc999]

For me a good AI is one that make me think in "what should I do?", "what he is trying to do?"...

But you are right none wants a game that is impossible to beat (meanwhile we like to have that possibility of an unbeatble game).

 

[Johnny_Physics]

I'm still not sure I understand what people really want when they talk about better AI. do you really want to be challenged by a single character in an FPS, to me it seems that the crux of most of the single player FPS's is you against an army of enemies. Make them smart enough to be competitive one on one and the game is going to be pretty frustrating.

An AI with a vider range of responses to various situations, not just the ordinary "idle" and "search & destroy".

Deus Ex(the first one) got a lot of flack for its AI but I think it worked great - and my opinion is entirely based around my ability to fool it so that I felt really smart.

An example from Deus Ex would be when I was chased through Hell's Kitchen by a bunch of soldiers, I ran around a corner which they saw and they naturally followed me, but after that I quickly rounded another corner and hid in the shadows behind some garbage - something they didn't see. So when they arrived in the alley they didn't just turn around and started shooting at the trashcan cloaked in darkness that I was covering behind, instead they just kept running until the road split and instead of choosing one way the group formed into two search parties each taking a separate way trying to hunt me down and it felt really sound and natural that they did that. The game responded like I hoped that it would and it really made the game a lot more immersive.

I would like to see more of those type of situations where you can actually play against the AI instead of just competing against it.