The Elder Scrolls IV: Oblivion

Quote:

---

I always figured it was hard, but if more developers develop AI engines (like a physics engine in the realm of an Unreal or Havok) it could progress a bit faster then it is at the current time.

---

We've already discussed this and I explained it once. An "AI Engine" isnt feasible because AI is specialized per game. AI is little more than an elaborate set of if-then statements, tuned to each individual game. Even games that you may think are similar likely have very different AI to fit the need of the game in question. They share some core concepts like what you describe but thats just the low-level type stuff, all the behavior is individualized for the game's engine and programming (a guard that reacts in MGS3 will act totally different from a guard that reacts in Half-Life 2, and how they react and are told to react is in part dependent on the tools used to make the game). As for totally different games, say Warcraft 3 and MGS3 and Sims 2, not a shared bit of AI code in them. Physics is physics, a ball bouncing is a ball bouncing, developers can modify how that ball bounces by degrees but thats why its possible to make a physics engine.

What youre talking about is some sort of generic AI construct that can adapt to different game scenarios. This is decades away, at the very least, even if every game developer on the planet is working on this (and in any case, far smarter people, like MIT researchers, are doing AI research). Even the AI that can currently beat a Grand Master in Chess is just able to do so from brute force, not being able to learn in a way that can make that AI play Go or Game of Life well. You may remember that there was a spat because the developers fed Kasparov's moves into the AI, well all the computer did was assign probabilities to the potential actions and compute all the scenarios. It didnt actually learn. Computers cant learn or adapt.

And honestly, if we get to that point, then we will have to deal with computers that can learn and grow, and thats kinda scary and reaches far beyond the "better guards in games" concept.

Quote:

---

Originally Posted by diffusionx

We've already discussed this and I explained it once. An "AI Engine" isnt feasible because AI is specialized per game. AI is little more than an elaborate set of if-then statements, tuned to each individual game. Even games that you may think are similar likely have very different AI to fit the need of the game in question. They share some core concepts like what you describe but thats just the low-level type stuff, all the behavior is individualized for the game's engine and programming (a guard that reacts in MGS3 will act totally different from a guard that reacts in Half-Life 2, and how they react and are told to react is in part dependent on the tools used to make the game). As for totally different games, say Warcraft 3 and MGS3 and Sims 2, not a shared bit of AI code in them. Physics is physics, a ball bouncing is a ball bouncing, developers can modify how that ball bounces by degrees but thats why its possible to make a physics engine.

---

I know we've talked about it before, but just because it's a decade or more off doesn't mean it isn't good conversation.

Quote:

---

What youre talking about is some sort of generic AI construct that can adapt to different game scenarios. This is decades away, at the very least, even if every game developer on the planet is working on this (and in any case, far smarter people, like MIT researchers, are doing AI research). Even the AI that can currently beat a Grand Master in Chess is just able to do so from brute force, not being able to learn in a way that can make that AI play Go or Game of Life well. You may remember that there was a spat because the developers fed Kasparov's moves into the AI, well all the computer did was assign probabilities to the potential actions and compute all the scenarios. It didnt actually learn. Computers cant learn or adapt.

---

What you can do is set the basic parameters, and have different modules for different types of games as they come up. So if you need AI for an RTS you can program the AI for that (perhaps using a couple of the parameters already defined to save some time). The idea should be flexible to accomodate the differences. Think about an RTS game merged with a Commandos style small squad gameplay. You'll still want individual units patrolling the streets to react like a Splinter Cell if they hear or see something suspicious, but in the same realm you can have the groups of soldiers guarding the front of the town react as a commanded unit. I'm sure there's work arounds for all of these problems.

Quote:

---

Originally Posted by OmegaFlareX

Like they said, if you want the best performance, you need a PC that will run circles around the 360.

---

That's not the way I interepreted the dev's statement. Rather, he indicated that it will look "perfect" on 360, and only the most cutting edge pc would otherwise suffice. I'm not especially tech savvy, but from what I've read, the 360's capabilities will not be matched by any pc chip sets available at the time of its release. That won't last forever, of course, but apparently PCs will have some catching up to do, and even then I wouldn't expect a perceivable jump in performance in the short term.

Quote:

---

What you can do is set the basic parameters, and have different modules for different types of games as they come up. So if you need AI for an RTS you can program the AI for that (perhaps using a couple of the parameters already defined to save some time). The idea should be flexible to accomodate the differences.

---

AI from RTS to RTS is different, though.

Like look, lets take the guard example. Lets say we are writing this theoretical AI engine, and it has a function, Patrol(). Okay, but how far does the AI patrol? When does it stop and turn around? Does it look in other rooms? What is the noise threshold and the sight threshold? All of this stuff is highly dependent on the *individual game design*. Its not dependent on genre, its dependent on each individual game. Not to mention that pathfinding and what not is highly dependent on each game's logic. When we are talking about graphic engines, well drawing a pixel is drawing a pixel. When we are talking about physics, if we want physics in a game theres a certain way we want it to behave.

But anyway, with that RTS AI example, we need to define how the AI acts when the game loads, like a build order. Well, that is different based on each game and what we want from the AI. Then we need to build how the AI will attack, but again, different in every game and different based on difficulty levels. Theres no benefit to building a "module" because every game uses different AI. Yea, there's a BuildOrder, and a Respond, and a GoToLumber, or whatever. But from there it becomes totally different based on the game.

All these modules would end up being are a couple of empty If-Then statements that would have to be filled in by the programmers of the game. Theres no point. I think thats what youre not realizing, AI is hard-coded into games now because its specialized for each game.

Quote:

---

Originally Posted by diffusionx

AI from RTS to RTS is different, though.

---

Each developer would have to tweak the programming and code for customization to their game, but if the basic parameters have been set it helps them speed up the process in defining the parameters then it would from scratch (i'm sure people tweak physics, effects, gravity and stuff like that for their game)

It just helps the developers focus more on creative and less on the technical issues hampering the experience.

Quote:

---

Like look, lets take the guard example. Lets say we are writing this theoretical AI engine, and it has a function, Patrol(). Okay, but how far does the AI patrol? When does it stop and turn around? Does it look in other rooms? What is the noise threshold and the sight threshold? All of this stuff is highly dependent on the *individual game design*. Its not dependent on genre, its dependent on each individual game. When we are talking about graphic engines, well drawing a pixel is drawing a pixel. When we are talking about physics, if we want physics in a game theres a certain way we want it to behave.

---

Okay, but in order for a guard to patrol the developer would have to define that themselves in the level design. The threshold variables would could be changed. The threshold for human hearing and sight will be defined by a standard. What is the standard human hearing threshold? Start with that, and tweak accordingly.

From there there could be different levels of threshold defined for a level 2 soldier (I haven't programmed since high school so I don't know all of the syntax). If the soldier is a boss or mini-boss they can see and hear further. If the guard hears something to the left you can tell that one guard to look, then investigate anything to the left of him. If there is a door or room on the left he could go in and check it out. It's infinite, but if all you have to do as a programmer is tweak variables and level design you're ahead of the game already, giving you more time to customize it for the experience.

Quote:

---

But anyway, with that RTS AI example, we need to define how the AI acts when the game loads, like a build order. Well, that is different based on each game and what we want from the AI. Then we need to build how the AI will attack, but again, different in every game and different based on difficulty levels. Theres no benefit to building a "module" because every game uses different AI. Yea, there's a BuildOrder, and a Respond, and a GoToLumber, or whatever. But from there it becomes totally different based on the game.

---

There are always common threads in the way people react. If someone is going to attack something there are always similarities in the way they go about it, and animation can define the difference between a woman's punch, a man's punch and a kick boxers punch. If we're talking about soldiers here the way they physically approach to attack can be similar. As these scenarios come up you'll be able to update the engine. This is how Adobe adds in new features to their software. They cover the basic functions, but none of it is perfect. They continue to add in new portions and modules as the market demands them. In doing this you can help improve AI and retain settings for use in future games without having to redefine them from scratch.

Quote:

---

I think thats what youre not realizing, AI is hard-coded into games now because its specialized for each game.

---

I realize that every game is different. The engine I'm talking about is one that can handle these differences to speed up development time. Every developer will have to spend time customizing AI for particular instances and the way they want in for their games. An engine like this will make it much quicker to do (maybe not, as you point out).

Quote:

---

It just helps the developers focus more on creative and less on the technical issues hampering the experience.

---

The thing youre not realizing here is that AI is creative. Its not the same as drawing a pixel. Its part of the creative part of the game. What youre talking about is akin to those awful graphic libraries that end up making the first-gen games all look the same.

Quote:

---

It's infinite, but if all you have to do as a programmer is tweak variables and level design you're ahead of the game already, giving you more time to customize it for the experience.

---

I dont know about you but I dont want all my games to have the near-identical AI.

Quote:

---

There are always common threads in the way people react. If someone is going to attack something there are always similarities in the way they go about it, and animation can define the difference between a woman's punch, a man's punch and a kick boxers punch.

---

There are ways they go about it in the real world but not necessarilly in-game. The logic of the game and the rules of design will dictate how they do it, not because they fit into some determined genre. Furthermore, the coding and scripting of the game will in large part dictate just what you need to do to get a guy to punch.

Like I said, yea in Metal Gear Solid 3 they notice you and shoot and in Half-Life 2 they notice you and shoot. But the differences are what matters, and in each game they are TOTALLY different, from in-game scripting to noise/vision thresholds to running away and so forth. The code for patrolling in MGS3 Im sure looks nothing like the code for patrolling in Half-Life 2 and that shows at least they are writing it for the individual experiences. A generic Patrol() function would have to be completely rewritten based on the logic and scripting of the games themselves, eliminating its usefullness.

Like I said, whatever youre proposing will end up being an empty bunch of if-then statements. Not really worthwhile. Besides, what youre talking about here, like noise thresholds and such, it doesnt take long to write. Im sure the experienced programmers can write it in a couple days.

Also, even if this theoretical engine were to surface, it wouldnt help improve AI any, because like you said the nuts and bolts would have to be customized for each game (and good AI is primarily in the details) and AI is pretty much a direct function of how many clock cycles you give it - and this is a decision that the designers make.

Quote:

---

Originally Posted by diffusionx

The thing youre not realizing here is that AI is creative. Its not the same as drawing a pixel. Its part of the creative part of the game. What youre talking about is akin to those awful graphic libraries that end up making the first-gen games all look the same.

---

No I do realize that AI is part of the creative. But AI is how a character reacts to you (or doesn't react) in certain scenarios. Some of the variables will always be constant in many situations. (a guard exploring the dark to investigate a sound) I don't think it will end up like graphic or clip art galleries because you have to physically mend the code to your game in order for the AI to react accordingly (so it won't be like plugging in something and letting it go) but if you have these basics already programmed to begin with it helps eliminate the need to redefine these for every game you try to make.

Quote:

---

I dont know about you but I dont want all my games to have the near-identical AI.

---

Me neither, which is where the customization for the experience, level, story, and style comes in.

Quote:

---

Like I said, yea in Metal Gear Solid 3 they notice you and shoot and in Half-Life 2 they notice you and shoot. But the differences are what matters, and in each game they are TOTALLY different, from in-game scripting to noise/vision thresholds to running away and so forth. The code for patrolling in MGS3 Im sure looks nothing like the code for patrolling in Half-Life 2 and that shows at least they are writing it for the individual experiences.

---

Right, which is why you won't be able to simply plug it in. The engine that has defined parameters will enable you to customize these patrolling differences quicker than it would from building everything from scratch (these similarities will be defined for you so you don't have to). Basically, it will help you make a good creative AI experience the same way Adobe Photoshop helps you create a graphic. If you aren't a good programmer your game will suffer. If you are a good AI programmer your game will benefit. If you are a bad artist or designer your designs from Photoshop will be bad, much in the same way your designs from Photoshop will probably be good if you're a good artist or designer.

These are simply tools to help you control AI faster and easier.

Quote:

---

Also, even if this theoretical engine were to surface, it wouldnt help improve AI any, because like you said the nuts and bolts would have to be customized for each game (and good AI is primarily in the details) and AI is pretty much a direct function of how many clock cycles you give it - and this is a decision that the designers make.

---

But like a Photoshop or Illustrator is helps them get to the meat of the creative process a lot faster. In the old days of print you'd have to film process everything for print. Nowadays you can do it all pretty much in Quark Xpress, Illustrator or InDesign. The only thing it did was eliminate hassle and allow artists and designers easier technical access to do what it is they want to do.

You're given a paintbrush and mask in Photoshop, but you have to customize it for a piece of work you're doing to get a good result.

And as people come up with new needs and problems for different genre types the new additions can be added in to help solve those problems quicker in the future.

Quote:

---

but if you have these basics already programmed to begin with it helps eliminate the need to redefine these for every game you try to make.

---

The "basics" you talk about are very small and trivial, like maybe .5% of the AI code. Everything else is customized. Which is my point.

I guess I'm just willing to bet more that fundamentals of how people respond to stimulus have more in common than simply .5% Finding common beads in what startles a soldier and how they jump and respond to different action will help define a toolset and set of rules to help aid developers.

I think you're missing the point. Yeah, in real life, many different things react to similar stimuli in a predictable manner, but the great majority of video games do NOT reflect real life. They're built around certain rules and certain playstyles to the point where having a common AI routine is impossible for all but the most rudimentary features, making it all rather pointless.

And Photoshop is more a toolbox than anything.