Gameplay and AI

[Rodion Romanovich]

What do you consider most important in terms of AI and gameplay? A lot of modern games in fact have much more code and budget for AI and gameplay than all older games, but they're perhaps spending the time on doing the wrong stuff? For example, in recent sneaking games the NPCs can move around doing natural-looking activities when they've not discovered the player, and so on, but in return the behavior once they find the player may be less believable. To what type of game elements, in your opinion, should the most effort in AI and gameplay go? Is the example above waste of resources or bad prioritization, in your opinion?

[Kekvit Irae]

Friendly AI has to adapt to the player's movements and orders. Enemy AI has to adapt to the player's behaviors and learn accordingly. Those are the fundamental rules of good AI. Game programmers can take a lesson from the military AI in the original Half Life. They flank, flush out, throw grenades when the player is hiding, and adapt.

Gameplay has to be precise, but not touchy. Crate jumping puzzles are fun, but only when they are sparse. Optional exploration of a level is vital, but not fun when every level looks the same (Assassin's Creed). Innovations to new ideas are good, so long as they integrate themselves into the game well. But most of all, gameplay has to be FUN.

[Martok]

(Note: I play primarily strategy games, so just bear that in mind when reading this.)

I think it's important to make sure that when contemplating adding new features to a game, that the AI can utilize said features with a reasonable degree of competence. There are too many examples of game features -- RTW's diplomacy system and MTW's mercenaries, to name only a couple -- that the human player can take advantage of, but the AI cannot (or can make use of in only the most rudimentary fashion). And while I readily admit that a human player will almost always do a better job than computer-controlled opponents of using certain features, one can at least avoid implemeting things that the AI can barely use at all.

[Ramses II CP]

Good AI should be aggressive rather than passive. It's all too common in gaming to have the 'creep forward until I can snipe him' attitude sneak in. I prefer an AI that forces the player to adapt by actively attacking him.

The next generation of AI will, hopefully, be more flexible and adaptive on it's own; i.e. if you're using the sniper rifle a lot the enemy should set more ambushes and look to surround you or engage at close range. If you're shotgunning through hordes then the AI needs to step back and pull it's own sniper rifle. Cooperation walks hand in hand with that adaptation.

I'm waiting for the AI that will pin you down with HMG fire and execute a leapfrog advance in pairs while tossing grenades to overrun your position.

[Fragony]

Halo's AI is particularily good, they adapt, flank flush, still the best out there.

[Bob the Insane]

I'm waiting for the AI that will pin you down with HMG fire and execute a leapfrog advance in pairs while tossing grenades to overrun your position.

The enemy soldiers (not the aliens) are almost up to this in Crysis...

At least they have a positive aggresive response to your attacks and never really return a a default, nothing happened state...

[Veho Nex]

kinda sounds like RTS ai from opposing fronts and Company of Heroes just a little except its an allied blitzkrieg that i can never seem to counter

[Geoffrey S]

Problem appears when they games gain in complexity of environment and player abilities, and the AI requires that much more ability to deal with the numerous variables. Programmers can get around that through very careful scripting and telling enemies what to do in certain situations (done particularly succesfully in Halo 3, for instance), but it's very easy for players to start recognizing the patterns and force the AI to do certain things.

[TinCow]

I think the main problem with poor AIs is the way in which the games themselves are created. I have no factual basis for this theory whatsoever, but I suspect that many developers create the game system first, then attempt to create an AI that can handle it. Naturally, the designers will want to include many 'cool' features that their players will enjoy having access to. When that is done, they then turn to the AI team and say, "Create an AI that can do all of this." Unfortunately, coding an AI is extremely difficult and the more complex the game rules, the harder it is to get a good AI system.

I think the best way to create an AI is to consider an fundamentally important part of the design of the game from the very first step. Every addition to the game should be accompanied by the question, "Can the AI handle this rule?" If the answer is yes, it is included in the game. If the answer is no, it is not. That will certainly result in less complex games, but I think it would result in an overall improvement in gaming entertainment.

Of course, I could be totally wrong about how games are developed. If they are already doing what I suggest and still getting poor results, then I'm not quite sure what the solution is.

[Fragony]

I think the best way to create an AI is to consider an fundamentally important part of the design of the game from the very first step. Every addition to the game should be accompanied by the question, "Can the AI handle this rule?" If the answer is yes, it is included in the game. If the answer is no, it is not. That will certainly result in less complex games, but I think it would result in an overall improvement in gaming entertainment.

Take FEAR, the AI is just about flawless because of just that. Leveldesign is wrapped around the (impressive) skills of the AI, saves an otherwise soso title. AI is the most important aspect of an excellent shooter imho.

[Husar]

There are also concerns like system requirements, an awesome AI would use a lot of CPU power, maybe multi cores can help with that in the future as you can have one or two cores just to calculate complicated AI routines.

[Fragony]

There are also concerns like system requirements, an awesome AI would use a lot of CPU power, maybe multi cores can help with that in the future as you can have one or two cores just to calculate complicated AI routines.

How old are Half-Life and No One Lives Forever (best shooter ever made by the way)? These games are technically ancient but have better AI then 99% of the other shooters even today, FEAR and Halo top them, and that's it. I think it just asks for some more dedication, nobody would ever say Half-life and NOLF were underdeveloped when it comes to other aspects.

[Bob the Insane]

We also have to recognise that horror of the modern games industry...

Making things easier...

Even if they could produce an FPS AI that would be practically identical to playing MP against living opponents without the human error, would that be fun?

I recall Rainbow Six 3: Raven Shield playing MP co-op against the AI on Hard... All the R6 guys were human and all thr tangos AI...

They would kick our collective butts round after round...

I doubt that would occur in newer games...

[TinCow]

I doubt that would occur in newer games...

That's true. There seems to be a tendency towards skipping AI altogether and substituting a human opponent. That is, of course, a wonderful thing since I have never seen any AI ever that is more fun to play against than a human. However, the game companies have to make that multiplayer ability easy to use and fun to play. I wouldn't be anywhere near as annoyed with the TW AI if it was possible to play the Grand Campaign against a human opponent in a convenient manner and with the ability to fight battles. Hotseat does not even really qualify as multiplayer in my book.

[Husar]

I actually had strategic AI in mind Frag, shooters are somewhat easier I think unless you want the AI to use dynamic objects as cover etc which may be a bit harder but not impossible.
Concerning those Rainbow games, I remember coming in behind some guy who was some 30 meters away and in the moment I sneaked around the corner he turned and headshot me, that's an aimbot, not an AI.

[Martok]

I think the main problem with poor AIs is the way in which the games themselves are created. I have no factual basis for this theory whatsoever, but I suspect that many developers create the game system first, then attempt to create an AI that can handle it. Naturally, the designers will want to include many 'cool' features that their players will enjoy having access to. When that is done, they then turn to the AI team and say, "Create an AI that can do all of this." Unfortunately, coding an AI is extremely difficult and the more complex the game rules, the harder it is to get a good AI system.

I think the best way to create an AI is to consider an fundamentally important part of the design of the game from the very first step. Every addition to the game should be accompanied by the question, "Can the AI handle this rule?" If the answer is yes, it is included in the game. If the answer is no, it is not. That will certainly result in less complex games, but I think it would result in an overall improvement in gaming entertainment.

Of course, I could be totally wrong about how games are developed. If they are already doing what I suggest and still getting poor results, then I'm not quite sure what the solution is.

I completely agree with you, TinCow. You actually made the precise point I was trying to make myself, but you were far more eloquent than I.


[That's one reason why I enjoy Galactic Civilizations 2 so much. Its AI is so superb in part because Brad Wardell (the game's creator & main AI programmer) refuses to add features that the the computer-controlled opponents can't handle. No matter how "cool" and/or exciting a potential new feature may be, he won't implement it unless he feels he can program the AI to utilize said feature competently.]

[CountArach]

Good squad cohesion is one of the most important things for me. The AI for example in F.E.A.R is great because they'll chuck a grenade in, then rush you with several men, while simultaneously setting up a flanking attack.

[Fragony]

We also have to recognise that horror of the modern games industry...

Making things easier...

Even if they could produce an FPS AI that would be practically identical to playing MP against living opponents without the human error, would that be fun?

I recall Rainbow Six 3: Raven Shield playing MP co-op against the AI on Hard... All the R6 guys were human and all thr tangos AI...

They would kick our collective butts round after round...

I doubt that would occur in newer games...

AI cheats, they always score a headshot.