Editorial by Dan Cheer

Despite significant technological advances in gaming over the past decade, one area of games is still largely bereft of any meaningful innovation: artificial intelligence. Is it just too hard, or will this decade see a paradigm shift in our approach to realism?

Quote: IT'S BEEN TEN YEARS since the release of Black & White, Lionhead's critically acclaimed God-simulator designed to bring us one step closer to realistic artificial intelligence. By concentrating on procedural choices made by the player that were then fed back into the environment, Black & White aimed to present a world in which gamers could explore the overarching effects of their actions. Likewise, developers could push the boundaries of artificial intelligence by examining the interaction between the player and in-game characters, and use the results to build personalities. On a rudimentary level, basic player input was mapped using a reinforcement model. Should the in-game pet – referred to as the Creature – displease the player, swift justice would be enforced with a slap. Affirmation of tasks completed satisfactorily would be achieved with a stroke. It's a time-honoured way to teach the fundamental construct of any social group, after all.

Continue reading...

Is it because game reviewers are still nearly wholly obsessed with graphics?

I loved Black and White.

Over fixation on graphical niceties instead of the technical nuts and bolts?
We haven't really seen any major progression in AI for years, most of my suspicion points towards the huge shift towards online play providing the majority of interaction leaving singleplay to highly scripted events. Even Civ5 which is a primarily singleplay games has terrible AI (still after all these years) and relies on higher levels of difficulty providing boosts to the AI instead of making them play smarter.

I like playing online sometimes but would KILL for a RTS that has excellent AI that actually is AI instead of a scripted build priority list that it sticks too without any sort of adaptation.

Seems it's been relegated to the 'too hard' basket, Oblivion was supposed to have mind-blowing new AI, what happened there?
Be interesting to see if Skyrim does anything new/different in that respect.

Quote: There's no better place to start than the integration of truly intelligent computer-generated foes.

I remember one of my lecturers saying a couple of years ago "If real artificial intelligence was the moon, we'd be about as close as if we climbed a tree".

It's not like AI is something that is just sitting there but game programmers are too lazy to code it in. What we consider to be 'AI' is really just behavior that appears to be intelligent (hence the Turing test). Even with such a (relatively) low bar, we're not even remotely close to accomplishing that except for severely restricted domains. It's a problem which computer science researchers have been working on for decades. Demanding that it's about damn time we get it in our video games is lol.

Great article. Hopefully with Skyrim, Bethesda (this time around) will deliver with their 'Radiant AI' and get other developers to pick up their game. I await with great anticipation.

damn, ninja'd :P

exis wrote: I remember one of my lecturers saying a couple of years ago "If real artificial intelligence was the moon, we'd be about as close as if we climbed a tree". It's not like AI is something that is just sitting there but game programmers are too lazy to code it in. What we consider to be 'AI' is really just behavior that appears to be intelligent (hence the Turing test). Even with such a (relatively) low bar, we're not even remotely close to accomplishing that except for severely restricted domains. It's a problem which computer science researchers have been working on for decades. Demanding that it's about damn time we get it in our video games is lol.

We're not talking about sentient computers here.

There's plenty of room for improvement in the AI routines used in videogames - it's just something that developers have not considered a priority.

The soldiers in Half-Life 1 are still smarter than the enemies you encounter in most shooters today.

F.E.A.R has the best AI in shooters imo. Rather than enemies just having more health, do more damage to you etc. You actually had to think and not just sit in cover and pop out every now and then.

The gaming market bulk got a lot more casual, making smarter opponents, and making your game too hard for that large sector is going to hurt sales.

Egor wrote: We're not talking about sentient computers here. There's plenty of room for improvement in the AI routines used in videogames - it's just something that developers have not considered a priority. The soldiers in Half-Life 1 are still smarter than the enemies you encounter in most shooters today.

I agree there's room for improvement, but taking a quote from the article, "truly intelligent computer-generated foes" are just not possible right now.

crysis 2 AI is laughable but expected for a consolized game.

Murakami wrote: Is it because game reviewers are still nearly wholly obsessed with graphics?

Its not that decent AI is hard, developers just aren't focusing on it. They just spend their time elsewhere.

You can't have open world and good AI it isn't possible just yet.

I still remember in Deus Ex (which is generally regarded as one of the greatest games ever) you could basically snipe and kill everyone on the entire level one by one -sure everyone would run around for a bit but if they didn't find you they just settled back into their normal patterns.

You can however have highly scripted AI in much more linear games like ME2 the fact that you don't is just lazy programming...

As a layman I wouldn't think it would be that hard to craft good AI for a game.

SirGrim wrote: Its not that decent AI is hard, developers just aren't focusing on it. They just spend their time elsewhere.

what? Decent AI is one of the hardest solutions facing modern programming. As a seasoned programmer myself I can tell you right now that the deterministic logic (if X is true then do Y) you encounter in a lot of modern games is not the same thing as having a decent AI. A decent AI must be able to adapt and very few AI's in games can do that.

Not really a game for some of you impatient FPS arcade shooter lovers, but Arma2 has very good AI. And the community has spent endless amounts of time improving them.

They will cover each other, split and flank you while trying to suppress. Call in nearby squads to help out. Use and seek cover. Lie in wait to ambush you at their most effective ranges. Will pull back when you are hammering them. Resupply and help out their injured team mates. Select the most appropriate weapons at the time e.g. RPG18's to smash your cover.

They do pretty well to get the AI to work like they do with the sometimes clunky Arma2 engine. The good thing is that in a game like Arma there is always someone trying to improve it. In arcade shooters like COD the AI will never be improved upon after release.

The AI in F.E.A.R used to put the shits up me sometimes

Mikos wrote: what? Decent AI is one of the hardest solutions facing modern programming. As a seasoned programmer myself I can tell you right now that the deterministic logic (if X is true then do Y) you encounter in a lot of modern games is not the same thing as having a decent AI. A decent AI must be able to adapt and very few AI's in games can do that.

The thing is, it's being done before. It's not some new spectacular goal. Good AI has being done in certain games and yet developers fail to replicate what should be basic AI by now.

exis wrote: I agree there's room for improvement, but taking a quote from the article, "truly intelligent computer-generated foes" are just not possible right now.

I think you're trying to refute a point that was never intended in the first place.

Nobody is talking about making real artificial intelligences in the computer science definition of the term. Real thinking, learning machines are obviously way out of our grasp at this point in time.

Developers just need to program computer-generated foes with a wider range of behaviours to make them less predictable, and appear more realistic. There's nothing impossible about that at all, it just takes effort.

Egor wrote: I think you're trying to refute a point that was never intended in the first place. Nobody is talking about making real artificial intelligences in the computer science definition of the term. Real thinking, learning machines are obviously way out of our grasp at this point in time. Developers just need to program computer-generated foes with a wider range of behaviours to make them less predictable, and appear more realistic. There's nothing impossible about that at all, it just takes effort.

All I'm saying is, words like "truly intelligent" have an actual meaning. And it goes beyond 'appearing more realistic'.

werewolves? wrote: As a layman I wouldn't think it would be that hard to craft good AI for a game.

I was writing something here but it was just turning into a compsci wankfest. How would you go about programming an AI character? I'm just not sure I understand how anybody could see it as a trivial task.

exis wrote: I was writing something here but it was just turning into a compsci wankfest. How would you go about programming an AI character? I'm just not sure I understand how anybody could see it as a trivial task.

I'm meaning for a high-end professional working for a major developer whose job it is.... it shouldn't be that hard to at least give the appearance of good AI, even if it's based on a lot of scripted actions.

I suppose I'm thinking about it from a first person shooter perspective where it would be based on AI for a given scenario.

Like staying in cover and trying to suppress attackers, when there is only one or two left retreating like hell to other allies. Not doing stupid shit.

Can't be that hard to code retreat or regroup orders, based on how many of the AI buddies are killed, suppression etc etc

Sanguinius wrote: relies on higher levels of difficulty providing boosts to the AI instead of making them play smarter.

This fills me with rage.

UnholySoldier wrote: The gaming market bulk got a lot more casual, making smarter opponents, and making your game too hard for that large sector is going to hurt sales.

See: Difficulty option

Options --> Game --> Difficulty
Is where it is usually found.

SirGrim wrote: Its not that decent AI is hard, developers just aren't focusing on it. They just spend their time elsewhere.

Modders show the truth of this statement

exis wrote: I was writing something here but it was just turning into a compsci wankfest. How would you go about programming an AI character? I'm just not sure I understand how anybody could see it as a trivial task.

Find some way of adapting the question and applying actual intelligence. Ie. Labcat hunting mouse. Surely the hunting instinct would be superior to human inputs.