Support NeoGAF

[entremet]

Every new generation there's always talk about greater power available for stuff like sophisticated AI, however, it never really pans out.

Most AI is pretty rudimentary, easily exploited, or just downright cheats to create challenge--see rubber banding in racers.

I'm not a programmer nor do I have a background in how AI works, but I do have theories.

-Really good AI is extremely computational taxing. Although it happened so long ago, Deep Blue beating the World Chess Champion at the time, while a landmark, took huge undertaking in both labor hours, ingenuity and hardware for the Deep Blue team.

-Creating competitive AI may not be worth it to game studios, especially since MP is more prominent and the return on investment is so low. So you can't create bullshots from AI.

-And similar to the point of above, limited resource in terms of labor and project priorities.

I've been playing games for over 20 years and I can't think of any game where I was impressed of the AI. It's usually rudimentary.

Is chasing AI something that is just too difficult to implement in an authentic way due to lack of computational power, resources, and return on investment?

 

[Calabi]

Their acting stupid so they can catch us unaware.

 

[DragonSworne]

What do you mean by suck?

Game AI is designed to be fun. Are you saying the current AI is not fun? If not, why? What can be done to make it more fun?

Simply saying it sucks doesn't make sense. If you want smarter AI that wins all the time I don't see how that will be fun.

 

[Wavebossa]

You pretty much answered your own question

It is extremely taxing the more robust it is (to program AND to compute). Usually the shortcut used is to just have the npc "cheat" instead of having them play by the same rules introduced by the game's universe.

In addition to that, the more robust the ai is, the higher chance there is for glitches/bugs. This may make for hilarious "Let's Play"s but it would also reduce the overall polish of the game.

 

[Krejlooc]

AI doesn't work like people think it does. It's just a state machine. There is no learning going on.

We make states that can be defeated, because AI that is too "good" isn't fun.

 

[JayEH]

Really good AI is pretty easy to make I assume. Fun AI to play against is harder.

 

[sn00zer]

Third person shooter AI isnt dumb, you just get a god-like view of everyone moving around. If the camera was obscured when you went into cover, I imagine it would be a much different experience.

 

[Figments]

There's a balance between smart AI and fun AI. While some people like playing against an incredibly intelligent AI, I would hazard a guess that the majority of people wouldn't find that particularly fun.

 

[entremet]

What do you mean by suck?

Game AI is designed to be fun. Are you saying the current AI is not fun? If not, why? What can be done to make it more fun?

Simply saying it sucks doesn't make sense. If you want smarter AI that wins all the time I don't see how that will be fun.

I'm talking about competitive games.

I love fighters for example. But playing the AI is never as satisfying.

Also for expert runs. AI is easily exploitable.

 

[Nastrodamous]

Because better AI can't be shown in screen shots and on boxes

 

[John Kowalski]

what if SHODAN comes from a game

can't have that

 

[Brazil]

Investments in graphics generate more profit than investments in enemy AI.

 

[Laurentius]

e9iOE

 

[Captain Tuttle]

For the same reason framerates are the way they are; people prefer pretty games over better performing ones.

 

[commedieu]

I can't imagine its worth the investment.

Multiplayer pretty much gets rid of the need to develop good A.I. Twitch shooters need a basic A.I that allows you to shoot them.

What A.I masterpieces have been successful to where it is an issue that a developer feels they need to improve, or be left behind?

 

[thelastcrusader]

Man I'll never forget Army of Two 40th AI on Hard. Those fuckers would drag wounded enemies out of your LoS, revive them, and coordinate to flank you. My brother and I were always like WTF!!! We'd always end up getting split up due to the chaos.

 

[Messofanego]

It's not a big selling point to gamers like other technical things are.

 

[DragonSworne]

I'm talking about competitive games.

I love fighters for example. But playing the AI is never as satisfying.

Also for expert runs. AI is easily exploitable.

Ah, I see.

I think investment in challenging AI stopped since online multiplayer became more common. If you look at games on the N64 and PSX, you'll find AI that easily outpaces current gen games. Games like Perfect Dark have far more impressive enemy AI then you see now in new shooters.

Companies assume you'll play online to find more challenge.

 

[BeeDog]

I won't have any hopes for realistic AI in a game any time soon, but what grinds my gears more is when a gaming series regresses for each subsequent game. Take the Killzone games, for example; Killzone 2 had really excellent AI, but both KZ3 and especially KZ:SF took several steps back. I can't for the life of me see the reason for that.

 

[Fbh]

Because better AI can't be shown in screen shots and on boxes

This.


Also, by making enemy smarter you asume the risk of making the game less fun.

Single player shooters would be impossible with waves of enemies that play like humans would. The fact you can exploid it, easily get heashots or find out patterns you can take advantage of is what makes the often used "1 vs many" scenario fun to play

 

[Ninja Dom]

It's been frustrating. I'm a big fan of fighting games and character action games and I've always wanted the AI to be more difficult through being more skilful and not by simply doing more damage to the player.

My answer as to why AI hasn't developed as fast as hoped is that it's simply that devs cannot be bothered with trying to improve it. The games industry has always suffered from a "Good Enough" attitude. If what they do still sells millions of copies then why improve?

Another reason is that some players aren't too interested in challenging AI and harder difficulties. Consumers just wanna consume games as fast as they can without paying too much attention to the intricacies.

 

[cormack12]

Wasn't there an article a while back (maybe from Guerilla) that said basically all the various teams fight over resources and some teams are obviously partitioned more. I think the AI guys end up with the short end of the stick - didn't they get like 4MB RAM for AI for shadowfall for example. So, it's just not seen as a priority I guess.

 

[Sapphire Dreams]

I made a thread about it a while ago and yeah the gist of it that making AI while it is hard, it's the balance between making it fun and fair that makes it a tricky thing to deal with.

 

[Violence Jack]

AI varies. I've played games where I've found that the AI gives just the right amount of challenge while remaining fun, and I've also played against the "guy just stands there and lets me shoot/hack/beat him to death" type of AI.

I'm afraid the answer lies in the middle. Challenge me, make me think, but be fair about it.

I'm confused as to what you want. Are you saying that you want AI that behaves more randomly or more human-like? In my opinion, that could just about be the same thing.

 

[tbm24]

I'm talking about competitive games.

I love fighters for example. But playing the AI is never as satisfying.

Also for expert runs. AI is easily exploitable.

Mostly cus Fighters are designed for player versus player first, AI second if that. That's why the harder difficulties they are always super cheap and react in ways humans never could. Though, a drivatar type setup for the next Street Fighter would be quite dope.

 

[Tobor]

Personally, I'm fine with stupid AI. I like to relax and have fun. If the AI is too good, it's stressful and less fun.

If I wanted to get killed every 10 seconds, I'd play a competitive shooter against humans.

 

[Guile]

https://www.youtube.com/watch?v=CpBimU0vcJg

The GOAT.

Funny how games these days fail to match up with AI from 9 years ago.

 

[entremet]

Even though rudimentary now, I remember being impressed at the soldiers actions in MSG2 versus its prequel.

Opening, lockers, calling for reinforcements.

Again, it's rudimentary now, but coming off MGS1, such a big leap.

Mostly cus Fighters are designed for player versus player first, AI second if that. That's why the harder difficulties they are always super cheap and react in ways humans never could. Though, a drivatar type setup for the next Street Fighter would be quite dope.

High level difficulties in fighting games is basically the computer reading your inputs and reacting accordingly.

 

[Fragamemnon]

AI doesn't work like people think it does. It's just a state machine. There is no learning going on.

We make states that can be defeated, because AI that is too "good" isn't fun.

Great post.

AI in games is all about providing a means to produce enemies and competitors that make the gameplay interesting. It's not about being the best, or making optimal decisions-it's about ensuring that the overall outcome results in a satisfying player experience.

There's another aspect going on that making an AI that is plays the game well gets exponentially more difficult and resource intensive as the complexities of the game increase. We've had "good" AI for games with simple rulesets like Chess for some time. But you can't just take the same kind of approach to making an AI for a game that has a lot more moving parts (and is thus much more difficult to simulate outcomes) for a game like Starcraft 2. The way out of this is to have games where the player and computer controlled competitor(s) aren't playing by the same rules, but that is still something a lot of people aren't comfortable with.

 

[Skux]

Most enemies are killed before they can do anything smart.

I remember that Minerva fan level for HL2. All the guy did was add some hp to the combine soldiers and put them at range with a lot of space to work in. They would flank and throw grenades to flush you out. They seemed a lot smarter but this is what they would have done anyway, the only difference was that they were actually alive long enough to do something.

 

[Alcoholikaust]

https://www.youtube.com/watch?v=CpBimU0vcJg

The GOAT.

Funny how games these days fail to match up with AI from 9 years ago.

Fear was soooo good.

Shout out to Halo Combat Evolved and Half Life 2 as well

 

[Omega]

What do you mean by suck?

Game AI is designed to be fun. Are you saying the current AI is not fun? If not, why? What can be done to make it more fun?

Simply saying it sucks doesn't make sense. If you want smarter AI that wins all the time I don't see how that will be fun.

or maybe AI that challenges you?

There's games that would actually have some depth to them but it's never necessary to explore the games mechanics.

For example off the top of my head, Batman. You have like 20 different attacks you can do but you never need to do more than 3 of them unless you want an achievement. Why not get better AI that does more than circles around me waiting their turn to get beat up? Make me have to quick batclaw a guy from grabbing a gun or maybe when they throw objects, have me need to throw the batarang to break it or stop it.

Nope, the whole game is just normal attack and counter. maybe an evade here and there. Other than that, outside of required sequences you'll never need to use any of the other gadgets or attacks.

 

[Arc Christelle]

What games have you been playing? Fucking Payday 1 has been making me and my friends its bitch lately.

 

[Muppet of a Man]

Games are not challenges anymore. They are "filmic" experiences.

 

[Sn4ke_911]

MGS2 still has one of the most impressive AI's in any game.

 

[xerzewatt]

Take a look at Shamus Young's blog post (http://www.shamusyoung.com/twentysidedtale/?p=19327) about AI in general and Thief AI in particular.

The cost of animations is a usually overlooked point. Even if your NPCs are really smart will you have the gameplay, graphics etc. to communicate that to the player? Otherwise it will look like your AI is cheating.

 

[Messofanego]

Wasn't there an article a while back (maybe from Guerilla) that said basically all the various teams fight over resources and some teams are obviously partitioned more. I think the AI guys end up with the short end of the stick - didn't they get like 4MB RAM for AI for shadowfall for example. So, it's just not seen as a priority I guess.

That is messed up. What was it like for Killzone 2?

I think Alien Isolation has been the only recent game along with RAGE that advertised its superior AI.

 

[JoeNut]

Apparently alien isolation has great Ai

 

[Pocky4Th3Win]

It's basically because people don't find it fun if the ai is too good and often say it's cheating if it is. Kotaku had a great article a while back.

http://kotaku.com/maybe-we-cant-handle-smart-enemies-in-our-games-1542300804

 

[block tower]

honestly whats the point of really smart AI in games, difficulty? i could think of more challenging ways to make a game harder instead of making tge AI smarter. if you want a true challenge in most games then multiplayer seems like your best bet. nothing harder than playing against other people who actually think and play like people. now i definitely get the frustration when playing against "dumb" AI even on a hard mode. but im not a programmer so im not sure how hard is too hard for AI, even on a hard mode, before the game looses its fun.

 

[Interfectum]

Graphics sell more games. AI does not and is much harder to test and code.

 

[commedieu]

Apparently alien isolation has great Ai

Alien Isolation has great everything.

BUT, in Alien Isoation A.I is hampered by having to start/stop animations. If you change up an A.I's choice, you can watch it stop an animation, then start another one. Its quite lame, but has only happened a few times. Reminded me we've got a ways to go still, but its good enough to stay entertaining, and you can use the A.I to your benefit with multiple A.I enemies. Seems they will go after eachother if around eachother.

 

[Messofanego]

Is it a priority? I don't see many critics calling out a game for having mediocre or bad AI, and gamers don't seem to either in enough of a volume that developers have to prioritise resources for it.

 

[Inuhanyou]

Halo has always had great AI, which is surprising going to how dumb the AI is in destiny, pure bullet sponges. Reach was the last game i thought to have truly exceptional AI.

Its not something that can be solved just with a stronger CPU, it has to be for the most part done at a manual programming level, and that takes time effort and skill to implement

 

[Air Zombie Meat]

I've always thought it was because it's difficult to do and doesn't give a good return on the investment required.

 

[tryagainlater]

Because AI isn't graphics and that's all people care about. Plus, it's probably really hard or something.

 

[ShinMaruku]

Those who are in artificial intelligence are finding making intelligence is extremely hard. Even for Google and IBM, would you really expect such stride from Game Developers?

 

[kencey]

Good question, AI doesn't seem to evolve much over the years.
I think we'll be stuck with NPCs forgetting everything in seconds.

"I guess it was just my imagination"
https://www.youtube.com/watch?v=lHbF9ynGZV0

 

[Loofy]

I usually play on medium, hard if its a naughty dog game. Dont think I need harder AI.

 

[Messofanego]

Take a look at Shamus Young's blog post (http://www.shamusyoung.com/twentysidedtale/?p=19327) about AI in general and Thief AI in particular.

The cost of animations is a usually overlooked point. Even if your NPCs are really smart will you have the gameplay, graphics etc. to communicate that to the player? Otherwise it will look like your AI is cheating.

It's basically because people don't find it fun if the ai is too good and often say it's cheating if it is. Kotaku had a great article a while back.

http://kotaku.com/maybe-we-cant-handle-smart-enemies-in-our-games-1542300804

Thanks for these, will read.