Why videogames never have good AI?

Why videogames never have good AI?

AI in general isn't as smart as a human yet. What the fuck are you on about, OP?

Good AI doesn't sell a game. I would stop believing that story about how the devs of tLoU scrapped out the AI because of muh cinematic experience and rushed casual reviewers if tLoU stop being treated like the second coming of Christ.

This

I like to think it's because designers are rarely programmers and AI development HEAVILY impacts gameplay, if the programmers and designers aren't joined at the hip you'll get a fun game with some unpolished tech or a tech demo that isn't much of a game (everything HE WHO CONSORTS WITH BEASTS does).

Games are an orchestra of different talents.


Good gameplay doesn't necessarily sell games either, there's a wide range of appeal to video games.

Do you have any idea how fucking hard it is to write AI? The more you write the worse it gets. It will either be obtuse with overly situational responses which make it look retarded as soon as anything weird happens, or overly simplistic stupid-proofed AI that just follows a few major cues and otherwise performs vague behavior in an attempt to fool the player into thinking it's doing something. Nearly all devs tend toward the latter model because it takes a lot less time and processing power, and most of the time it functions just as well. Actual good AI will have to wait a while, because a lot of experimentation with genuine AI (not fucking university science project-tier situational response robots) which will go through many stages and look like dogshit at first. Big devs and publishers won't take the risk to gain the practical experience, and even if they did, players would a) shit on it for being bad and naturally have little-to-no insight on how to improve it and b) praise it like good little drones, which of course defeats the purpose because it provides no meaningful feedback. The most we'll get is trash AI that looks really smart in one extremely specific context to grab headlines about how CUTTING EDGE and REALISTIC it is, because the gaming press knows people are too immature to care about the process and just want a golden pheasant to worship.

tl;dr: Greedy uncreative devs + mindless entitled gamers = no point in trying to pioneer better AI.

Want to back that up with a game that had good AI and didn't sell?

You forgot:
-It will be a cheating bastard that is fed your inputs, basically reading your mind
-It will be twice the cheating bastard and the same rules won't apply to it (racing game speed boost when you're too far ahead of opponents)

Because AI doesn't exist in any video game, it's just scripting.

Let me rephrase that, you don't need good AI to sell a game.

You don't need good sound, physics or graphics to sell a game either…

Mr Gimmick had some pretty advanced enemy behavior for an NES platformer, and it sold jack shit everywhere.

Didn't Mr. Gimmick release on NES after the SNES was out?

Yeah, in 1992. The Super Famicom was already out for 2 years in Japan, and people didn't care for it because they wanted the flashier SNES games.

I get you OP, this is why my game will have great AI. Well, at least when it's finished… I have to redesign everything in it to make it better, and I couldn't/can't work on it for quite a long time. But the AI will be a primary focus in this game, the NPCs are persistent, active at all times. They'll have their own desires to kill, sell, make friends and enemies, guilds, parties, hunt bosses, do quests and all of that. They are just another player like you, so they have to be smart enough to play the game on their own.

Return on investment, it's hugely expensive to develop decent AI and you won't see any return on that money.
From a business standpoint that money would be better spent on graphics, E3 demos or marketing .

Because AI takes effort

That sounds really, really, cool, user. Almost sounds like you're making a spiritual successor to STALKER there.

Thanks man. I remember stalker used some zone system for it's AI, so if I'm not wrong then yeah, the AI will be simulated like that but globally.

Oh, yeah, the A-Life AI system of STALKER was to be a very major focus of the game, global simulation and everything - back before the giant cuts to the game that comprised the last half of its development process it was a development hype line that NPCs would be able to follow the game's main quest just as well as the player character would. Hence the comparison. It'll be really cool to see that A-Life concept really brought to life the way you describe.

...

man someone on an anonymous image board espousing arguments perfectly in-line with the narrative, it must be true

Playing Bioshock made the idea of devs designing games for the mentally handicap in mind not that unbelievable. Besides it changes very little.

Maybe they thought it was just a gimmick.

More importantly you don't need good AI to make a good game in the first place.

Because it doesn't exists.

doesn't change the fact it's obvious bullshit

Isn't it because AI in general is too basic to do jack shit of interest? Instead it's just scripting?
You're question should be why are so many devs hacks at it. Especially strategy game AI.

Making an AI is extremely difficult. It takes computer science and math PhD tier knowledge and skill to even approach making anything adequate by modern standards for any complex game. Go try making an AI for a fucking tic tac toe and see how long does it takes for you to realize you made a total piece of shit of an AI despite the game being as simple as it can possibly get while still being a game of skill rather than chance.

I'm not sure if tictactoe is all that great of a choice for your example. I can see it as a lengthy effort, but not exactly difficult. Maybe I'll try it later for fun.

FEAR had great AI because it would flank you and take cover but the AI itself was actually very simple. You could do an approximate build of one of their AI builds yourself.

When in view of player, move away from him continually away while shooting back, when behind cover turn and run. Shit like that for AI.

Normally this would make for some cowardly AI but if you funnel them through a series of loop corridors or designed levels it can make for a formidable enemy because they'll loop around to flank you as you advance and shit like that. And that's where the actual genius lies in games, level design that takes advantage of AI.

Isn't fear's AI just a standard AI that yells out it's commands? I do heavily agree with how a good AI should be designed in a way that the level design makes the most out of it.

It IS a standard AI with nothing special to it compared to other AIs of it's time, but the difference was ENTIRELY via level design.

It's like looking at a pipeline and intentionally giving it a leak or two so the enemies funnel themselves in different ways. You can set events that funnel enemies in, turn off access to certain pathing nodes for each in their code when they've been "let loose" or even when the player goes past certain points or makes a certain amount of kills to change behavior and tactics along with it.

F.E.A.R. wasn't genius work either, it was just work with a little elbow grease and passion, which is what we lack in today's industry.

Are you trolling the people that don't know how to program? I'm pretty sure you could even draw a state diagram for an optimal tic-tac-toe strategy in like ten minutes, much less code one.

Well, what's stopping you from doing it if it's only going to take 10 minutes?

sounds like you don't play a lot of games that actually capitalize on unique AI behavior. Play uncharted 4.

What do you mean by good AI?
One that can read your inputs and shut you down?

Why OP cant into grammar?
Why OP a nigger?

…Is there some reason that I would?

I don't think you can "play" a movie.

filtered

HA

That's a beautiful reaction pic


You can always make multiple simplistic courses of action for an AI and have it roll a pseudorandom number to choose on the fly a new course of action to trick the player into believing the AI has other plans.
But really, AIs are really fascinating. I remember reading about Crash Team Racing's AI following a predetermined course while outside of view of the player, then letting the few characters closest to him play much more aggressively and with logic and almost putting everyone else on a halt so as to rubberband it without the player noticing it immediately.

Not obvious bullshit just completely unverifiable. It isn't out of line from a company that fired its lead writer for not being feminist enough.

Believe it or not but the Halo games of all things have pretty good AI, or at least CE and Reach does.

I made a perfect tictactoe AI once because just making a tictactoe random-move AI for an assignment was boring.

Tictactoe is actually a really simple game to make an AI for, that can be summarized in a few rules:
1. If you can complete a line of 3, do so.
2. If you can prevent an opponent's line of 3 from being completed, do so.
3. If you can create 2 simultaneous sets of 2/3rd complete lines, do so.
4. If you can prevent an opponent's 2 simultaneous sets of 2/3rd complete lines, do so.
5. Get the middle
6. RNG for the rest cause it don't matter
BAM unbeatable AI fuck you.


The REAL answer to OP's question is this:
Because good AI isn't fun to play against.
Either the AI has a flaw you can exploit, making the AI dumb as fuck; or the AI has all its flaws covered, in which case you're just fighting against a (near) unbeatable opponent and most likely need to rely on gambling. Remember that whilst you take time to aim your shots or plan moves ahead, the AI can instantly do everything it wants AND weigh out the odds of every single fucking move you can possibly make against it, throw your previous decisions into a neural network to guesstimate your decision making on top of that and essentially become the most unfun multiplayer match you've played: but then for every single match every fucking time.

But then you're just adding flaws on purpose again which are easy to exploit, making the AI look dumb as fuck.
There's no way around this. The in-between of good and dumb AI doesn't exist.

False equivolence GalCiv sells decently and it is all about it's awesome AI as part of it's 4x package. But ask any veteran to Civ games and they'll tell you that the AI is good but it is completely "vanilla" or "bland" or "boring" because it's perfect in doing everything right and actually thinks and strategizes as it's supposed to and therefore ends up being sterile cause it never just says to itself "fuck this" and just do something retarded cause it's bored. Then there is how trivially easy it is to make an NPC do something with extreme and utterly perfect precision and timing always without needing to think it because that's exactly what a computer is good at and it doesn't need to be smart to do it. Even without the gimp that sterility in it's patterns is always readily apparent.

AI doesn't operate on smartness until it actually learns, even then it is closer to operating on base instincts than understanding. Putting some learning to AI in video games is more work than it is worth.

You should try making an AI to an infinite board 5-in-a-row tic tac toe.

Half Life 1 has pretty good AI. The soldiers will flank you when possible. They'll throw grenades if you don't fight them closer. I'm sure there's even more trickery to them. There's also how houndeyes are dangerous in a group since the damage amplifies. Half Life 2's AI seems a bit silly due to physics being involved. You can hide behind a soda can and be safe.

Why don't you do it?

...

Hey, it's where the money is. Can you blame me?

Because you're defining 'good AI' as intelligent, learning, adaptable, and simulation-oriented rather than a collective direction and toolkit used by obstacles and enemies in your game to fit encounter and level design.

Games like Doom and Quake already have all the "good AI" they need, the monsters themselves are just another facet used by mapbuilders and modders; rather than relying on them to justify their own design, placement, etc.

If you want AI to "learn" then you have to define player actions (usually by harshly limiting them), what actions they can recognize and react to (usually by only interacting with a few prominent examples), severity and priorities (which just makes the AI 'bad' once the player learns about and manipulates them), and incorporating this into their already implemented behavior and immediate course of action. It's a tedious process that's easier and easier to circumvent or simply be overlooked the more tools and opportunity is presented to the player.

Tim Schafer sure lost weight.

Making AI itself is very easy, you just change the NPC's behavior depending on context. Making it interesting is more tricky, but making it good shouldn't be that difficult. It certainly doesn't require 6 dimensional neural networks that look like mozart's musical notations.

You just need to give the NPC enough behavior to make it compelling, and toss in some RNG to prevent it from being precisely predictable. For example if the NPC runs away, it doesn't run to the best place, but randomly picked from possible ones, and with some random variation to exact position. Then you can add variations to the running away, for example if there's another NPC nearby, they could run and tag team with the other NPC, both trying to approach you from different sides. That alone is "very" simple to do, but it could create interesting AI behaviors. Then you do similar things all over the place until it's interesting enough.

I think the most likely reason why this doesn't happen, is because different kinds of NPCs would require different kinds of AI and tactics, thus it would take too much effort in the eyes of money driven studios. It's hard to sell gameplay, but easy to sell graphics and flare.

The subparness of the Combine's AI is because of a combination of unavoidable glitches like being able to hide behind the soda can because they couldn't really figure out a way around that at the time that wasn't to complicated when Source was first made and the Combine being really really weak in pure physical attributes like HP so the Combine die before they can actually carry out their tactics. The HECU marines that you praise so highly though is just hard scripting a tree branch of "if W then do X, if Y then do Z" and some of those tactics you praise the HECU marines of doing is just scripted events they can and will fuck right up immediately. The Combine's AI is in fact based on this scripting and was expanded upon greatly not only being bigger squads (16 but the optimum is 2-6) but the Combine are capable of actually making plans and utilizing tactics without a scripted sequence telling them to but again they die like chumps before anything can come of it. Another flaw of the Combine's AI is the limitation of how many actions the Combine can do as a group while remaining independent the HECU don't have this flaw because they're retarded and as a result will stop and shoot you because the script broke so the limits of the HECU's scripting doesn't become apparent until the more advanced Combine show up.

...

Because graphics cards don't include a logic processor.

I've fucked around with the combine AI in garrysmod, with some hackery and attribute editing they can actually be made extremely brutal
they will search random areas of the map until they find you, utilize grenades correctly, snipe from windows and actually group flank you if you hide (ie they will move to cover every possible angle so you no longer have any effective cover to hide behind)

hl2 devs just offset their smart AI with extremely subpar HP and damage like this guy said, I imagine they were probably killing the game testers too often in their earlier state

I don't think you understand how costs in a video game development budget goes, the guy coding and programming the AI is being paid on a salary (AAA) and gets a percentage of the cut from the game selling (smaller dev teams) you're not going to adjust this salary just because he goes from developing the AI to developing graphics. In fact the graphics and marketing ends up being bloated when you concentrate on it at the expense of practically everything else.

These fucking videos are shit, it ruins videogames for me.

fun fact: aliens colonial marines had AI that could have potentially been actually incredibly clever, same with fallout 4. in ACM's case, the guy making that un-suck mod changed a couple of hex values and the human AI now only communicates data when they actually receive vocalization from another AI human, ergo if they yell "I see him over there" they're actually sharing that piece of data, and if they're alone or out of earshot of one another they can't share data and get exponentially less effective. in fallout 4's case, a modder changed literally one line of code to disable the AI's retardation and all of a sudden they start playing their "class" properly - characters with grenades will throw them more, characters with guns will try to suppress you, and groups will actively try to flank you or flush you out of cover

let that sink in, two of the worse games released in recent memory almost had genuinely interesting and intelligent AI, but were either intentionally hamstrung or were pushed out the door in an almost-acceptable state

Basically.

I don't know if you are aware how software works but you can copy the Ai and it's "experience" infinitely. You can just put it in you game and it already learned from all the play testers and devs. Still on that fact it's not good.

youre so fuck of shit guy. if you made the greatest ai nobody would get past the first stage and you wouldnt have to make the rest of the game

Threads like these are always some incoherent mess because people confuse scientific AI with (complex) conditionally scripted game behavior.

wow this is a perfect example of people being way out of their depth. Tic tac toe has an optimal strategy that most people figure out when they are like 10. Programming an AI to do it is beyond easy.