Gameplay

>gameplay
>graphics
>story

What about AI?

Other urls found in this thread:

youtube.com/watch?v=Nt1XmiDwxhY
twitter.com/SFWRedditGifs

>AI in games past 2007

lel

t. Cowadoody kid

>he thinks old games utilized AI better
>clearly hes a ''Cowadoody kid''

how fucking retarded do you have to be to reach that conclusion?

i found an old magazine from 1996. they were talking about AI in alien trilogy and saying "in the nearest future, enemies will learn your moves and constantly adapt to your style of play!!"

Falls under gameplay. 2bh a lot of the time pretty basic AI is sufficient for a good game (conditional attack patterns etc), but I guess if you want to get sophisticated it can result in neat emergent behavior.

Much easier to design a cheating AI than a fair and good one.

AI can't progress further in games yet. We're at a plateau with it where you have modern AI in games, then movement reading, and there's no in between. It's too advanced for people to code.
>inb4 fear
literally voice clips and the same AI as any other game at the time.

>AI thread

F.E.A.R 1, obligatory.

Shodan is the best AI.

Nah, its because developers don't want to put forth the effort needed to actually code a decent AI. Its easier to crank the graphics up to 9000. Its what "sells". They don't even try any more.

Still haven't found a more satisfying AI to fight.

I only care about AI in stealth games. I have plenty of fun tearing it apart, and finding how bad it actually is.

Ayylmao in Isolation does that, to an extent.

See every X4 game ever.

Games stopped trying to do anything with A.I. past 2005.

I remember playing the first Far Cry and a bit after a firefight the few remnants of the squad I was in combat with someone yelled "He's got us overwhelmed! We have to fall back!" and then they proceeded to do so. And FEAR is a whole different story in terms of advanced A.I.

A.I. has never impressed me since those games.

Ayy in isolation is the exact opposite. He doesnt roam around the level. He spawns near you all the time. Lazy devs.

We had 500 threads about this.

Neural network based vidya IA WHEN?

This.

Shodan is cute!

That's weird, I played Far Cry a lot and I don't remember them specifically calling retreat. They do retreat sometimes, though it's often a questionable tactic, especially since their body armour doesn't work when you shoot them in the back. They do work in groups though, which is really cool. Also they have no idea how to throw grenades, but still do, which is slightly less cool.

FEAR AI is very exploitable and simple if you actually take your time killing them without slo-mo. Wouldn't say I dislike it though, they're pretty satisfying to kill, lots of responses.

>Neural network based vidya IA WHEN?

It wouldn't help too much.

We can't get AI to naturally make mistakes in the same way that an animal or human would

So far we can either have AI that is very easy to read, or AI that will be perfect at everything.

There is no inbetween

You can't make a dog neural network for a dog enemy for example

Anyone have the screencap of the programmer-user who got fired for making the game AI too good?

never

The AI would just decide it doesnt want to play with you

>AI
The damn game devs today can't even crawl, good AI is expecting them to get up and start sprinting.

Too bad FEAR is CS:de_office ad infinitum with some lazy jumpscares thrown in.
Could be bretty gud game otherwise, but lack of variety kills my interest in it very fast.

>de_office

Starcraft 2 is trying to push it, dunno what happened

Gamedev user here, most games don't need super advanced AI but it's usually a good thing to put as much effort in as possible so it feels fun to interact with them, even if they're not very smart. There are a hundred billion things to worry about when making a game, and time must be managed well. The more time spent on AI is less time somewhere else, and more often than not it's better spent somewhere else. Unless there's a big break through in AI development it will remain this way unfortunately, there are just too many things to worry about and if the AI functions well enough to be provide smooth and fun gameplay then it's doing its job.

>make a game where enemies begin with a randomised virtual genome which determines their behavioral tendencies
>each one killed returns its genetic material to the general pool proportionally to how well it performed during its lifespan (damage it caused to the player etc.)
>successive generations develop traits that make them harder to play against
>genetic crossover and a low mutation rate help maintain diversity and introduce unexpected behaviors
Would it work?

>spending time making good AI for a multyplayer game

FEAR 1's AI was only good because it had godlike pathing designed into the levels themselves.

That said, developers do need to put more effort into making the CPU fun to fight. Even if it's a hacky way to do it like FEAR.

Could be interesting depending on how the game is played. I've thought about games with adaptive enemies but more in a line of

>If you headshot a lot enemies start getting tougher heads/working without heads
>If you hide a lot they develop more sophisticated methods of tracking you

Combine these two ideas and you could have some pretty scary enemies.

Too easy to game the system, allow yourself to get attacked by a shit enemy a lot and its fitness would shoot up, making the next generation weaker.

>linear shooter
>intelligent AI
The AI in FEAR was heavily scripted.

Miasmata dedicated a lot of programming to the Creature's AI and I think it showed. It could track your scent on the ground and when it finds you, it paces around and waits for an opening, like a real predator.

google's alpha go is currently trained to play against some pro

last good AI I saw in a multyplayer game was UT2004

People don't want clever IA (or AI if that's what you meant) opponents because it's more fun playing with people but it'll have other uses, but I doubt it'd be used client-side rather than by devs (like procgen to make environments or animations maybe). At least not for a while.

youtube.com/watch?v=Nt1XmiDwxhY

I agree on the comments on spawning though. There's a mod to make it more unpredictable and not hovering around your general area all the time.

It's considered in gameplay.

>le genetic fitness meme
I don't think you can expect the computer to randomly outsmart a person reliably enough for that to work, I would also assume that a lot of the feedback would be noise because of people dying for stupid reasons rather than being outplayed

All you need to do is have genes that are appropriately linked with physical/stat traits. For example the gene that determines how easily an enemy spots you could be expressed on the model as having more exposed sensory organs that take critical damage. Then you allow enemies to earn 'fitness' points for auxiliary actions like spotting the player for another enemy, and you can end up with completely emergent symbiotic relations between enemies appearing like more vulnerable 'scouts' spotting for big bruisers with armored heads that can't see shit on their own.

The problem is that as soon as you make an AI that actually outsmarts a player, every single one of your playtesters will flip their shit and say the game is broken and they can't win.

AI has to be massively gimped to make the game fair.

This, the AI in Isolation is laughably bad. It's easily spoofed in most sections. The devs made up for it's short comings by making it spawn close to the player.

FEAR is highly scripted. Basically put in ever room there are designated spots and actions enemies supposed to do depending on 3-4 scenarios player choses. if you make a level without those, enemy becomes as stupid as in every other shooter.

>I don't think you can expect the computer to randomly outsmart a person
Right, you'd obviously need some traditionally-written baseline behavior so that you're not throwing spazzy vegetables at the player. No reason you can't add a load of parameters to that general behaviour that are determined by genetic operators though. Also yeah, assigning credit to individuals would be critical but not impossible.

You expect game devs to spend more than half of development time to research/perfect their AI when that time can be used to polish the other aspects of the game?

Space Rangers 1/2 - NPC's posess all the abilities and can perform all the actions player does. They can even complete quests and government assignments. For instance you can wait until someone takes a mission that you refused to do, then track them and watch them do it. Governments can also assign same task to several rangers beside you, so for instance if you got ordered to catch a pirate, and if you wait long enough you will see other ranger take him down, get back to quest planet and get rewarded.

If you play on easy-normal difficulty you can get to a point where NPC rangers beat the whole game for you - takes 20-30 in-game years. In 2 though they cant beat one of the 3 final bosses, as that requires ground battle, which is the only thing AI can not do without a player.

By estimate, to make it at least somewhat better than specifically programmed script-based AI, you'll need to run the evolution algorithms for a few years on a supercomputer, and resulting neurological network would occupy about 500GB, which immediately renders it unuseable for VG purposes

>They can even complete quests and government assignments.
They can't.
>you can wait until someone takes a mission that you refused to do
You can't.
>Governments can also assign same task to several rangers beside you
Never happens.
>if you wait long enough you will see other ranger take him down
Can happen because nobody likes pirates.
>get back to quest planet and get rewarded
Nope, but if you take a mission to kill somebody and that ship dies, for instance, in a dominator/klisan invasion, you can still get paid.
>as that requires ground battle
You can actually defeat Terron just by shooting him (it?) until he (it?) dies, it just takes a lot of time. Alternatively, send him to terraform a star.

see
I'm not suggesting the AI start entirely from scratch with the eventual goal of becoming unbeatable - not only would that be unfeasible but it wouldn't even be fun to play. What I'm suggesting is much less ambitious and doesn't involve NNs, but could still provide the player with organic surprises and challenges.

games are targeted at dummies and dummies dont need AI to be challenged

There is literally no decent single players that utilize AI properly now because how lazy devs are at only focusing on multiplayer.

BotW has great AI, it's the only game that comes to mind that put effort into its AI.