Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[feelotraveller]

The purpose of game AI is to be fun to play against.

Yep, that's the line people like Soren Johnson have been running with for decades, and I wouldn't disagree at a macro level.  But there are different kinds of fun - someone like wierd wants a challenge and presumably that involves getting feedback which changes as the 'AI' learns, whereas a bunch of spotty teenages just want something to repeatedly wipe their boots on that lets them get consistent 'success' by repeating the same thing over and over again.  Guess who the market (generally) caters for?

It's machine learning, which is sort of an intelligent behaviour. The other thing is that neural networks are, at least in origin, based on mimicking the human brain - and what else do you call an artificial brain? Also people like calling it AI. I don't know anyone in the field who actually calls it anything other than machine learning, though.

Again I largely agree - although with a large dose of reserve, what chatbot (even) is actually credible today? - the ability to learn is generally thought to be a prerequisite for (or mark of) intelligence.  My point was that just about no game AI "even aspires to such lofty heights" but rather generally involves a set of unchangeable predefined triggers without any capability for learning.

Now AI toothbrushes on the other hand... 

[Naturegirl1999]

I wonder what coding language would be good at making neural networks

[Arx]

Yeah, pretty much. Specifically the TensorFlow library, but PyTorch is rapidly catching up/overtaking it. Last I heard from my research group before I left was that the next step would probably be code in PyTorch, deploy to TensorFlow.

Reelya-style edit:

The purpose of game AI is to be fun to play against.

Yep, that's the line people like Soren Johnson have been running with for decades, and I wouldn't disagree at a macro level.  But there are different kinds of fun - someone like wierd wants a challenge and presumably that involves getting feedback which changes as the 'AI' learns, whereas a bunch of spotty teenages just want something to repeatedly wipe their boots on that lets them get consistent 'success' by repeating the same thing over and over again.  Guess who the market (generally) caters for?

I agree with you, but I don't think it negates my point - I have no problem with an AI cheating to bring itself up to be able to compete with a player, and I don't think it casts the skill component of the game into question at all.

As for compstomps vs. challenge, that's what difficulty settings should be for, provided you're in a context where those are relevant. I just don't think it inherently torpedoes the game's credibility if increasing the difficulty gives the AI, say, a resource handicap, as long as it's executed in a way that doesn't make playing against the AI blatantly unfun or unlike playing against a human.

[Max™]

import strong_ai

[Egan_BW]

I agree with you, but I don't think it negates my point - I have no problem with an AI cheating to bring itself up to be able to compete with a player, and I don't think it casts the skill component of the game into question at all.

As for compstomps vs. challenge, that's what difficulty settings should be for, provided you're in a context where those are relevant. I just don't think it inherently torpedoes the game's credibility if increasing the difficulty gives the AI, say, a resource handicap, as long as it's executed in a way that doesn't make playing against the AI blatantly unfun or unlike playing against a human.

It annoys my inner roleplayer when AI players work differently for the sole reason of not being the player. Strategy games are more interesting to me as simulations than as challenges. And if the player and their opponant are the same except the opposition gets an arbitrary boost, it feels as if the simulation's integrity is compromised.
Then, I say this as someone who has never really gotten into strategy games.

[wierd]

You guys are missing my real point--

The AI for a strategy game, should permit the player to use--- Strategy.

Strategy is more than just "Oh, I will abuse this set of mechanics to zerg rush and shit."  It is "Surprise fucker, i'm not where you thought I was" too.

Omniscient AI completely torpedoes this latter kind of play.  AI that spontaneously generates units just outside the FoW likewise is bogus.  Similar story for AI that does not obey resource rules or cooldowns, because cutting supply access is a valid strategy for defeating an otherwise superior enemy through attrition.


See where i am getting here?  When your ai cheats like a mofo, you lose the right to call it at STRATEGY game. 

[Arx]

It annoys my inner roleplayer when AI players work differently for the sole reason of not being the player. Strategy games are more interesting to me as simulations than as challenges. And if the player and their opponant are the same except the opposition gets an arbitrary boost, it feels as if the simulation's integrity is compromised.
Then, I say this as someone who has never really gotten into strategy games.

I'm a hardcore competitive RTS player (am I good? Questionable), and have friends who are the opposite - completely casual. In the former case, either the AI has to be good enough to compete with you (hella difficult for the average game studio to pull off without cheats), or single player is only good for practicing build orders (basically true of all current competitive RTS). In the latter case, the game is more about having some fun without having to go full tryhard, and the computer playing like a human isn't that important 'cause it's just about having fun.

The real issue is that the player and their opponent are fundamentally not the same. Sadly, the game is rigged from the start, and we can't have nice things.

You guys are missing my real point--

The AI for a strategy game, should permit the player to use--- Strategy.

Strategy is more than just "Oh, I will abuse this set of mechanics to zerg rush and shit."  It is "Surprise fucker, i'm not where you thought I was" too.

Omniscient AI completely torpedoes this latter kind of play.  AI that spontaneously generates units just outside the FoW likewise is bogus.  Similar story for AI that does not obey resource rules or cooldowns, because cutting supply access is a valid strategy for defeating an otherwise superior enemy through attrition.


See where i am getting here?  When your ai cheats like a mofo, you lose the right to call it at STRATEGY game. 

"If the AI is blatantly unrealistic, if feels blatantly unrealistic to play against."

Yep, I never contested that. It's pretty obvious. That's why my comment was the AI must be fun, and cheating does not preclude fun. It can ruin it, but its mere presence isn't enough to do so.

[wierd]

I liken a cheating RTS AI, to a chess AI that moves pawns like a queen (without reaching the far side of the board first), because "the player is too good!"

Whatever game that is, it is NOT chess, and that is NOT a valid strategy.

[Arx]

You're entitled to that opinion and I cannot dissuade you from it, but unfortunately the reality is that every shipped RTS AI will disappoint you, then. AlphaStar is the only AI I can think of that can actually play an RTS properly, because the problem is simply too complex for dev studios to solve within budget.

[Max™]

Do we have strong ai yet?

Sorry guys, I'm not very good at python I guess.

[Reelya]

You're entitled to that opinion and I cannot dissuade you from it, but unfortunately the reality is that every shipped RTS AI will disappoint you, then. AlphaStar is the only AI I can think of that can actually play an RTS properly, because the problem is simply too complex for dev studios to solve within budget.

Pretty much this. All those broken promises from studios weren't just them hyping up non-existent tech. It was hubris from super smart programmers going "good AI, how hard can it be?" and thinking they're going to brew up some batch of "magic sauce" within budget that's going to totally play this game without needing kludges. Anyone saying "just put good AI in, how hard could it be to balance that, just vary the level of resources the AI has" is offering you Peter Molyneux level bullshit.

[wierd]

Which is why I said this:

I would be happy with a simple algorithm if it simply had properly restricted inputs that better represent what is available to a player, and not an omniscient computer-- and obeys actual game rules. (Resource costs, build times, and cooldowns.)


The actual HOW it decides where to move, as long as it is not "Move mass of units to player base location" in nature-- and instead, "Triangulate base location from vectors of enemy unit travel as they enter my visual space"-- is moot.  Use whatever algorithm you want. It need not be NN or a GAN, or anything fancy like that.

Just don't fucking cheat, and dont pretend that the game is even remotely about skill if you do.

EG, I am just fine with something that operates on branch tables, and is in no way a modern flavor AI.  It just needs to follow the game's rules, and I will be OK with it.


The problem, is that for varying levels of "Buuuuuuuuuuuuut!"  from developers, they RELY on the computer being omniscient to do ordinary things, like find paths.  Make it obey FoW, and suddenly the computer can't even get out of its own base.

Rather than fix this, they just go "Oh, the player will never notice!"

Bullshit.  I notice.


Like I said, the genre is STRATEGY.
When simple strategies, like base relocation, DO NOT WORK AS EXPECTED, and where decoy operations are likewise ineffective, 100% of the time, the laziness of the developers to properly limit their pathing routines to obey FoW results in entire classes of strategic combat being impossible, because "buuuuuuuuuut!" and it becomes painfully fucking obvious.

"Make it cheat!!" is poison fruit.  Like I said, once you decide that the rules don't apply anymore for your computer player, you aren't playing the stated game anymore. 

[Trekkin]

Rather than fix this, they just go "Oh, the player will never notice!"

Bullshit.  I notice.

They know perfectly well you'll notice. They have just (correctly) determined that there are more profitable ends to which to devote their development resources. Once the AI can satisfy casual players by setting up units for them to knock down, it's good enough for one large fraction of their player base, and the other large fraction ultimately wants to play against humans anyway. Sure, they could chase some vaporware ideal of "good AI," but they could also spend that programmer time making the engine run more efficiently or across a wider range of hardware and make the game more widely accessible -- and what is more, even if they don't have time to make it perfect, every incremental improvement in efficiency is ultimately helpful. A wonderful AI that will be ready a day after the deadline is not. The former is therefore a safer bet.

Sure, they could completely restructure their development process to make that not true, and if you think they "should" do that, you know how to prove it. In the absence of that proof, though, the existing process does make games that sell infinitely better than vaporware.

[Naturegirl1999]

A video about the various hypothesis about how the universe began and the assumptions we make

[Il Palazzo]

I kinda like Sabine's outreach efforts, especially her 'talk to a physicist' service - which, to me, seems like one of the better ways to deal with wannabe Einsteins peddling their pet theories. Her blog posts and videos are also generally very measured and informative.
Not sure such an enumeration of hypotheses as in that video is actually of any use to anyone who isn't already in the know, though.