Playing complex strategy games for many years, one of the things that irks me the most is that hard AI levels often just give the dumb AI cheats to simulate it being smarter. To me, it’s not very satisfying to go against cheating AI. Are any games today leveraging neural networks to supplant or augment hand-written decision tree based AI? Are any under development? I know AI can be resource intensive, but it seems that at least turn based games could employ it.
trying to live train AI against your playstyle is both expensive and unnecessary. Hard bots have never really been too much trouble. We don’t really need to use AI to outpace humans in most games. The exceptions would be an extremely long play games like chess and go.
There’s been a lot of use in AI for platformers and stuff like trackmania, but not for competition, simply for speedruns.
yeah I would like to leverage AI for stuff like RPG NPCs. instead of hearing the same filler lines for 200 hours of gameplay, barely reacting to the context of your game you could have a vibrant array of endless dialog that actually keeps up with your game progress (or lack thereof).
That would be a pretty good use. Llms are a little slow on most home hardware still. Hallucinations could also be a little scary. I wonder if that would affect your ESRB rating, That’s technically it could say anything…
The fear of hallucinations is so great for a commercial company that when square enix tried it on a remake of a detective game of theirs, it became the poster child of how awful LLMs are for videogames, it’s one of the worst rated steam games, it’s like talking to a wall because they nerfed it so hard it’s worse than a normal text parser.
It would certainly be nice to have for the fighting games I play. A few have toyed with the idea of “shadow fighters”, but it never really feels like playing against a person. It might get their habits down, but it doesn’t replicate the adaptation of facing a person and having them change how they play based on how you’re playing. If someone could crack that nut, everyone would have someone on their level to play against at any hour of the day, no matter how obscure the game is.
Hard bots have actually been so much trouble, that literally the only way to make them hard at all is to make them cheat by allowing them to operate outside of the ruleset the player is bound by. It’s a humongous issue with every strategy game on the market.
The most advanced AI I’ve seen is in Hitman WoA, and Zelda: Breath of the Wild.
Both games don’t have “learning” AI. They just have tons of rules that the player can reasonably expect and interact with, that make them seem lifelike. If a guard sees you throw a coin twice in Hitman, he doesn’t get suspicious and investigate - he goes and picks it up just like the first one. Same for reactions to finding guns, briefcases, or your exploding rubber duck.
ECHO, the 3rd person action\puzzle game was a fun concept to script in your machine dopplegangers to learn on you (and repeat after you one of the set actions you can do) and reset every cycle.
I don’t think it would work by itself without such limiting.
I always got the impression it wasn’t a learning AI but rather a very limited “Has the player pressed the run button? if YES: AI can use run next cycle”
Yes it is, it’s 100% scripted. And yes, in the environment where you can do like 10 different actions, they start to do their routine adding ones that you used in that cycle before they get reset. In a sense, they act no more natural than monsters from a tabletop game.
But these do make me think that if we talk gamedesign with a LLM as an actor, it should too have a very tight set of options around it to effectively learn. The ideal situation is something simplistic, like Google’s dino jumper where the target is getting as far as it can by recognising a barrier and jumping at the right time.
But when things get not that trivial, like when in CS 1.6 we have a choice to plant a bomb or kill all CTs, it needs a lot of learning to decide what of these two options is statistically right at any moment. And it needs to do this while having a choice of guns, a neverending branching tree of routes to take, tactics to use, and how to coexist with it’s teammates. And with growing complexity it’s hard to make sure that it’s guided right.
Imagine you have thousands of parameters from it playing one year straight to lose and to win. And you need to add weight to parameters that do affect it’s chance to win while it keeps learning. It’s more of a task than writing a believable bot, that is already dificult.
And the way ECHO fakes it… makes it less of a headache. Because if you limit possible options to the point close to Google’s dino, you can establish a firm grasp on teaching the LLM how to behave in a bunch of pre-defined situations.
And if you won’t, it’s probably easier to ‘fake it’ like ECHO or F.E.A.R. does giving a player an impression of AI when it’s just a complicated scri orchestrating the spectacle.
The only issue with current systems is that the “AI” is tweaked to the specific game mechanics. You can easily enough build multiple algorithms for varying play styles and then have it adapt to counter the play style of the player. The problems is that the current way that many games are monetized is through expansions, gameplay tweaks, etc., as well as those being necessary when a game mechanic turns out to be really poorly implemented or just unpopular and the mechanics change. If the “AI” isn’t modified at the same time to rake advantage of the changes, then it becomes easy to beat. The other issue is that eventually a human can learn all of the play style algorithms and learn to counter them and then it becomes boring.
Unfortunately, generative “AI” is not a true learning model and thus not truly intelligent in any sense of the word. It requires that it is only “taught” with good information. So if it gets any data that includes even slight mistakes, it can end up making lots of those mistakes repeatedly. And if those mistakes aren’t corrected by a human, it doesn’t understand which things were mistakes and how they contributed to winning or losing. It can’t learn that they were mistakes or to not do them. It doesn’t truly understand how to decide something is wrong on its own, only that things are related and how often it should use those relationships over others. Which means manual training is required, which due to the sheer volume of information required to train a generative “AI”, is not possible in a complex game where the player has thousand of possible moves that each branch to thousands of possible combinations of moves, etc.
Have you read about Alphastar, super computer that can whoop the best StarCraft players in the world? https://youtu.be/ZsCnuDgDcPo?feature=shared
