Print

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

[kj1225]

The thing is, all strategy games cheat, because they have to. CPUs just aren't good enough to fight people.

It's just they're usually less in your face with their special favors.

Oh, I know. I've got plenty of hours with shogun 2: total war. Those shitty tiny 1 province clans always manage to pull full muderstacks out of their asses. But at least they don't get units with abilities that I can't get and they don't magically get extra turns or something, so I can deal.

Well, technically it's kind of different. Depending on the type of game, the AI could destroy you with ease. Like, imagine if the AI in something like Civ was doing tactics that all the best players did.

[AbstractTraitorHero]

I'd rather have me be absolutely wrecked by tatics and stradegy then fight a cheating opponent.

[Empiricist]

I think the point is that they can't use proper tactics and strategy, hence why they get all those privileges. I mean, videogame "AI" often isn't really AI. I mean, it's not like anyone would want to make and train a neural network or something for that (assuming they even have the resources in the first place), especially not with the risk of them coming to very strange conclusions, when they could just take the easier method which costs less and takes less time.

The issue being brought up isn't that they have those privileges, more that they're too in your face about it.

[AbstractTraitorHero]

I can't wait till ai is cheap and easy as hell maybe in eh 5 or 6 years?

[Empiricist]

Don't hold your breath. Even if it becomes cheaper and easier, which I don't think that much progress will be made on in only five years, it still needs training which is a bit of an issue seeing as at the time of training, no metagame would even exist (it may even be a chicken and egg issue, you want player strategies to train the AI with so you can give the game AI, but the players can't form any strategy without an AI to use them against). I mean, what if you find a balance issue and need to make changes? Well, have fun retraining it when you could've just done it the old fashioned way.

I mean just coding it instead of relying on machine learning works, for the most part at least. So it's arguably best to keep it that way and save the machine learning for problems where just coding it manually isn't feasible.

[AbstractTraitorHero]

Eh maybe Ai is better for npc characters in roleplaying games then?

[Nunzillor]

What about a search algorithm instead of a neural network?  Good heuristic, don't search to a great depth??

[AbstractTraitorHero]

Yeah if it's a balance issue you can just fix the issue in the game and the ai will adapt!

[Empiricist]

What about a search algorithm instead of a neural network?  Good heuristic, don't search to a great depth??

Isn't that what some videogame "AI" use already, such as for chess? I mean, yeah, I'd imagine it'd work quite well, but I don't think it'd be actual AI by itself though it is certainly used in actual AI. Not that actual AI is really needed for most videogames.

Unless I'm just going crazy and making a fool of myself ranting about my own delusions.

[AbstractTraitorHero]

Yeah but actual ai would be awesome vs matches agasint them could be truly entertaining.

[Nunzillor]

What about a search algorithm instead of a neural network?  Good heuristic, don't search to a great depth??

Isn't that what some videogame "AI" use already, such as for chess? I mean, yeah, I'd imagine it'd work quite well, but I don't think it'd be actual AI by itself though it is certainly used in actual AI. Not that actual AI is really needed for most videogames.

Unless I'm just going crazy and making a fool of myself ranting about my own delusions.

No worries, I'm sure I'm the one ranting about my own delusions, or saying things that make little sense.  Oh well.

Game theory applied (through search algorithms) is part of AI, but I think you mean that it isn't machine learning?  I just feel like the old approaches can be improved with better heuristics to judge states (they use more development time for that) and more computational power... which is getting better constantly...

Training and implementing a neural network for a video game seems like a lot, but maybe I'm wrong?  If you could actually design, train, and implement it within a dev's limited timeframe, that would indeed be awesome though.

[Empiricist]

No I mean, I don't think search algorithms by themselves count as AI, though they can definitely be used in it. But yeah, I definitely do agree that the old approaches work quite fine and that they can definitely shine well if given new stuff, just that it may not be economically feasible to implement a full learning AI. At least, not for strategy games, I know that bullet hell(?) called Warning Forever implemented some means for the bosses to adapt to the player, though I don't think such a trick will work as well for single-player strategy games unless you manage to justify why the first few enemies are utterly brain dead. Then again, my definition of AI might just be too restrictive since I've been assuming it includes the capacity to learn or at least be taught.

[Nunzillor]

I just always thought it was AI, dunno.  Performing intellectual tasks artifically approximating humans in ability is what I think of, but that may be short-sighted?  It's taught in (beginner) uni AI courses, but maybe it doesn't count because it's not "intelligent," I guess.  Perhaps it all depends on your perspective?  That's just what I think.

Like, IBM's research that produced Deep Blue was considered a breakthrough in AI research, I think?  That was basically just applying a search algorithm with lots of computational power, right?

[Empiricist]

Yeah, I suppose my uni's distorting my perspective as well, seeing as it just offers a Machine Learning course rather than an AI course which is probably affecting my perception of it. Well, that and the main resource I currently have in reading up about it is focused on the learning aspects (since for whatever reason, though I can take Machine Learning now in second year, one of it's prerequisites is a second year, semester two subject, meaning I still can't take it until next year >__<).

[syvarris]

On the topic of XCOM 2's AI difficulty, I had actually thought it looked rather decent, though I've only watched the first two hours of the first video.  One of the things that I liked about Long War was that one of the primary sources of increased difficulty was the higher number of enemies.  Especially EXALT missions, which had absurdly massive hordes of weak enemies.  It was difficult because they had such numbers, but it didn't feel like cheating, and it made me feel like a badass mowing down so many enemies.

XCOM 2 seemed to be doing something similar, by having the basic Advent troops be weak and numerous, without special effects.  Then again, maybe the change from having lots of different types of basic enemies, each with their own quirks, makes the game less interesting and they cheat more to compensate.

...Mind control is certainly cheap, though.  I mean, seriously, how are you supposed to counter that?  Just run away?  Boring.  Only use soldiers with high will/will-increasing items?  Annoying, not strategic.

I think the point is that they can't use proper tactics and strategy, hence why they get all those privileges. I mean, videogame "AI" often isn't really AI. I mean, it's not like anyone would want to make and train a neural network or something for that (assuming they even have the resources in the first place), especially not with the risk of them coming to very strange conclusions, when they could just take the easier method which costs less and takes less time.

The issue being brought up isn't that they have those privileges, more that they're too in your face about it.

You underestimate nerds, Emp.  Just a little while ago, a single guy made an AI for Dark Souls PvP, using a neural net even.  Yes, Dark Souls is much easier than XCOM, but there's certainly people who are interested in making AIs, even at significant cost with little return on their investment.

I honestly question whether XCOM would be a better game with an extremely skilled AI.  I mean, who do you want to be?  The guy who has inferior capabilities, but might succeed because he's smarter?  Or the comparatively stupid person who will generally only win by dint of having superior firepower?  If a match is challenging, both people will almost certainly exist.  A good AI would need to be an AI which dynamically adjusts how stupid it is based on how smart the player is, so that the player can always feel like he's the smart guy who triumphed over the idiot.


I think the Dark Souls PvP AI is a pretty decent example of that.  Yes, it's vastly superior to the game's normal NPC AI, but it's also vastly harder.  It's not skilled enough to beat the best human players, but it utterly stomps most people, and most likely could do so even if it had significantly inferior stats.