Game development and why most games have "bad" AI

[unmerged(63310)]

Just was watching Sid Meier's presentation at GDC http://gdc.gamespot.com/story/6253256/meier-on-crafting-the-epic-journey-full-keynote-video-inside and while I think he is out of touch on a few topics most of what he says makes sense. You can see some of the examples with HoI3 and it seems Paradox did learn alot of things from that experience that are already being affected in Vicky 2.

Personally I think a greater number of difficulty levels would help make games have more of a learning curve with also rewarding the learning by giving more challenges but since most developers decide the budget doesn't need to waste money on the few hard core players that will learn all the details of the game mechanics and give 3 difficulty levels if any very few games provide real challenge once the game is learned. The vast majority of people play a game once or twice then move on to the next game and will not put up with a game that is difficult from the start.

Of course most Paradox games are having a much steeper learning curve than an the average game and I think that tends to make people who put the investment of time to learn everything really cry out more for some challenge once they mastered most of the mechanics so Paradox has an interesting challenge there.

I have definitely noticed though that people complain alot less about getting slaughtered in multiplayer against other humans(not that they don't complain, just less) but if the AI somehow beats them in a way which is not clear- the common assumption is the AI "cheated" and people can get very angry about that when they convince themselves they couldn't have lost any other way.

 

[Delegate]

Exactly. Like when the US has entire population of California, Ohio, and other states turned into soldiers and is sending hundreds of divisions at you... I always ended up screaming and banging the keyboard. What is so hard for developers to make a system where these insane things dont happen, so that the AI plays a legit game...? I mean, they can give the AI ways of making it challenging without letting it do stupid sh*t like that.

Look at Age of Empires 2, the AI plays a completely legit game with no cheating or boosts. It's just really efficient. Not that that is really applicable here, but that is the best example I can think of.

 

[Garek Maxwell]

The problem is that the more decisions you grant to the AI the more easily it will become "dumb" because...well...in spite of the name the Artificial Intelligence doesn't 'think'. It all boils down to a bunch of switches and when the right series of switches are turned on then it goes haywire and turns everyone into a soldier for its own little world conquest.

I don't really expect perfect AI from release, but the great thing about patches is that they can fix the AI later. This means on release Paradox can have a working (but "stupid") AI that cheats at times and later patch in something that would have taken too long and cost too much to impliment in time for release. I personally don't mind this. It gives me time to get used to the game mechanics.

 

[Chaingun]

Citing the relevant parts from that link:

[11:11] Now, Meier is going to discuss the "Role of AI."

[11:11] He thinks that AI needs to be part of the overall experience but should not be considered a person.

[11:12] He warns that making the AI do surprising things will have bad effects either way. If it does something bad, the player will assume it's stupid. If it does something overly clever, the player will assume it cheated.

[11:13] Meier likes to think of the AI as a metric. It will make the players better and better and will give players feedback. In the case of Civ Rev, the leaders give feedback, since they're the "only friends players have in a game."

As for why games have bad AI, it's actually not just that players want to win, but even more that coding an AI that appears intelligent is incredibly difficult (except for simple games, e.g. chess - and no I don't want to re-discuss the meaning of thw word 'complexity' or discuss chess programming in this thread).

It should also be noted that dumbing down a fundamentally good AI is much easier than making a poor AI better.

I totally agree more difficulty levels = good. Reason being a beginner will not like getting beaten to dust the first time he plays a game, nor does a veteran always win at the hardest difficulty level. Try losing as France on Very Hard as a veteran in EU2/3.

 

[Temrek]

The problem with making a good AI for a complicated game is not to make it good, it is to make it good AND fast. Complicated games have incredible many factors that need to be considered, many may have the be calculated using complicated algorithms before they rare of any use to the AI and resources are limited.

And even then, an AI who can play the game as a genius unless the player uses a very draconian strategy in the game will be considered stupid by a lot of those that play the game.

 

[Jerzul]

Exactly. Like when the US has entire population of California, Ohio, and other states turned into soldiers and is sending hundreds of divisions at you... I always ended up screaming and banging the keyboard. What is so hard for developers to make a system where these insane things dont happen, so that the AI plays a legit game...? I mean, they can give the AI ways of making it challenging without letting it do stupid sh*t like that.

Look at Age of Empires 2, the AI plays a completely legit game with no cheating or boosts. It's just really efficient. Not that that is really applicable here, but that is the best example I can think of.

Well you don't need to worry about this. One of the DD's said that POPs can't be switched manually. They have to be created organically, so now the only way for the AI to create a huge amount of soldier pops would be to pay them very high wages (probably while simultaneously hindering other lower/middle class jobs).

 

[Chaingun]

The problem with making a good AI for a complicated game is not to make it good, it is to make it good AND fast.

I'd say both are problems. Given infinite computing power an AI that fallows the rules of the game will still be finitely good... You'd need to be able to state an optimal solution to the problem of winning the game to exploit infinite computing power. Easy in chess, but try doing it in e.g. Diplomacy ;-).

From practical experience I'll say that making the AI for a non-standard TBS game not being retarded is difficult before you even start optimizing it, though admittedly that's partially due to fact you can't probably shouldn't use algorithms with O(n^2) and worse time complexity. Implementing AIs for checkers and chess were piece of cake in comparison.

 

[unmerged(75409)]

In the end the AI is just an automaton which reacts to inputs by the player. Just a software function. Not a replacement for human players.

PC games are not simulations of reality, they are entertainment software. ENTERTAINMENT software. Not works of science, not statements on history, not statements on civilization or anything. Just entertainment software, like Microsoft word or your favorite language learning program. A lot of gamers seem to expect games to be more than entertainment :wacko:

 

[Chaingun]

A part of the entertainment in strategy games is having an adversary that isn't brain dead.

 

[unmerged(75409)]

Yeah of course, but why do so many people get upset when the AI "cheats"? Do they feel deprived of what would otherwise be a proud achievement - beating a software function in an entertainment program?? It's still a machine!

Gamers who get outraged over "cheating" AI resemble a steed who complains that the wooden mare does not move properly... sure it doesn't but even if it did it would still only be a wooden mare

 

[twinxor]

Making optimal AI for a ridiculously complex game like Victoria is out of the question, and it wouldn't be fun anyway in a historical game (real-world leaders made irrational choices all the time). The AI will use heuristics to attempt to approximate reasonable behavior, and if you don't want it to be cripplingly incompetent you'll have to live with it getting some "unfair" advantages to make it a worthwhile opponent. Play a simpler game if you can't stand that, I guess.

 

[Andrelvis]

In the end the AI is just an automaton which reacts to inputs by the player. Just a software function. Not a replacement for human players.

PC games are not simulations of reality, they are entertainment software. ENTERTAINMENT software. Not works of science, not statements on history, not statements on civilization or anything. Just entertainment software, like Microsoft word or your favorite language learning program. A lot of gamers seem to expect games to be more than entertainment :wacko:

Or rather, are entertained by all sorts of different things.

 

[unmerged(139928)]

In the end the AI is just an automaton which reacts to inputs by the player. Just a software function. Not a replacement for human players.

PC games are not simulations of reality, they are entertainment software. ENTERTAINMENT software. Not works of science, not statements on history, not statements on civilization or anything. Just entertainment software, like Microsoft word or your favorite language learning program. A lot of gamers seem to expect games to be more than entertainment :wacko:

And a lot of gamers want some challenging opponnents while playing single-player, to keep them entertained. Others are also entertained better if the game is closer to a simulation of reality.

Yeah of course, but why do so many people get upset when the AI "cheats"? Do they feel deprived of what would otherwise be a proud achievement - beating a software function in an entertainment program?? It's still a machine!

Yeah of course, but why do so many people play "video games"? Do they feel proud about beating software functions and pre-determined challenges in an entertainment program by making various inputs at different times while going through a "story" of sorts?? It's still not real!

 

[unmerged(75409)]

Yeah of course, but why do so many people play "video games"? Do they feel proud about beating software functions and pre-determined challenges in an entertainment program by making various inputs at different times while going through a "story" of sorts?? It's still not real!

Well I admit I do feel a sense of achievement when I defeat the AI, playing as an underdog nation. I like immersion in a history like game even if it really is only a computer game and I would be better off using sunday afternoons to prepare for work. o It's just so much fun!

I just can't stand the whining you sometimes get in the HoI3 forums when they complain that the AI cheats by not having as severe attrition, and all sorts of bonuses to remain challenging.

 

[Garek Maxwell]

I just can't stand the whining you sometimes get in the HoI3 forums when they complain that the AI cheats by not having as severe attrition, and all sorts of bonuses to remain challenging.

I find that really funny since in just about every single game with an AI of any kind has advantages that could be considered cheats. In a game with some sort of "fog of war" (like a Real Time Strategy war game) the AI has the advantage of knowing what the player doesn't. Other things could be reacting to information not accessible to the player like internal information on a country. For something like Victoria, the AI might know what's going on in another country that could only be known if you were playing that country. (Or guessed from certain kinds of evidence.)

The only way to stop these "cheats" is to actually cripple the AI and this can often lead to unpredictable problems. Things like the AI assuming it's neighbors are warlike and will build up giant armies and go bankrupt over it decade after decade because it doesn't know the threat level of its neighbors is one example. Sometimes AI specific cheats are necessary, even if someone else doesn't like them.

Plus, Germany totally cheated in Victorian times. The German states weren't supposed to unite like that. Big grey blob all over the map? That's total hax!

 

[Kodos666]

what I don't understand is: computer science has showed many examples of AI-design that weren't considered in game-development. What happened to a 'grown' AI, genetic algorithms and neural networks, why does anybody stays with the old 'IF...,THEN...'-approach?

 

[Chaingun]

what I don't understand is: computer science has showed many examples of AI-design that weren't considered in game-development. What happened to a 'grown' AI, genetic algorithms and neural networks, why does anybody stays with the old 'IF...,THEN...'-approach?

Genetic algorithms and neural networks have very little application to game AI. Neutral networks are used mainly for pattern recognition, whereas a genetic algorithm essentially is a glorified random search...

Also, define 'old 'IF...,THEN...'-approach' more closely... If you mean finite state machines, yes those are still used. In fact, they will likely continue to be used. It's how you transition between states (e.g. when to go from EAT(object) to ATTACK(enemy) ) that's quite interesting.

 

[unmerged(135578)]

what I don't understand is: computer science has showed many examples of AI-design that weren't considered in game-development. What happened to a 'grown' AI, genetic algorithms and neural networks, why does anybody stays with the old 'IF...,THEN...'-approach?

The fact is that neural networking does not really work with games. Genetic algorithms (at least when I studied them long time ago) had no real application either. As far as IF-THEN is concerned, nowadays you do use finite state machines to set Game AI to different states (ATTACK, INVASION, DEFEND etc) which pretty much how things were done long ago.

"Advanced" stuff that does work is GOAP (at least with simple soldier actions, the E:TW that used it turned to be less than functional). Effectively having a hierarchical task network (HTN) with a bag of goals that can be arranged according to need (GOAP). How well it can really work in strategy game is still very much open question...

 

[knul]

what I don't understand is: computer science has showed many examples of AI-design that weren't considered in game-development. What happened to a 'grown' AI, genetic algorithms and neural networks, why does anybody stays with the old 'IF...,THEN...'-approach?

Rule-based programming (rules being if-then constructions) is broadly applicable. You can write about anything in it. Old as it is, it is powerfull and still very usefull. In fac, most computer models within cognitive science are built with rule-based like architectures (f.e. Soar, ACT-R).

Genetic algorithms and especially neural networks are IMO overvalued by many people. Neural networks require long training times and even then become black boxes, making it extremely hard to fix them if they do not behave exactly as you wish.

The problem with genetic algorithms is that they are "weak" algorithms, meaning that they are knowledge-poor and domain-independend. The experience within the AI field is that weak algorithms do not work very well. In order to behave intelligently, you need domain knowledge and lots of it too. General reasoning engines and such have performed much weaker than domain knowledge rich expert systems. In other words, there is no "panacea", no universal solution. Unless you incorporate lots of knowledge in your application of a genetic algorithm, it will not produce very good results.

In most cases the bottleneck in designing good AI is to obtain the vast amount of domain knowledge required and find out how to formalize it so that the computer can use it. For games, this requires an extensive research into the knowledge that a human gamer uses to play effectively and a way to program this into the computer. This is hard work and it seems that it is unavoidable, whether you use rules or ANNs or fuzzy logic or planners or whatever floats your boat.

 

[podcat]

neural networks are also fairly unpredictable and very hard to train well with back propagation etc when the rewards dont show until very very far in the future. They can be a fun academic toy and I know of one or two projects who use them for very specific problems, but they have no place in most development.