Why hasn't AI improved in the same way that e.g., graphics has?

[FOARP]

Interesting piece on AI in computer games featuring one of the lead CKII developers:

http://www.ign.com/articles/2013/04/08/whatever-happened-to-video-game-ai

My basic take away from this is that the reason why the improvement in AI over the last 15-20 years has been seemingly less than that in other areas of computer games is that there is eesentially a law of diminishing returns in coding AI - to give computer AI instructions making it capable of behving intelligently in 99% of circumstances takes far more coding that getting it to behave properly in 90% of circumstances.

Did I read it right?

 

[Pellaken]

Money.

More people will buy games with good graphics than they will games with good AI. Look at this http://forum.paradoxplaza.com/forum/showthread.php?665425-New-Sprite-packs-and-music/page6&p=15208973#post15208973 it's from a paradox game and it's a completely dismissive post when I suggested ANYTHING BUT new graphics. This is the kind of mindset many people have these days, and the poster is right - the vast majority share this mindset. This is frankly the answer. I don't blame the devs, I blame the players.

 

[FOARP]

Money.

More people will buy games with good graphics than they will games with good AI. Look at this http://forum.paradoxplaza.com/forum/showthread.php?665425-New-Sprite-packs-and-music/page6&p=15208973#post15208973 it's from a paradox game and it's a completely dismissive post when I suggested ANYTHING BUT new graphics. This is the kind of mindset many people have these days, and the poster is right - the vast majority share this mindset. This is frankly the answer. I don't blame the devs, I blame the players.

Thing is, at least judging by what the person interviewed in the piece above said, even if they did spend a long time on it, it wouldn't get that much better as a result. AI is something they work on to the point where, at best, it's good enough not to completely kill immersion.

Example: It's no secret that Panzer Corps was made as the spritual successor to Panzer General, which cam out 18 years before it. Now, Slith/Matrix aren't big games houses, but - at least speaking from a lay-man's point of view - you might have expected a really massive improvement in the quality of the AI just based on better hard-ware. Actually, whilst Panzer Corp's AI is definitely an improvement over Panzer General's, the improvement is no-where near that you'd expect based on the improvement in hardware available in the intervening 18 years.

Same goes with some of Paradox's releases. HOI3 was, especially after TFH, a definite improvement over HOI2 - but given the intervening time, this improvement doesn't seem proportionate to the improvements in available hardware. Again, I'm speaking from a law-man's point of view here - I can't see what's going on under the hood and wouldn't understand it even if I could.

You're right to say that a lot of people shoot for eye-candy, at least initially. I can't believe strategy gamers do this, but they do. Good AI, though, is necessary for replay value. This is what let Total War down.

 

[riknap]

you shouldn't be too harsh though when devs, and companies in general, spend resources on eye-candy.

a good AI, like GalCiv's for example, is guaranteed to not only ensure replayability, but attract interest on its own. But therein lies a problem, ie. it can only attract players (ie. consumers) who have an innate interest in strategy and/or familiar with the genre. this essentially means that the "consumers" it attracts are the "connoisseurs" and not the "casuals" (the latter who, let's be honest, outnumber the former the same way "elites" are outnumbered by "rank-and-file")

a good UI, however, is guaranteed to attract the passing-glance and tempt the "casual" into trying the game out in the first place. it also assists in the replayability aspect by making the game more relaxing, ie. less eye-straining in comparison to contemporaries. Total War, to use your example, may be a relative let-down in the AI department, but the average-player would rarely notice the difference but would more easily notice the UI-difference, which Total War scores high on.

It's ultimately an EU-ish slider between UI-focus = more casuals/less depth and AI-focus = more connoisseurs/less accessibility. And as a business, they need to decide how to stay profitable and expand their market without damaging their existing support.


As a personal example, I was a Total War fan myself, having gotten into the strategy genre via Rome Total War. I've been familiar with Paradox's other games for some time now, but what made the slide easier for me was EU3: Divine Wind, whose map was aesthetically pleasing enough that I could tolerate it (it does have much less detail than TW games) enough to engross myself into the mechanics and thus actually care about AI in the first place (hell, now I'm even a modder :laugh: )

besides, the article itself discusses how AI-coding is mostly now limited to the amount of man-hours spent in coding and optimising, and only significant computational breakthroughs can further improve AI. besides, just how accurate is the player's conception of the AI anyway? what may seem like a brilliant move of the AI by one player might be seen by another player as a retarded and suicidal action, given the exact same scenario.
at the same time, improving graphics is only limited by the hardware speed available, which regularly increases as time goes on. it's obvious that graphics would advance further and faster than the AI. and, unlike the AI, it's easier to be more objective about how "better" the previous iteration is in comparison to the former

I'd personally liken it to neuro-surgery and dermatology, as the simile kind of works :laugh:

 

[Pellaken]

In Victoria 2, any nation with a player can rise to GP status almost without effort. This is one example of where I see the AI failing. The AI should not be this far behind a player.

 

[No idea]

In Victoria 2, any nation with a player can rise to GP status almost without effort. This is one example of where I see the AI failing. The AI should not be this far behind a player.

Well, theuth to be said, that is not just the ai fault, but the player already knowing how to "beat" the game.

On topic. First of all, i think ai quality is a bit subjective. You cant always tell if the ai has made a good move or a bad one because you dont know everything the ai "player" knows. Prhaps you are not aware of other things the ai player is aware, and you think that movement was a bad choice when it might be a good one.

Second. I think The more complex games become, the more difficult it is to make a good ai. I mean, games made 30 years ago were not as complex. Take for example space invaders (if that game can be said to have an ai) or a chess game. They are far less complex, from making an ai point of view, than a total war game, for example. There are far, far more things to have into account and on top of thta, those things change every second

Third. I think it also depends on the actual gameplay of a game. If the gameplay relies on things computers are really good at (calculating, for example, a game that was about building things and getting a prfit as quick as possible) the ai can become better than a player (if the game is just about that). On the other hand, if the game is about being inventive, then the ai will be at a loss really soon, as the player will soon learn the ai tricks and will counter them, while the ai will never learn anything from the player.

 

[delra]

 

[FOARP]

Well, theuth to be said, that is not just the ai fault, but the player already knowing how to "beat" the game.

On topic. First of all, i think ai quality is a bit subjective. You cant always tell if the ai has made a good move or a bad one because you dont know everything the ai "player" knows. Prhaps you are not aware of other things the ai player is aware, and you think that movement was a bad choice when it might be a good one.

Have to disagree with you here. AI quality is not very subjective because there are many things we see AI doing in games circulating today that are objectively wrong. Examples:

1. HOI3: suicidal invasions
   
2. Panzer Corp: not loading up units to use in amphibious/airbourne attack
   
3. Victoria 2: wars every 5 years after being beaten half-a-dozen times already
   
4. Crusader Kings II: suicidal attacks by armies of less than ten men on other armies much stronger than them,
   
5. Empire - Total War: Units milling around in front of your units whilst you annihilate them,
   

Second. I think The more complex games become, the more difficult it is to make a good ai. I mean, games made 30 years ago were not as complex. Take for example space invaders (if that game can be said to have an ai) or a chess game. They are far less complex, from making an ai point of view, than a total war game, for example. There are far, far more things to have into account and on top of thta, those things change every second.

True, but irrelevant. AI has not improved in proportion to the way it has improved in other areas for games of exactly the same level of complexity. Examples:

1. Panzer Corps is better than Panzer General, but not amazingly better despite an 18 year gap between games.
2. Arma II was no great improvement over Operation Flashpoint or Arma I in terms of AI, despite a ten-year gap between OP:FP and Arma II. You still have much the same path-finding issues and AI stupidity in both.
3. Empire: Total War was not much more complicated in terms of game-play than Shogun: Total War, and the AI in both games was not too different, this despite nine years between the release-dates of the two games.

Hell, thinking about it, the first war-game I ever played had AI (Stonkers) that at least knew how to path-find its way across a map, over a bridge (the best place to beat the enemy) and into your base, and which would beat you if you gave it the chance, and it was a hex-based wargame that came out in 1983. There's games coming out even today where the AI simply can't beat you even if you leave it an open goal.

Third. I think it also depends on the actual gameplay of a game. If the gameplay relies on things computers are really good at (calculating, for example, a game that was about building things and getting a prfit as quick as possible) the ai can become better than a player (if the game is just about that). On the other hand, if the game is about being inventive, then the ai will be at a loss really soon, as the player will soon learn the ai tricks and will counter them, while the ai will never learn anything from the player.

Again, I have to disagree here. Vicky II is a politico-economic sim and the player still gets to out-pace the AI in terms of building their economy. Sure, I don't expect AI will out-match a human opponent during my life-time at wargames, but I would have expected to see a marked improvement in AI-quality in the past decade or so in proportion to the increase in processing power, similar to the improvement in graphics - and I haven't.

Have to admit that ARMA II was the last game I played that would qualify as an FPS (and then not really), before that it was FarCry 1 (which I enjoyed immensely). The map on the left is E1M2 from Doom? There was a big improvement in AI between Doom and FarCry, at least in my books, but I wouldn't be surprised if not much as changed in AI terms for FarCry 3.

 

[Kadanz]

The reason might also be that the game has to be playable for 'everyone'. And there will always be people who complain about everything just because...well they like to complain about everything. Making the AI to good, people start to complain about...'paradox hates their costumors because they make their AI too hard'. I can already see the many forum threads popping up...'AI is cheating, game is too hard...'

The AI cannot be too hard, many people hate to lose. some kind of pride thing I guess...losing against a computer software. It cannot be too easy either, games which are too easy get boring very quickly.

I can see why AI improvements stay behind than eye candy. It has to be accesible for new players as for the strategy veterans.

 

[FOARP]

The reason might also be that the game has to be playable for 'everyone'. And there will always be people who complain about everything just because...well they like to complain about everything. Making the AI to good, people start to complain about...'paradox hates their costumors because they make their AI too hard'. I can already see the many forum threads popping up...'AI is cheating, game is too hard...'

The AI cannot be too hard, many people hate to lose. some kind of pride thing I guess...losing against a computer software. It cannot be too easy either, games which are too easy get boring very quickly.

I can see why AI improvements stay behind than eye candy. It has to be accesible for new players as for the strategy veterans.

I don't think anyone would actually complain about, say, the AI being capable of pulling off a successful amphibious or airborne attack. Indeed, when HOI3 has seen improvements that allowed it to occasionally do so, all I've seen is praise for this.

It is true, however, that AI has sometimes been buffed by some companies by giving it all sorts of advantages - I don't particularly disagree with this in principle, since human intelligence is likely to always be better than AI, but it shouldn't be just an obvious way of compensating for poor coding. I can't think of a Paradox product in which this has happened yet though.

 

[Lothos]

Money.

More people will buy games with good graphics than they will games with good AI. Look at this http://forum.paradoxplaza.com/forum/showthread.php?665425-New-Sprite-packs-and-music/page6&p=15208973#post15208973 it's from a paradox game and it's a completely dismissive post when I suggested ANYTHING BUT new graphics. This is the kind of mindset many people have these days, and the poster is right - the vast majority share this mindset. This is frankly the answer. I don't blame the devs, I blame the players.

Being an AI nut job that I am for the HoI 1, HoI 2 and HoI 3 series what this poster is saying is pretty correct. Most people buy games due to eye candy and the AI is either assumed or is a secondary thought. But on the flip side the AI is the first thing people complain about after playing any strategy game for about a month. A good AI is what extends the shelf life of a game. So it all comes down to what the companies long term expectations are.

If they want a game where they can release lots of small content and make money off of over time then delivering a good AI from the start would feed that. They take an initial hit early on because the budget for the game would be larger which then makes profit margins lower but in the long term putting out lots of DLC content will replace that and then some.

If they are looking for a quick burst with a short shelf life then the more eye candy you put in the more money you will make from a short term boost. Most companies have been folloing this strategy as it brings in cash faster, makes share holders happy. Problem is they have to keep cranking out game and game because shelf life is so short but that is easier to explain to shareholders than to take a big hit in the begining.

 

[magritte2]

Graphics improvements come directly with technology and while in theory more number-crunching ability could be used for better A.I. in practice, people have to write better algorithms. In games like chess with a small number of possible choices, it's relatively easy to build good A.I. but that's not what Paradox's games are like. If you define an optimal strategy for the A.I. to follow, once the player has watched it for a while, it becomes predictable and can be exploited. If you randomize it to prevent predictability, you inherently introduce sub-optimal choices. And in Paradox's games, some players want historical A.I. which isn't necessarily good, since real countries didn't behave like optimal game players because they didn't know the "rules" and their objectives were often not ones players would choose.

 

[DanSez]

As many have said and I agree -- that the eye-candy/graphical interface is a necessary opening door to attract a large customer base, but a good challenging AI is needed for long term commitment to the product. PI walks a fine balance with a new game launch (looks good, works ok) and over time and expansions, some UI and candy are added, but a the holes in AI are discovered and patched to an extent impossible upon first release. Not a bad compromise and one that keeps people employed and entertained.

I think the real solution will have to start with a new hardware base. Think about it -- with the need for the eye-candy/improved graphics, pressure was put on chip and memory manufacturers to provide faster solutions - now we have separate graphic card processors and memory banks that relieve the computing stress on the main processor(s).

There will be an 'adaptive processor' developed (already in development?) where codes for 'Reason' are treated like sprites, balanced in a hierarchy of conditions (again, programmable), freeing up the main processors to handle things like coordinating multi-threads, handing human (slow) input and monitoring other feedback and input systems. The Graphic Processors we know today started off as a small set of assistant chips on the motherboard. In about 20 years you will be able to go to your 'electronic game site' and order a HAL 3000 logic board and upgrade your machine to take advantage of the latest 'Adaptive' programming available.

The initial drive comes from the needs of industry and the military - (vehicles that can drive established routes without accident, aircraft that can fly without human intervention, mining and manufacturing robots working independently under inhospitable conditions) - all these things are in the process of becoming reality and the pressure for decisions-consequences programming are only increasing.

 

[Accrsd]

Interesting piece on AI in computer games featuring one of the lead CKII developers:

http://www.ign.com/articles/2013/04/08/whatever-happened-to-video-game-ai

My basic take away from this is that the reason why the improvement in AI over the last 15-20 years has been seemingly less than that in other areas of computer games is that there is eesentially a law of diminishing returns in coding AI - to give computer AI instructions making it capable of behving intelligently in 99% of circumstances takes far more coding that getting it to behave properly in 90% of circumstances.

Did I read it right?

the answer is actually far simpler then all the other posts in this thread make it out to be. in fact i can summarize it in one sentence. Pretty colors don't make you smarter, they just make you pretty. translation: It's easier to change the art style of a game then it is to make the game smarter.

 

[Sn3ipen]

While it is true that graphics sell more than a good AI. The problem is what is a good AI? A competing AI thats almost impossible to beat, an AI that acts historical(Read scripted) or one that is trying to find the middle ground? The AI in paradox games try to act like it have free will and will make irrational choices sometimes. Sometimes its on purpose to ad immersion and sometimes it is something thats overlooked during development.

I also think that I have way higher expectation from the AI today then I had as a kid. Then it was all about the childish fun of playing computer games. I don't think I even noticed or cared if my opponents where simply standing there and allowed me to attack them as long as it was fun to do it and it gave me an ego boost. The more I play games, the more I tend to expect from the AI to keep me playing.

I don't know how many hundreds of hours I clocked in EU3 before I started to actually see AI glitches. It was simply because of every playtrough I got better and better at beating the AI. Nowadays when I have gotten used to PDS games, I see the PDS AI do stupid things after some years of gameplay in every new game I try. Not because it is bad, but because my brain semi-know how to beat it before I even try it. Thats probably the reason why people find it really great, when the AI finally manages to launch a successful naval assault in TFH.

However, if they ever release a new game with an AI simply made for powergaming, they would turn of allot of new players. There is also the fact that the only way to tweak the AI is to play the game over and over again. That is something that is taking allot of time and effort to do. Its also time consuming to reproduce AI behavior in these kinds of games, where the AI are taking most of its actions with calculating maths and the roll of a dice. Both the numbers it have to crunch and the numbers on the dice will change for every playtrough. It will even change on loaded saves.

PDS are taking the AI seriously. They recently hired Wiz from this board as an AI programmer. He have apparently been working on HOD AI and stuff. Hope he actually is an AI programmer and manage to do some amazing stuff in the future. That he isn't just payed with a lifetime supply of games to work as a scapegoat, so we have someone to yell at.

 

[Lothos]

Who is the kid they hired?

 

[Wizzington]

While it is true that graphics sell more than a good AI. The problem is what is a good AI? A competing AI thats almost impossible to beat, an AI that acts historical(Read scripted) or one that is trying to find the middle ground? The AI in paradox games try to act like it have free will and will make irrational choices sometimes. Sometimes its on purpose to ad immersion and sometimes it is something thats overlooked during development.

I also think that I have way higher expectation from the AI today then I had as a kid. Then it was all about the childish fun of playing computer games. I don't think I even noticed or cared if my opponents where simply standing there and allowed me to attack them as long as it was fun to do it and it gave me an ego boost. The more I play games, the more I tend to expect from the AI to keep me playing.

I don't know how many hundreds of hours I clocked in EU3 before I started to actually see AI glitches. It was simply because of every playtrough I got better and better at beating the AI. Nowadays when I have gotten used to PDS games, I see the PDS AI do stupid things after some years of gameplay in every new game I try. Not because it is bad, but because my brain semi-know how to beat it before I even try it. Thats probably the reason why people find it really great, when the AI finally manages to launch a successful naval assault in TFH.

However, if they ever release a new game with an AI simply made for powergaming, they would turn of allot of new players. There is also the fact that the only way to tweak the AI is to play the game over and over again. That is something that is taking allot of time and effort to do. Its also time consuming to reproduce AI behavior in these kinds of games, where the AI are taking most of its actions with calculating maths and the roll of a dice. Both the numbers it have to crunch and the numbers on the dice will change for every playtrough. It will even change on loaded saves.

PDS are taking the AI seriously. They recently hired Wiz from this board as an AI programmer. He have apparently been working on HOD AI and stuff. Hope he actually is an AI programmer and manage to do some amazing stuff in the future. That he isn't just payed with a lifetime supply of games to work as a scapegoat, so we have someone to yell at.

Wait, I'm supposed to get paid in something other than games?

Who is the kid they hired?

That would be the 30-year old kid that is me. Hi.

 

[1alexey]

It is surprising nobody mentioned multiplayer.

Since probably arround the time WC3 came in, the vast improvements in internet infrastructure made people playing games with people. A huge if not most of games have AIs for introduction/practice, and the "meat" in multiplayer. Take major mainstream games: DOTA2, LOL, Starctaft 2, COD, Battlefield, ex.

Why waiste money developing good AIs, when you can provide people with a better thing, game against real human opponent.

 

[Darkgamma]

Because people such as me sometimes plain don't care about multiplayer

 

[duhsveti]

Sooooo Wiz while you are here and we have your attention, are you really working on the AI?