What Would You Like to See from Game AI?

jtogel asks: "As someone working in new (bio-inspired) AI research with an eye to applications in games, but within an academic setting, I often hear that game developers are not incorporating cutting-edge academic AI into their projects because it's too "risky" (they can't really predict how gamers would react), and because they don't see the point in it. As a gamer, and as someone who cares what gamers think, I am often surprised by the sorry state of current commercial game AI - it has hardly moved since the 1980s. However, maybe the problem is that no-one really knows what we want from game AI. Academics keep coming up with innovative AI technologies, but what we should we use it for? What do you think? What sort of intelligent behavior would you like to see in games, but don't at present? Which are the most obvious intelligence deficiencies of current NPCs that need to be fixed?"

Comments

Turing

[eldavojohn]

For the purposes of this discussion, I will target the two extremes of games. We have game type I in which the game space is very much finite and complete search trees are exploitable and can easily be enumerated.

Let us take the simple game of tic-tac-toe (three by three) as our game space. Not too hard. Tic-tac-toe, checkers, cognac, are popular games that fall in this category. And when you use AI to its full advantage, the computer rarely loses.

As we move towards type II games, we see games like chess or go where the game space is too large to search but still 'algorithms' (or heuristics) can be defined that prune the tree space or match/strive patterns out of raw moves. Look up tables are also useful but not really "AI" in the strictest of sense.

A full fledged type II game would be something like Warcraft or a complex computer game in which the internal engine of the playing board (server in this case) has an innumerable amount of states that it can exist in.

The NPCs and AI in games like Doom or even Halo are still fairly simple. They rely on heuristics and Euclidean distances (conceptually within the game) to overcome their opponents. If they are faced with obstacles, they deviate from their path.

Now you ask me what I would like to see. I have very finite desires and I will list them here:

• AI that learns. I am so sick of AI not keeping a log of encounters. There are many learning algorithms out there and none of them apply to full emersion games (like SWG or WoW) and its because there are too many variables--too many parameters to build the small dimensional vector fields of the Win or Learn Fast (WoLF) algorithm. This isn't the only kind of learning though, what about statistical analysis of prior opponents? "I'm a level 54 mage encountering a level 52 warrior, my odds of survival are x% based on past encounters with warriors two levels below me..." Yes, it's pre-canned statistical analysis but things like this could seriously act as a good heuristic for many games.
• If you reduce the game space or find some really good heuristic (as in the above), dumb it down. Meet the user's pleasure level. It should be a challenge but not such a challenge that victory is unattainable.
• Use your imagination. That's all I'm going to say about that one. There is no cookbook for heuristics ... learn to throw things at the user.
• Introduce long term "bounty hunter" style AI. I would enjoy an AI that sleeps and is constrained like I am. Introduce simple work parsers to our enemies. The closer you can come to interacting with me on the level of a real human, the more I'm going to like it. I know there are robots out there that understand basic human speech, why can't there be games where we receive phone calls or e-mails from assassins in real life? I know this sounds ridiculous, but make me afraid of my foe outside the game. Get in my head.
• Introduce random variables. I don't care if you have to build in mechanisms to your game that are illogical so that the AI sometimes goofs up. All humans are fallible at a certain point and if you have differing levels of predictability in your AI, the user will love it.
  
  

In the end, my opinion is that the largest deficiency for game AI is that it fails the Turing Test.

Horribly.

I can easily tell that I'm not playing another human.

Re:Turing

[Anonymous Coward]

> why can't there be games where we receive phone
> calls or e-mails from assassins in real life?

EA's Majestic.

Re:Turing

[Anonymous Coward]

I can't believe you managed to write all that so succinctly and still be fast enough to get first post... :-o

Re:Turing

[Wind_Walker]

So what you're saying is, you'd prefer that Ragnaros emerge, kill all of the healers in the raid, then begin picking off the DPS one at a time as they die from a lack of heals? Because he keeps a log of what has happened in the past and he realizes that it's the healers keeping the others alive, so logically he would kill them first? Or, he sees the same 40 people as last week, and decides not to emerge at all, thus denying your guild the loot from another kill?

That sounds like fun to you?

Re:Turing

[Anonymous Coward]

I don't play WoW, but the first part does indeed sound fun. If your group falls apart at the loss of the healers, then it's time for a change in strategy.

I mean, what happens in PvP? I'm sure if "dead healers = teh win", then that's what other players are going to do, no? And you must defend against that.

Re:Turing

[Anonymous Coward]

But you don't -- that is, not really. In PvP you die so much more frequently than in PvE. PvP is more like Unreal Tournament deathmatch. You want every part of the game to be like that -- multiple deaths to achieve a single goal, no "lifespan" longer than 5 minutes? yeesh, no thanks.

Re:Turing

[joshdick]

"Introduce random variables. I don't care if you have to build in mechanisms to your game that are illogical so that the AI sometimes goofs up. All humans are fallible at a certain point and if you have differing levels of predictability in your AI, the user will love it."

Sure, NPCs should goof up now and again, but most of the time, they should at least be rational; they should always take steps not to die.

That means:
* No NPC should stand by idly as I charge toward it with weapons drawn.
* If I swiftly and easily kill all N-1 enemies in a group, the Nth one ought to get a clue and run away with its tail between its legs, beg for mercy or at least try a different strategy.
* No NPC should forget about me simply because I walk out of its view.
* NPCs should use teamwork.

Also, there should be variability among NPCs, even of the same type or class. For example, in "Halo," all Elites are the same, so once you figure out how to beat one, you'll be able to beat 'em all.

Re:Turing

[Spaceman40]

You do realize that half the strategy in an MMO depends on the idiotic AI they're using? That, if the AI was actually intelligent, the games would have to be different?

You think it's fun because the reason you're winning is that the enemy is too stupid to beat you? You concede that if Ragnaros actually thought about it for a second (even before dying the first time), he would go for the healers first; therefore, you're winning not by strategy or tactics (which are meaningless against a pattern - you're just responding with a pattern at that point), let alone initiative, but by brute strength and strict orders? Where every fight with him is the same as any other - random patterns aside?

Personally, I'd rather fight a much weaker (but much smarter) boss, where I actually had to show some initiative, where the fight is different from week to week. Ever played the same chess game twice with an experienced player?

Eh, I understand the draw for a 40-man raid, though. I'd rather play 32 on 32 Tribes, though.

Re:Turing

[Anonymous Coward]

You could get around that problem by giving him HP and armor much closer to that of what players can obtain. If he used some sort of strategy that actually made him hard, vs hard by having a lot of health and/or insane armor, it would certainly make it more interesting.

At this point, once a guild learns how to manipulate him to their advantage, it's really not much of a contest any more, just a huge time sink to get more equipment/cash. How is that fun?

Re:Turing

[Anonymous Coward]

Called off in a hurry after 19 participants in the game started taking things a little too seriously.

Re:Turing

[Minwee]

You do know that the reason that "boss" encounters are what they are is that they were designed around the staggeringly dumb and predictable AI that most MMORPG mobs use. If the AI were particularly good then ten players on a "raid" should get all the challenge they need from running into a group of ten monsters of the same level.

Since the games are all balanced around the AI, of _course_ changing the AI would unbalance them. The thing is that even if the encounter you described were to change all it would mean is that this sequence of events would follow:

1) A few hundred people will scream bloody murder on the forums since they can no longer get free candy from the Easter Bunny just for showing up and following the same script that everyone has used since November of 2004.

2) Regularly scheduled Loot Collection Visits will stop as everyone takes their toys and goes somewhere else for Ridiculously Easy Money.

3) A few People With Too Much Time On Their Hands will actually try to figure out how they can beat the new, more challenging encounter. Eventually they will come up with some loophole or flaw in the AI's logic which allows them to win.

4) The same hundred idiots will hear about this and continue to scream about how this only proves that Blizzard hates them and is in league with the girls who won't go to the Prom with them because the rewards haven't been changed.

5) Then the servers will crash. Despite what the conspiracy theorists think this won't have anything to do with steps 1) through 4), it's just something that happens a lot.

Re:Turing

[Meagermanx]

Not if the monsters are weaker than you. And anyway, I'd be happier paying for a challenge than paying to fight a 25 line AI script.

Re:Turing

[mshurpik]

why can't there be games where we receive phone calls or e-mails from assassins in real life? I know this sounds ridiculous, but make me afraid of my foe outside the game. Get in my head.

You mean like, "Case...this is Wintermute. We need to talk."

Or have you seen the movie eXistenZ by any chance?

Re:Turing

[KDR_11k]

That would require that there's a way to stop the monster from reaching your healers. If he can just walk over to them and hack away it's less of a challenge and just unfair. Game boss battles usually rely on the player being outgunned and having to outsmart the enemy to succeed. If the enemy was smart enough to use his full power effectively he'd usually be impossible to defeat.

Re:Turing

[aitikin]

Wow. I have to say, I was going to make some crack saying "'What do you want in AI?' just that, AI". But now that I've read your post, all I can say is wow. That described it right to the bottom line for me. The only major problem I see with this is mentioned already. I don't know of many gaming systems (computer or otherwise) that has anywhere near the processing power necessary to implement such algorithms or heuristics. For a game like WoW or SWG, that'd take a shit ton of power (never played but from my understanding very intricate). Regardless, I almost completely wish games had gone that line of logic rather then the "If it has tons of eye candy, people will LOVE it" mentallity. Don't get me wrong, looks are a plus, but they aren't everything.

Re:Turing

[I Like Pudding]

Personally, I'd rather fight a much weaker (but much smarter) boss, where I actually had to show some initiative, where the fight is different from week to week. Ever played the same chess game twice with an experienced player?

So your ultimate scenario is to play chess with a weakling? Companies must be throwing money at you to design the next big MMO.

Re:Turing

[xtracto]

Yeah, I as the original poster are an AI researcher (currently trying to do my PhD). I read a really nice book about AI in videogames, I would recommend it a lot to anyone interested in that, it is called (mm could not find the name, I promise to add the name of the book tomrrow as it is in my house). Anyways, the book explains the different AI techniques currently used in the different kinds of games, which currently are very standard searching techniques.

AI that learns. I am so sick of AI not keeping a log of encounters. There are many learning algorithms out there and none of them apply to full emersion games

Sure, there is enough AI development for that, take fore example genetic algorithms for evolution, the concept is not hard at all and surely will improve NPC behaviour a lot.

As for the risk for development houses in using CSc.-AI sure it is, that is why it will be a startup company with nothing to lose the one that will use the new approach.

One of my current projects is using the RPG toolkit to create a world with only NPC characters, I am still thinking the idea but I plan to implement plenty of AI concepts in the "game". For example, using JADE & JadeX for the NPCs, and use some of my research (about uncertainty management) too. Of course I will add genetic algorithms and neural networks.

The idea is to throw as much AI I can. The graphics will be very simple but I will concentrate on the AI.

Oh! one of the problems of using classical AI concepts is the processing time constraints, however there have also been advances in that direction as research in "Real Time" solutions is being done too (look at www.scopus.com)

Re:Turing

[Bad Boy Marty]

See http://memeticai.org/ for some work that was done for BioWare's Neverwinter Nights series. Interesting stuff, and pretty promising, but *hard*. The great thing about using NWN as the game is that there is a very accessible scripting language available, so the only thing you need to concern yourself with is the AI itself. It might not be the approach you want to take, but it does illustrate the potential for building AI in an RPG.

Re:Turing

[DerWulf]

no he is saying: instead of giving bosses 10 queens (chess) but no intelligence, make the board even and and give the computer the ability to learn and be smart.

Re:Turing

[ShakaUVM]

>>In the end, my opinion is that the largest deficiency for game AI is that it fails the Turing Test.

Sorry, but people don't actually want that. They *think* they do, but what they actually want is an AI that varies its behavior occasionally.

If I were a GM and could play Onyxia in WOW, you know what I would do? I would walk over to the nice squishy priests, kill them all, then proceed to kill the druids and shamans, then kill the mages and locks, followed by rogues, hunters and finally warriors.

Which is the exact opposite of how it goes, and how it should go.

A boss that could pass the turing test is an unbeatable boss.

On the other hand, I'd like to see boss encounters where the PCs actually have to react and change their tactics, instead of just reading a web page somewhere on how to beat it. Bliz does a couple stupid random things, like varying the breath weapon element in BWL, but that's about it.

So, in conclusion, I think the AI in WOW sucks, but for a totally different reason. =)

Re:Turing

[Mattcelt]

Hmmm, perhaps it wasn't written by a human...?

---

(Seriously though, the little * next to the userID means the user is a /. subscriber, and gets to see the stories a little earlier than the rest of us. S/he had plenty of time to write it before posting was available.)

Re:Turing

[meringuoid]

If I swiftly and easily kill all N-1 enemies in a group, the Nth one ought to get a clue and run away with its tail between its legs, beg for mercy or at least try a different strategy.

THAT'S something I'd like to see. I've been replaying KOTOR this last week. During much of this game you're moving between four planets looking for parts of a Star Map to lead you to the villain's secret base. On each one, somewhere there's an ambush by three Dark Jedi who are out to kill you.

The first one of these you encounter is a hard battle. The fourth, however, is likely to be terribly easy, because you've levelled half a dozen times in the interim, while they've been hanging around in the cantina ogling the dancers.

So, along come these three Dark Jedi, to kill me off for their master. They charge towards me... BZZZZZT! Force Lightning, one shot, all three of them are instantly reduced to one-quarter health.

Do they

a) run like hell the other way?
b) kneel before me and swear allegiance to me as their new Dark Lord?
c) give up the whole Sith thing and go and become bantha herders?
d) just shite it and stand there in a panic?

Nope. They keep coming. Admirable loyalty to their cause, I suppose, but really, really stupid.

Same with all these RPGs... in Neverwinter Nights, my warrior was wandering the docks and getting jumped by the local bandits. And killing them with a single katana blow. And then using the Great Cleave feat to do it again, and again, and again, butchering up to six of them in a single turn. And still they come, rushing eagerly forward to be massacred. I mean, come on, show some initiative, fuck off and hide and throw darts or something, you know?

Or how about the gang of pirates who attacked my party in Baldur's Gate 2? Just after coming out of Spellhold? Here we are, a bunch of mighty heroes in bloodstained armour, openly carrying massive swords most of which are glowing and one of which is talking, on a street full of ordinary civilians, and these pirates choose us to mug. What the hell were they thinking? We chopped them all up in thirty seconds flat and fed 'em to Boo.

I'd like AIs to realise when they're out of their depth and leg it. It would add a certain something to the experience... your sorcerer fires off his Super-Duper Slaughtering Bolt of Greater Mayhem into the midst of the crowd of orcish assailants, half of them drop dead on the spot, and the rest of them run. Might add a little to the old underused lawful / chaotic axis, too: do you mercilessly butcher fleeing enemies, shooting them in the back with Crossbow Bolts of Sycoraxic Vengeance or perhaps a spell of Belgrano's Greater Torpedo, or honourably let them live and risk running into an ambush they prepare a little further down the road?

Re:Turing

[Anonymous Coward]

As well as because of concerns that someone's grandmother was going to receive one of these calls and have a heart attack and die\, leaving EA liable.

Re:Turing

[Jett]

I can't find any info about this - everything I see says it was called off temporarily after the 9/11 attacks and then called off permanently awhile after that because it wasn't making any money. I can't find anything about people taking it too seriously.

Re:Turing

[Anonymous Coward]

> Or how about the gang of pirates who attacked my party in Baldur's Gate 2? Just after coming out of Spellhold? Here we are, a bunch of mighty heroes in bloodstained armour, openly carrying massive swords most of which are glowing and one of which is *talking*, on a street full of ordinary civilians, and these pirates choose *us* to mug. What the hell were they thinking?

/snarfed to .sigfile.
//should be engraved into the plastic on every monitor on every game designer's desk at every game company in the world.

Re:Turing

[mugnyte]

Things are sadly more complex than this. AI dealing with a fairly consistent audience can be tuned to delight that audience. But once you ship a game, it needs to satisfy the "game quality" judgement from a large bunch of players.

  When shipping a game, there are checks and balances that need to take place. AI can learn, and in fact can learn quite well. This could mean your encoutners are vastly more difficult than you expect. Certain players may amp this up, and certain down, but there are so many variables to AI that summing it up to a few "dials" is too idealistic, IMO.

  Also, AI can learn incorrectly. For example, everyone has their favorite memory of creatures fighting one another in a game without the player's interaction (perhaps their instigation though). In this scenario, the AI may cause a "lesson" that completely changes the game, where creatures avoid one another, or seek to kill one another over the priority of fighting the player. Or they may simply learn to avoid fighting altogether, or always fight to the dead, and so on.

  Simply put, AI with too much of a learning-pattern change takes away from the "gamey-ness" of your purchase. In fact, this "gamey-ness" is a collective opinion from many hours of demo with test groups, so it's hard to know where to even limit such learning patterns.

  Perhaps if a game were to build several "races" of creature, each with a learning pattern that sweeps both the "shipped" intelligence and the "learning capability" settings. Best, one would have a big config file to tweak such things, enabling folks to change the entire feel of the game (along with the look, as done now). When I say big, I'm expecting hundreds of variables per race. To even test all your settings, you'd want a sandbox with scripted encounters. Hook in a feedback loop and suddenly you have a semi-evolutionary system. Now you've written several man-years of code, and not much game. A development house simply can't dedicate that much to AI.

    I'd look to a future where inputs for "creatures" are normalized, so that every game can re-use a variant of the AI engine. For example, simple inputs like Others, Obstacles, Context and a myriad of parameters for the race's behavior, and then an output signal (quite complex) - that any game could use to push their creatures around. I'm not aware of such an engine as yet, but that may be because the game world as a group has just too many variables to try and normalize them.

Re:Turing

[Jett]

You hit the nail on the head - this is exactly the reason I prefer playing multiplayer first person shooters over RPGs - chances are if I've got a rocket launcher and my opponent is out of ammo they are going to make a calculation: Can I take him with my melee weapon, get more ammo quick, or escape before I get blasted to hell. They are actually assessing what kind of a threat I am and responding to it in a realistic way. Even better, you can't really predict that response because every person is unique and there are a lot of other variables to account for. And even better than that they are going to try to survive as long as possible UNLESS they decide to go kamikaze - it's not like with AI where as you point out, the typical response is to fight to the death every single time, even when death is assured and the best the AI can do is scratch you. Playing against the "AI" in games just doesn't compare to an actual human opponent - it's like playing Chess against a mentally disabled child most of the time. The sad thing is it really hasn't gotten any better - compare the AI in an old NES game to even something as recent as Oblivion - NPC conversations are just as inane, NPC behaviors are usually just as stupid. The great advance in Oblivion: Most non-enemy NPCs have schedules - they go to sleep at a certain time or stand in a specific location at a certain time (often that location is staring at a wall).

Re:Turing

[meregistered]

Everyone worries far too much about 'balance'.
The 'balance' in most games is artifical. Not actual balance, but perceived balance.
We tend to think of things as being fair or unfair in our lives and this reflects the balanced or unbalanced thinking in our games.
To make actually balanced games (which challenge our intellect and skills instead of our sense of fairness) a new approach must be made.
Unfortunately it will offend many peoples sense of 'fairness'.
With the millions of people playing I think actual balance would engender an impressive following fairly quickly of those of us who despise synthetic balance.

-ME®

Re:Turing

[Zemrec]

Actually, the daily schedule idea for NPCs goes back to the old Ultima series. I think it was Ultima V that did it first, although I could be wrong.

On Oblivion, though, you make it sound more simplistic than it is, there is some level of AI/scripting intelligence to enemies, but still, lower-level and weakened enemies still keep coming for the most part. Occasionaly they do other things. There's a mod that makes animals behave more real world-like, over at http://www.tessource.net/files/cache/index.html, lots of other good stuff there too.

On the other hand, most online multiplayer games I've played aren't much better. In Shooters, for instance, when you die you almost always respawn instantly afterwards, and it's a simple matter of finding a better gun than the default and you're back to where you were before. So most players don't use any strategy and aren't too concerned if they die on a straight ahead attack at you; they'll just respawn and give you payback.

Re:Turing

[twistedsymphony]

As dumb as it is I think the AI in Oblivion is better the most games we've seen recently.

Most animals will blindly attack you if you enter their territory, they have no idea that the sword you're wielding can take them out in a single blow. Similarly animals like deer will always flee.

NPCs have varying levels of intelligence, if they're an ally and following you, the smart ones will sneak when you sneak and follow your lead. The dumb ones will charge in ahead of you, get in your line of fire and often get themselves killed. Similarly if you attach someone who is un armed or weak, depending on their personality they'll run away. Smart Archers will keep their distance, zombie archers will stay in one spot. Smart blade fighters will block and dodge, keeping a small distance away unless attacking, if you sit there and block they'll wait for you to get your guard down before attacking. And dumb blunt fighters will run in Kamikaze.

It could still get a lot better but just walking down the street you hear the NPCs starting conversations with each other and I frequently hear stories about different weird scenarios that played out due to the AI. It still has a lot to be desired but it's a whole lot better then anything else I've seen, when battling computer controlled NPCs I often feel that I'm on a level playing field and they're bound by the same rules that I am.

I think to make the best AI experience possible you have to imagine how you'd build the game if the opposition was player controlled, then simply write AI that replaces the "player".

Re:Turing

[Lonewolf666]

By all means, let him try.
Make his decisions fallible (random glitches in whom he attacks?) and reduce his strength to compensate for his better smarts, but allow him a few clever moves. That will make the game more interesting.

Re:Turing

[Anonymous Coward]

Everytime this comes up at /. I can't help but reply because I love this subject. I (think and hope) I see it coming that the insatiable game community moves to standard AI APIs/HW and code in the same way that 2D moved to 3D over the last decade. It looks as though rather than AI we have to at least go through real physics first which is just now making the transistion from SW to HW. I would dearly love to see the transition from infant AI SW implementation to mature SW implementation to proprietary API to defacto industry standard API to HW accelerated AI.

Re:Turing

[meringuoid]

The great advance in Oblivion: Most non-enemy NPCs have schedules - they go to sleep at a certain time or stand in a specific location at a certain time

Haven't played Oblivion, but there was one game I played in which scheduled NPCs were done well: Majora's Mask. Clearly this only works because the entire game takes place over Groundhog Weekend, and so the designers can schedule every last move over those three days in intricate detail... I doubt you could realistically do it with a more open-ended game.

Re:Turing

[Jett]

I guess I haven't noticed how smart the enemy AI is because I always sneak around and snipe people long distance with arrows. Most of the time they just stand there and take it like they don't even notice - the few who figure out that they are being killed charge right at me, sometimes they might throw some spells at me first but it seems like they always end up charging.
I have not noticed different behaviors from companion NPCs, I haven't had many so far but the ones I have had seemed to rush and attack even when I'm trying to sneak around and snipe with my arrows.

Re:Turing

[Jett]

Re: multiplayer shooters - I like the mix of skilled and unskilled players and I like game rules which add a little bit of structure so there are clear objectives (e.g. capture the flag). You end up with a variety of challenges - you've got the noob who always gets a certain weapon and always tries to do the same thing with it, you've got the clan guys who team up and try to complete the objective, you've got the campers defending bases, etc. Most shooters these days seem to be pretty balanced as far as weaponry goes too - it's not really a matter of getting the best gun and rushing the enemy with it, that can certainly work but if you use a little strategy you'll be a better player.
Which reminds me - another major issue I have with game AI is that more often than not a "harder" enemy is just as stupid as the easiest enemies except they have beefed up attributes. Compare that to a multiplayer FPS where every player has the same attributes and access to the same resources - a "harder" enemy means a smarter opponent and is thus an actual challenge, i.e. you aren't overcoming stacked attributes you are overcoming someone who is actually better than you at what you are doing. To me that is a hell of a lot more fun than beating up on graphical representations of a bunch of stupid scripts that can't tell I am even attacking them most of the time.

Re: Oblivion - I did install that mod, as well as the one which makes takes away guards stupid psychic powers (and a few others which are supposed to improve enemy AI). They definitely help improve the game. I'm not saying Oblivion sucks or anything, I'm not big into RPGs to begin with and the AI in it just doesn't live up to the hype. Also that game is buggy as all hell. All that aside, it is pretty fun and I have spent several hours playing it. I think part of the problem is that they tried to create an open-ended immersive world and succeeded quite a bit which makes the areas where they didn't succeed stand out a lot more so than it would in some oldschool isometric RPG.

Re:Turing

[Spaceman40]

Precisely.

Re:Turing

[twistedsymphony]

yeah there's really only 2 or 3 NPC "personalities". Most of the time when you're cave diving the baddies you'll meet there have a kill or be killed mentality. I found that most of your targets in the Dark Brotherhood will flee if given the chance, and most necromancers you fight in the Mages guild will keep retreat if you get too close but keep their distance in general.

I wouldn't call the AI "intelligent" but it's definite more adaptive then the scripted AI we're used to. Basically certain actions happen if certain criteria are met and the criteria you meet is completely dependent on how you play the game, So two different people can have two completely different experiences if they have different play styles. Some interesting glitches though. I broke into a Church undercroft and after 3 verbal warnings from the NPCs there they started attacking me. I yielded and they stopped but immediately after stopping they started attacking again... because I was still trespassing. Another one was I attacked an Archer guard in the Imperial City and he missed me and hit another guard. So that guard attacked the one who shot the arrow, and then more and more guards came and started attacking each other, this continued until they all died. Because they're programed to attack and kill anyone who attacks a guard, but apparently they don't distinguish between intentional attacks and accidental friendly fire. So the whole thing caused an AI chain reaction and killed off all the guards in the immediate area.

Again, a lot more could be done to improve this, and it's not anything we haven't seen before, but it seems to be pretty rare in games these days. I think one of the things I really like is some key NPCs can be killed and prevent you from finishing things later in the game. A good example is in the Thieves guild you have to steal something from the Arch Mage in the Mages Guild, if you kill the arch mage while trying to steal it, you can't complete the mages guild questline later because the NPC that gives you your quests is dead. Is that kind of dynamic playing that makes the game really interesting. Yeah I'd be really pissed if I killed someone who's intrical to the plot later on, but it makes me think long and hard about my actions in the game. If everything is scripted then there's no real consequences and I don't have to think too critically about my actions. Worrying about long term consequences give the game a very REAL feeling. If you were to tie that in with AI I think it'd be even better.

Re:Turing

[Anonymous Coward]

I enjoy City of Heroes in part because follows all of the parent poster's rules.

It's particularly fun / weird that they have set up the game so that once you are virtually impervious to a mob's level (mob lvl + 5 lvl), they will continue training, arguing, jumping around, "fighting" among themselves and other interesting scripted animations, like you're not even there. Unless you actually punch one of them. In which case, most will actually run away and yell "I'm outta here!" or something to that effect.

The last men standing may escape when you have gulped down a mob from N-1 to 1 or two. If you use "Taunt" (to make them stop to attack you so you can catch up to them) they could stop, look at you closing up on them, and run away no matter how strong your taunt levels are. This "scared to hell" behavior is seemingly random, so it doesn't get too boring and you just go find yourself a new target mob.

When they run away, they tend to join stronger mobs too, as if taunting you to try and disturb their newly found protection.

One other thing I like is that the server never generates the mobs while players are looking. This is becoming more common, and I applaud not having to see mobs spawn out of thin air, or even all around me so fast that I'll die the second I try to jump away. That's not AI but immerses me just the same.

Re:Turing

[Anonymous Coward]

Actually, I think he's referring to games like Majestic. And really, if that turns his crank he should be playing online ARGs (ilovebees, the new ones based on Lost, etc.).

Re:Turing

[Anonymous Coward]

everything I see says it was called off temporarily after the 9/11 attacks

Right, and how many people were involved in the 9/11 attacks.

Re:Turing

[Suicyco]

Absolutely that sounds like fun! Do you enjoy cookie cutter recipes for defeating monsters? Thats just rote bullshit, boring as hell. What if the enemy actually thought about things, changed its strategy, etc.? What if the big boss changed his minions around, moved them to better spots, changed the order in which you see them, etc? Man, that would be awesome. Guess what bro - "fun" is not guaranteed kills. That is boring as shit. Fun is the fear of death, strategizing, running away going "WTF!!!" and having to completely rethink the encounter. I would love that. As it is, these encounters are lame, boring and simply following the recipe for victory.

Since when is "fun" equated with completely linear known results from repetitive activity?

Thats why I hate listening to folks whine about twinks in WoW. Because its not "fun" to lose. Just think of those guys are really smart bosses, and good luck. I would love to be able to finally defeat a super smart, devious, powerful boss. Victory should be hard fought and truly be an accomplishment. Not "gather this many buffs, sneak in here, shoot that guy, wait 5 seconds, hit him with 10 of this, switch to the other guy walking by (wait 3 secs to shoot him) blah blah blah blah." BORING.

Frigging twitchy kids. You have no sense of accomplishment, just a sense of entitlement to "fun." You whine and complain that things are "hard" and not fun for you. I just don't understand how a known outcome of a known sequence of events is fun in any fucking way. Of course, I have a fully functional brain that hasn't been spoon fed "accomplishment" because I am breathing and somehow entitled to it.

Re:Turing

[Scoth]

"Goin' to the bear pits tomorrow, you wanna come with?"

Still the best NPC convo I've ever heard. Had me rolling on the floor.

Adapting

[The_Wilschon]

I'd like to see game AI that adapts to the ability of the player(s). That way, someone with little patience could start playing and see a reasonable rate of success immediately, but still enjoy things later as they become more skilled.

Re:Adapting

[vaporakula]

Dynamic Difficulty Adjustment is a really hot topic in games at the moment. Choosing a difficulty level is just soooo last gen ;)

Some really horrible examples are seen in most racing games - ever notice that if you want to win at Mario Kart, it's best to not be in first place on the final lap?

The techniques are becoming more sophisticated though. The biggest danger is currently that it's really hard to get the challenge "just right" - when balancing games, the smallest adjustment can make things swing really wildly... so you often end up with a game that is oscillating between too difficult and too easy.

Re:Adapting

[GloomE]

Yeah, difficulty settings should adjust relative to your skill, not define some absolute skill.
I like being pushed hard in a game, but some may like to casually throw a few rail slugs around while concentrating on the scenery, so it's still nice to have control over difficulty even if it is dynamically adjusted.

Re:Adapting

[Anonymous Coward]

The other posters are right, it has been done - more often than you'd expect, actually. Elder Scrolls IV: Oblivion bases almost everything on the player's level - monsters, randomized loot, even the items the merchants sell. Sports games sometimes have "catch-up logic," where the AI for both teams dynamically adjusts based on the score.

I agree that games do need to adopt it more often - even if it's as simple as turning it down for every death and up for every finished level.

Re:Adapting

[steveo777]

Well, with Mario Kart, I think it has to do with the 'rare' items. They tend to appear more often during the last lap (lightning bolt, blue shell). Whether or not it's multiplayer.

Surprise!

[Kjuib]

I would like to be able to come from behind on an enemy and have them be surprise.. maybe a surprising shout.. and shooting all over the place cause they don't know where I came from. That would be good. And difficulty not relying on how well they aim randomizer is or how long until they shoot, but depending on how much they learn from their previous attempts to kill you. A setting for auto adjust difficulty is always nice.

Learn routes, etc. from the player

[i.r.id10t]

Why can't the game AI learn routes, etc. from the player? One of my biggest gripes with bots is that they either only follow a (few) set pattern, as opposed to a player displaying (to the game) many other diverse routes and locations.

Re:Learn routes, etc. from the player

[technos]

Not just that, but learn from a couple different types of player.. Right now, the AI in most games seems to settle on one and only strategy to win. While learning from players is a step, the logical next step is to learn from several types of players.

Say you're playing a FPS. Among players you have kamikazes, base campers, spawn campers, and the guys that run straight for a place to snipe the enemy, to name a few.

How much more interesting would it be to say "AI1, you're a kamikaze. Go draw their fire. AI2, you're a base camper. AI3, you're a sniper. I'm going to go grab some grenades and have fun Kentucky fishing by their flag" when forced to play with AI in your faction?

Re:Learn routes, etc. from the player

[Cyberscythe]

I thought Quake III displayed this sort of AI learning. One of the programmers wrote his CS thesis on the Quake III AI.

Re:Learn routes, etc. from the player

[saintp]

This reminds me of Battlefield Vietnam, a generally great game where the AI left something to be desired. In one level, there are two islands linked by a long bridge. If, as the Americans, you could take the low-lying island and the bridge, snipers at the bottom of the hill on the second island could rack up insane amounts of kills -- hundreds in very short games -- because the VC never learned to either a) take a different route (like going through the trees, or around the hill, or anything besides straight down the road, towards the snipers); b) spawning more snipers, to direct some fire at the American snipers; or c) using their tanks or artillery in an intelligent fashion to take out the very concentrated American base of fire. On another level -- one of the Ho Chi Minh trail levels, IIRC -- the American start with five control points and the VC start with one, separated by a river and a bridge from the American control points. The Americans can easily take a bunker near the bridge, lay prone on top next to an ammo box (and out of the firing line of the VC) and lob grenades over the top, onto the bridge, and kill hundreds of VC. The VC never try lobbing a grenade back, or firing a mortar, or using any indirect fire weapon on the Americans on top of the bunker. They never swim the river and flank the Americans. They just die like lemmings. These are two examples where a learning AI would have been of great benefit. The VC would have seen their fellow soldiers being blown to bits, adapted to it, and made the levels much more challenging and fun.

Re:Learn routes, etc. from the player

[Anonymous Coward]

Have you ever played FEAR? The AI in that game can be somewhat predictable at times, yet in some cases you may be blown away by the actions they undertake. I'll give you a few examples, the second of which I've experienced once and it was very interesting:

1) You'll be up on a catwalk, and 3 troops walk into the room on patrol. You open fire, but they all manage to take cover before you can finish one off. From your vantage point, you can pick any of the 3 off, should they reveal themselves. While waiting for your eventual reward, you are instead surprised to hear "FIRE IN THE HOLE", and notice a grenade hurled at you from behind cover. You jump down from the catwalk, to take cover of your own. You peek out to see two of them approaching, you begin firing and take one of them out. The other takes cover, and you're back to the waiting game. But wait, wasn't there a third? News flash, he's already to your left and about to unload 00 buckshot into your rib cage.

2) You've been ambushed by many soldiers, and death seems inevitable. However, you unleash your "inhuman" ability, and manage to clean the floor with all but one. This last one is taking cover in an office with 2 large windows, and only one door. He screams "Oh shit! I need cover!" Of course, there's no one left to cover him, you just took out the whole platoon. This last soldier only has an SMG, you have healthpacks and plenty of armor, and you want to have a little fun. You decide to just run in and go "Chuck Norris" style and kick him through the window. As you run in, this last soldier gets up, proclaims "Oh shit!" once again, and promptly runs backwards while firing. Just before you deliver the kick heard round the world, he jumps through the window you were planning to kick him through, in order to escape your furious feet.

The first examle can happen quite a few times. The AI can and occasionally will use smart tactics to catch you off guard or gain an advantage. I can't tell you how many times I was searching for the third guy before I realized he had already worked his way up through cover to get to my side or even behind me, while his buddies kept me busy. The second example has happened to me only once, however, it greatly surprised me. I never expected the NPC to jump out of the window, glass and all, to escape. In seemingly 99% of games, he would have stood straight up and stood there firing. In this case, it appeared to me he knew he had no chance of winning, and took the only route available to him to have a remote chance of living. It was surreal, as he put his chin down and dove through the window, I just stood there with my jaw dropped, as he hit the ground in a roll and came up firing, hoping to get away.

The AI in FEAR was surpisingly refreshing at times, since you were forced to make progressivly smarter decisions to quickly dispatch all the enemies laid out in front of you. I certaintly recommend it, and I hope AI of this sort becomes the norm for shooters out there.

Re:Learn routes, etc. from the player

[RFC959]

This reminds me of part of what impressed me about the Combat Mission games: the AI will learn very quickly to avoid dangerous areas. If you set up on the defensive, covering an area, and then when the enemy rolls in you open up and zap a few of their tanks, they will figure out how to take cover from you and flank you.

Not that the AI was perfect, but it was very good at reacting like that. The thing that bugged me most about the AI of troops under your command was this: You order a unit to run from cover, across 100 yards of open terrain, to some other cover. They'll get up and start running, taking some fire, and then when they're 90% of the way there, they'll decide to TURN AROUND AND RUN BACK, across 90 yards of open field AGAIN, instead of forwards 10 yards to cover. I guess this is because the AI doesn't know that the place they're running to is cover - it only knows about places the unit has already been.

Yes, There Are Many

[Anonymous Coward]

Is there a question in there?

Yes, in fact, there were three or four at least:

Academics keep coming up with innovative AI technologies, but what we should we use it for? What do you think? What sort of intelligent behavior would you like to see in games, but don't at present? Which are the most obvious intelligence deficiencies of current NPCs that need to be fixed?

Now would you actually care to discuss them?

I'd like to see...

[grub]

... A CHALLENGE!

The only times I can recall game AI really impressing me was when I played Thief, Thief 2, System Shock 2 and Far Cry. Not all the games' AI have high 'wow factor' now but at the time they rocked.

Re:I'd like to see...

[Anonymous Coward]

I recall watching a google video about "blackman" which is a brood war player, and in that video the comentator is telling us that blackman actualy practices against specially programmed AI that he made himself using strategies of high level players like nada or yellow just to name a few. I think this is very interesting because the AI only seems chalenging before you actualy go online and play against some people, who use very different strats(rushes and so on). By incorporating good strategies, such as those of high ranked players, the AI can be much more challenging and that can be a great help to the player. When you play against human opponents, what happens is that 50% of the time you either face a very strong opponent or a very weak one, and this can be either good or bad on your gameplay.

Re:I'd like to see...

[sophanes]

It's not just the quality of the AI, it is how effectively it is put to use.

For example, the AI in Thief/Thief2 wasn't particularly advanced, even by the standards of the day - simple pathfinding and patrol paths, with AI behaviour linked to the alert status of that entity. However, unlike other games where the AI appears to be an afterthought, LGS made the AI in the Thief series integral to their level design and game-play philosophy. By making the AI react to player sounds and environmental cues such as blood stains/bodies/missing items they were made to appear much more intelligent than they actually were. The range of dialogue uttered by the AI made them appear more real than they would have been without it.

Re:I'd like to see...

[jesdynf]

What are you talking about? SS2 didn't have AI, just terrifying atmospherics and cunning placement of creatures. The "AI" had two settings, "run at you and kill you" and "shoot at you from a distance".

Re:I'd like to see...

[Anonymous Coward]

SS2 had similar AI to the Thief series. Monsters could hear you and react.

Maybe games have to advance for AI to advance

[snookumz]

The developers of Red Steel for the new nintendo wii said that they upped the AI players combat skills, because the new controller allowed faster and more accurate aiming than a traditional controller. They're point was that game AI is usually braindead to make it reasonable to the players who are slightly hindered by an unnatural interface. It's common in video games for a character to stand without any cover at the end of a long hallway. This isn't because the programmers couldn't program the AI to look for cover. Game characters aren't often built to be tactical, because they'd creme the gamer. That's my take.

Re:Maybe games have to advance for AI to advance

[Anonymous Coward]

Usually the only difficult "AI" opponents in fps games are the ones with statistically superhuman aim. They aren't doing anything smarter; they're just shooting straight at you when you're in their line of sight.

What I'd like is an AI opponent with emotions, motives, plans, and expectations, but with a handicap on its theoretically perfect aim. I don't want a flat statistical handicap either. It's not enough to give it an aim spread without also taking into account its current emotions, motives, plans, and expectations.

I don't believe it when you say game AI is intentionally made braindead to let players compete. Even at its top notch, current AI would appear braindead in most games; it's the ability to perfectly and instantaneously perform certain actions (like aiming at an opponent) that makes it seem good.

Re:Maybe games have to advance for AI to advance

[cgenman]

Very True.

You really don't want artificial intelligence. You want the game to behave in a way that seem plausably intelligent but are actually beatably stupid. It's much more exciting to be swarmed by 7 or 8 nameless mooks who die at the end of your sword than to spend 1/2 hour chasing down one guy who keeps hiding behind cover, shooting you from the back, and running away when injured.

Game AI isn't about AI any more than buildings in games are about livable architecture. AI is part of the illusion, but a small part. You have enemy groups that have goals, and feelings, and tendencies. And it's your job to kill them all. Or make them all like you. Or roll them up into a giant ball. Or what not. But they can't be too good at whatever it is they're trying to do, or the player's actions won't matter.

When AI is good, it takes center stage. It plays like it is the most important thing in the world, and to it... it is. But the player is what matters. And for the player to matter, he cannot generally be working against AI that is too powerful.

THAT HAVING BEEN SAID, it would be good to have better AI for NPC characters working with the player. NPC's that can figure out what the player's goals are. NPC's that can run through a level without bumping into freaking walls (Twenty years later and they STILL DO THAT!). They can't be too intelligent, or the player won't have anything to play, but they shouldn't be completely dumb either... An RTS where the player's commanders on the ground made tactical decisions based upon the enemy they were facing and the location they had been deployed to would be great. Imagine playing a game where you tactically deploy intelligent units, rather than micromanage the dumb shoot-and-runners we have now.

And for story-based games, it would be great to have more complex emotional simulators. Games like Oblivion or Final Fantasy fall down because simple human interactions are actually quite complex, and having characters that may care in heavily pre-scripted ways is a little dry. If you constantly take Tifa's materia and give it to Cloud, those two characters are eventually going to develop animosity. If character B keeps getting killed and character A keeps resurrecting them, a bond of either friendship or jealousy will form, depending on their internal states.

It would also be good to have some form of AI toolkit abstracted out. AI middleware, so to speak. I know they exist, but I also know they suck.

As a side note:

An AI research student came over to one of the companies that I had worked at. He was really excited to talk to the game AI programmer and find out what he did. "Do you use genetic algorithms? Adaptive data structures? Neural network pattern recognition?" The programmer looked over at him. "Uh... kid? We use timers."

Re:Maybe games have to advance for AI to advance

[Eideewt]

Actually, I would like to chase one guy for a half hour. I always liked chasing the Thief-bot in Descent II. It would sneak up on you and steal a weapon, then run away and do it again.

Re:Maybe games have to advance for AI to advance

[shadow demon]

Whoa, that brings back a ton of memories. Descent 2 was the first real computer game I (owned and) played. Unfortunately I never got very far, but I defnitely remember the thief-bot, it was annoying as hell! I think that that is a great example of an AI that can really get to at least some of the players. It was especially annoying when playing in the tons of hidden (and dark) tunnels, where it could easily sneak behind you and run away. Although I must admit that when I killed it after chasing it around all over the place, I felt a sense of success that I rarely even felt when finishing a level. All in all, I think it was a great addition to the game.

The AI is also why I prefer real-time strategy games. It seems like that's the only place where good AI can be used without fear of players whining about it. The game that I got after Descent 2 was Total Annihilation. It had pretty damn good AI, at least on the higher difficlty settings. RTSs are probably the only games that really benefit from AI nowadays, mostly because increasing its intelligence doesn't hinder, or even really change gameplay. I still play TA once in a(n admittedly long) while, and it is the last computer game that I have owned.

Re:Maybe games have to advance for AI to advance

[mcvos]

The AI is also why I prefer real-time strategy games. It seems like that's the only place where good AI can be used without fear of players whining about it. The game that I got after Descent 2 was Total Annihilation. It had pretty damn good AI, at least on the higher difficlty settings. RTSs are probably the only games that really benefit from AI nowadays, mostly because increasing its intelligence doesn't hinder, or even really change gameplay. I still play TA once in a(n admittedly long) while, and it is the last computer game that I have owned.

I think the reason why AI in RTS games seems so good, it that the AI isn't hindered by the click-fest/limited field of view interface that the human has to use. The genre that really needs some good AI is turn-based strategy. There you really can see how pitiful the state of the art still is.

Re:Maybe games have to advance for AI to advance

[Cyberscythe]

If the AI was perfectly human, you can always give the player some sort of super-human abilites to compensate. Like in Max Payne 2 where you get bullet time, or in Deus Ex where you can get nanotech upgrades for speed/sight/etc.

Re:Maybe games have to advance for AI to advance

[ShakaUVM]

>>Games like Oblivion or Final Fantasy fall down because simple human interactions are actually quite
>>complex, and having characters that may care in heavily pre-scripted ways is a little dry.

Oblivions "role-playing" is just choose-your-own-adventure, and rather horribly done because they use different voice actors for the different dialogue options (which just comes off as creeeepy). But -- their NPC AI is actually pretty damn sweet. I knew I was going to kill a bad guy, so I pick pocketed him and took his weapon off of him, then stabbed him in the back with his own weapon. He got pissed, but instead of punching me, ran over to some nearby corpses (other NPCs that had attacked me) grabbed a weapon off their corpse, and then proceeded to beat the crap out of me.

It's implemented with fairly simple rules, goal-fulfillment and some other stuff, but the interactions that result are actually pretty complex. I watched an NPC get hungry, and eat a head of lettuce. While he might have normally gotten away with it, a wandering guard was nearby and aggroed him. The guy took off at a run for the gates, but the guard chased him down and butchered him not far from his own house. I looted his corpse, took his key, and took everything of value in the house.

There's also several dead NPCs (all thieves guild) lying around in the docks district who at various times had been caught by guards and killed.

Whats funny is that all of those NPCs are alive and well in my girlfriend's game.

Re:Maybe games have to advance for AI to advance

[king-manic]

It's much more exciting to be swarmed by 7 or 8 nameless mooks who die at the end of your sword than to spend 1/2 hour chasing down one guy who keeps hiding behind cover, shooting you from the back, and running away when injured.

Err.. you just described the boss fight with The End in MGS:3. it was however fun.

Re:Maybe games have to advance for AI to advance

[MobyDisk]

It's much more exciting to be swarmed by 7 or 8 nameless mooks who die at the end of your sword than to spend 1/2 hour chasing down one guy who keeps hiding behind cover, shooting you from the back, and running away when injured.

That's the best summary of this whole issue I've heard. People like to be Rambo or Neo. There is great pleasure beating up 10 guys at once. But after a while it gets old.

Most intelligent gamers eventually grow up and don't play those games any more. They switch to multiplayer games because humans opponents hide behind cover, shoot you from the back, and run away when injured. Nothing is more fun than losing to a superior enemy.

Bots that aren't inherently aimbots

[drinkypoo]

I haven't played an unreal game in a little while (Except against humans) but last time I did, there were bots shooting me from across the map with a pulse rifle while I was dodging around, and they never missed. Bots should not depend on level design to make sure they don't whip your ass every time. They should be within the fallibility range of humans.

Re:Bots that aren't inherently aimbots

[MachDelta]

Unreal bots tend to have better than expected accuracy to compensate for their lack of 'creativity'. They just don't have the same kind of thinking power as a human player, so they have better aim and faster reactions to try and make up the difference. It does kind of suck on wide open maps where every enemy is a deadly sniper, but the very simple solution is to turn the difficulty down a notch.

Infact, Unreal has long supported a feature request i've seen mentioned in other posts on this topic - difficulty adaptation. You can turn on an option where the game will automatically crank up the difficulty if you're kicking asses, and (eventually) turn things down a bit if you're being stomped on. Eventually most people find a difficulty they're comfortable with and then lock it to that.

Oh and if the bots never missed (like, ever) then you're on Godlike. Take the rating literally - it is not intended for mere mortals. :)

Look to FramSticks for an example

[maynard]

FramSticks

Here is a great example of Artificial Life that generates truly unexpected behavior during runs. Perhaps this type of Alife simulation could be an inspiration for a new generation of game AI. Do not program a location and series of behavior patterns, instead make a population of AIs based on a variety of physical forms. Each form will have a limited set of possible movements within a the simulated world. It will need inputs in order to determine friend and foe, perhaps something similar to limited vision and hearing. It will need survival as a baseline goal. They will also need "food" of some sort. Perhaps the player or other, different, AIs can represent a food source. Give it some form of sex in order to reproduce its learned behavior through some genetic mechanism.

I suspect a hardware physics chip would help tremendously. But what I've seen of FramSticks was pretty damn cool. I have no idea how well it could be incorporated into AI gaming though. So this is just one of those: *shrug* hey, what about this? type of posts from someone ignorant and totally out of the field.

Re:Look to FramSticks for an example

[Anonymous Coward]

Give it some form of sex in order to reproduce its learned behavior through some genetic mechanism.

That really isn't as trivial as you make it sound to be. As the learned behavior gets more complex, the effort needed to copy that behavior (that is, extracting it from your learning mechanisms data structures and copy it with modification) increases as well. The problem becomes almost impossible to do in the case of between-gender inheritance of behavior. Remember that it took about 6 bln years to create the genetic code that allows humans to have between gender reproduction with inheritance of traits.

Nethack is already a sentient entity

[kognate]

Nethack has some of the best AI in any game, hands down. More games could learn from Nethack how to have creatures that are smart (like Jubelix, or a shopkeeper you have stolen from) and other actors (like pets) that do what you want them to, mostly, but still have enough brains to make gameplay fun.

Nethack: AI that's Int: 20 in my book.

Re:Nethack is already a sentient entity

[Anonymous Coward]

nethack's not really good AI at all, it's just a bunch of special-casing for all the different situations---artificial ai if you will.

of course, i still think it's the best game out there...

Re:Nethack is already a sentient entity

[Eideewt]

Dunno about that (I never got hooked on Nethack), but that reminds me of how impressed I am with Wolfenstein 3d's AI. It just goes to show you that AI need not be technically advanced to be good (although that surely helps). It just needs to keep you guessing. I liked how Wolfenstein did it because I never knew what the enemies were up to. They really moved around behind your back, so I could backtrack and find a roomful of bad guys where I expected none to be.

Radiant + tactics

[ToasterofDOOM]

For the most part, Oblivion's Radiant AI is more than I could have asked for. I love it that people have schedules and live out their own lives according to goals. Of course it has its rough edges, but it really is phenomenal. There is, however, one place it lacks more than others, and that is in combat. Sure, It's fine for a skill based RPG where split second decisions don't have the potential to turn the tide, but combat AI is something that only recently has really been stressed. There are far too few games where enemies actually plan out an attack and work as a team. Give me serious tactical processing in AI, plug it into a military shooter, and you've got me hooked.

Re:Radiant + tactics

[Greyfox]

Yeah, though one thing that really bugs me is that the AI will notice a fallen comrade and won't get distressed about it. I was shadowing through a cave of necromancers last night and picking them off one at a time and several times another necromancer would walk by a body and look at it and just keep walking. No "Hmm, Frank's dead... maybe I should investigate that or raise an alarm or something..." And if another AI bot notices me sneaking around and starts yelling, that should bring everyone in range running too. I'm pretty sure some of the Metal Gear AIs did this.

Re:Radiant + tactics

[Oniko]

Heh, yeah. Radiant AI is fine for civilians (yeah, it's not perfect, but it's pretty damn impressive), but the combat....

Notes to all friendly Oblivion NPCs: If your buddy shoots big angry light-making spell thingy at big angry gonna-kill-you-all monster, don't get in between. It probably isn't healthy.

Also (and this was, actually, the more frustrating one), if there's a bunch of you, and a large group of monsters in an area, do not, repeat, do *not* all go haring after different monsters that are very far away. This gets you all killed, and then the squishy little thief/mage needs to hop around dodging many many bad things until the immortal(s) wake(s) up. Or she quickloads for the 10% chance that at least a few of you will gang up and kill *something*. This does not make the squishy one happy.

Or, say, you're a certain nameless Orc warrioress with only melee abilities, and your squishy little thief/mage buddy has ranged magic, a nice new shiny black bow, and is currently running right the @#!%&$##!@ past you to try and get you to kill the big evil giant-weapon-go-smashy guy on her tail that is turning her into scrib jelly... it'd be rather nice if you took your attention off of that archer on top of the building (that you have no path to and no way to hit) and saved your panicking, dying, and probably-hilarious-to-watch buddy. Just a thought.

Seriously, the ways that NPCs target could be improved. A lot. I know it's better NPC AI than is generally available elsewhere, and I love Bethesda for all that they've accomplished, but it's just really reeeeally frustrating sometimes.

Re:Radiant + tactics

[Propaganda13]

Team AI and communication is a great addition, some games are getting better at this. If a bot is able to see or hear me, call for reinforcements and work together to kill me. Coordination between the bots makes an average bot a lot harder to kill. A near dead bot falling back while its teammates move forward would also help.

Re:Radiant + tactics

[DeafByBeheading]

And if another AI bot notices me sneaking around and starts yelling, that should bring everyone in range running too. I'm pretty sure some of the Metal Gear AIs did this.

A lot of the stealth-based FPS games do this now. I think I first saw it in No One Lives Forever, but I'd be surprised if that was really the first game to do this...

Re:Radiant + tactics

[MoriaOrc]

If you think that's pretty bad, I've killed AI's having conversations with other AI's (mostly bandits in dungeons talking about how scary mudcrabs are while they wear full glass armor, but that's another story) and their conversation partner just turns around and finds someone else to talk to.

Don't get me wrong, Radiant AI is pretty cool most of the time (outside of combat, and when not talking about mudcrabs), but when it messes up it messes up bigtime.

Re:Radiant + tactics

[Eideewt]

I first saw it in Thief.

Re:Radiant + tactics

[Anonymous Coward]

I'm not sure Oblivions citizens count as AI. They're pretty much scripted, with some very basic handling for interruptions. And as far as I can tell, most of their "work" consists of walking around and chatting with people :P

It certainly added to the feel of the game, in the sense that it would have been odd without it, but they still seemed pretty dumb on the whole. It's also funny when a group of people spot you, kill you (or any other threat), and as soon as you're dead immediately get back to chatting about random stuff :)

Re:Radiant + tactics

[Greyfox]

Yeah, but during one Dark Brotherhood quest the character AI gets increasingly freaked out so it would appear that the AI is capable of that, but it would appear that it's a special hard coded behavior or something. I know if I were in some creepy dark cave with corpses already hanging up all around me, I'd get increasingly freaked out if my buddies started turning up dead. Maybe some of the survivors would cowardly run away and some would decide to go hunting for the culprit. That sort of sensable behavior would be awesome.

It's still a good step in the right direction, though, and I'm sure it'll improve over time.

Re:Radiant + tactics

[olego]

You should try out FarCry. For an insanely fun game, start in "Realistic" mode. If you're detected, the AI will circle around you and attack you from the back, if they're able to. It's kinda creepy.

Re:Radiant + tactics

[Miaowara_Tomokato]

I'm not sure Oblivions citizens count as AI. They're pretty much scripted, with some very basic handling for interruptions. And as far as I can tell, most of their "work" consists of walking around and chatting with people :P

What you say may be true... but how does this not reflect real life fairly accurately?

Re:Radiant + tactics

[Anonymous Coward]

Holy fuck man, you haven't played Ultima. Say Ultima VII, have you? Oblivion is the most braindamaged game I've seen lately. Do yourself a favor, check Ultima.

Re:Radiant + tactics

[ToasterofDOOM]

True, I just recently got FarCry but I've been playing on Normal since it's my first time through, but theAI in that game is usually pretty smart.

Re:Radiant + tactics

[ToasterofDOOM]

I only really got into RPGs until about 2 years ago, and only really have played Morrowind (My all time favorite) and KOTOR (also superb) I'll check it out.

Smart not always smart

[teh moges]

The thing is, we don't always WANT to have smart/unpredictable AI. IN Goldeneye64 for example, half of the fun was learning where everyone was, and trying to beat your previous time for that level. If the NPCs are going to be in different spots everytime, then I can't be sure that my 2.20min run was any better or worse then my friend's 2.15min run, since I may have had to overcome more obsticles. Great storylines and AI make a great game. Time trials, high scores, and the ability to compete with other humans (whether comparing high scores or multiplayer) makes for an everlasting one. That being said though, AI for first person shooters NPC "grunts" should be easy to defeat. I don't want to have to do too much to escape from my jail cell in teh first level. I want to be able to defeat the guard by sidestepping his first shot and knocking him out. After that however, I want the opponent's force's higher characters to sidestep MY shots and knock ME out if I'm not careful enough. Games like Unreal Tournament have the right idea: the ability to shout orders at your NPC teammates in battle to get them to do things you want. Now, we may be a full console generation from being able to plug in a microphone and literally tell the AI that you want it to follow you around this corner and give you some cover fire, but I should be able to input a key sequence to tell it that. Likewise, the AI should be able to work out what your strengths have been throughout the last couple of deathmatchs and tell you that maybe you are better off going up top and snipering, which the AI covers you. OK, to summarise: At the start, I want predictability. Past those levels, I want the AI to play according to how I play, minimise the effects of my strengths, maximise the damage to my weaknesses.

Re:Smart not always smart

[NeMon'ess]

If timetrials were half the fun for you, I'm sorry you didn't have enough friends for 4-person multiplayer. The designers didn't go through all that work just for you to do the same level over and over again trying to shave off seconds.

Convincing Sex

[Anonymous Coward]

I want an AI that can do a convincing sex scene, that's different every time, and pays attention to what I want. Since I like to have sex with people I find interesting, I want the AI to be interesting, as well.

I plan on waiting a few (but only a few) decades.

Re:Convincing Sex

[Anonymous Coward]

You want hot coffee with that?

Re:Convincing Sex

[The-Bus]

"I plan on waiting a few (but only a few) decades."

For the sex or for the AI?

RTS AI that can cut down trees

[adpowers]

Honestly. Why can't the computer in Warcraft III figure out it needs to chop down some trees in order to reach me (for the few maps that isolate players in forests)?

Also, most AI doesn't deal with running out of money/credits/gold very well. For example, in Warcraft, I like to have a big stock pile of gold when all the mines are depleted, so I can keep building units as necessary. The computer always seems to use whatever gold they have, so as soon as the mines collapse, they are stuck with whatever units they have left. If you can kill their current forces (which shouldn't be too hard if you have lots of money), then winning the game is easy.

Re:RTS AI that can cut down trees

[Mprx]

That sounds like good AI to me. Resources saved are, with very few exceptions, resources wasted. Far better to defeat the enemy before your resource production ends than stockpile resources and risk loss.

Re:RTS AI that can cut down trees

[Some_Llama]

Actually the AI in WC3 doesn't spend all of it's resources immediately, often times it hordes a lot of resources but never spends them, this can be seen by wiping out their main set of people and unit structures but leaving the castle/town hall intact and peons alive, the computer will continue to gather resources but never build any supporting structures.

I think it's due to an AI routine that doesn't reset itself, once a barracks or whatever is built the computer no longer considers that building needed, so when you destroy it then leave their base it never rebuilds it.

What I want in a war simulation

[jon787]

What I want in a war simulation

I don't know what I want to see

[Pluvius]

But I'll tell you what I don't want to see: Dynamic AI. Or at least the kind of dynamic AI we see now-a-days. I'm talking about those racing games where you're beating the hell out of the computer driver but in the last lap or so he ends up becoming completely inhuman (perhaps even "cheating") and beating you or coming close to it. Stop that shit, it's not any fun.

Rob

Re:I don't know what I want to see

[CastrTroy]

This really bugs me too. It was fun in NFS 2, when you were playing network games against people of varying skill levels, but it's not fun when you're playing against the computer and have to win to move on. I hate how you can be in the same car as the computer, and yet they blaze past you, even though you're going at top speed.

Re:I don't know what I want to see

[Anonymous Coward]

*cough* mario kart *cough*
fucking pisses me off

Re:I don't know what I want to see

[0olong]

I hear ya. I remember that from Mario Kart. You drive a terrible first few laps but you still win the race because instead of capitalizing on their lead the NPCs adapt their speed to your average speed. On the other side of the coin, if you drive a fantastic first few laps the NPCs adapt to your speed as well, so if you then drive a mediocre last lap you'll be one of the last to finish.

I think it's arguably the only real flaw in Super Mario Kart (SNES).^^

Re:I don't know what I want to see

[vaporakula]

The "cheating" part is the problem, there. There's nothing inherently wrong with a dynamicly adjusting difficulty level - it's just crude implementations that make it really obvious and steal the fun from you. It can be done, it just takes finesse...

Re:I don't know what I want to see

[Anonymous Coward]

Actually, I think you'll find that getting the shit pounded out of you and then coming back to win it at the last possible moment is an entirely human trait. It's called "the Rocky Effect". If you listen carefully, the best AIs even have very quiet "Adriaaannnn" SFX playing in the background.

Re:I don't know what I want to see

[Anonymous Coward]

This is horribly noticeable in Max Payne.

I don't like "catch up" AIs in driving games either, but in a way it's more tolerable there than in a shooting game. If you do well in Max Payne, the AI gets lightning fast reaction and extremely good aim, and that just takes you right out of the game, especially when you consider that the enemiess are all supposed to be human beings.

If I see the enemy first, I want that to be a huge advantage. One level in Max Payne 2 starts out with an enemy nearby with his back turned to you. If you've done well up to this point, try zooming in on his head with your sniper rifle, enter bullet-time, pull the trigger... and watch the guy react and dodge the bullet immediately.

If they can't make games challenging without resorting to tactics like these, something is wrong. I don't know how to fix it myself, but I definitely know something's broken.

Re:I don't know what I want to see

[AgentDib]

I definitely disagree with this. One of my favorite features in Max Payne 2 was the adjusting AI. It does get a lot harder if you are doing well, and that's what I like about it - it keeps things challenging. I am a longtime clan FPS player and there were several sections that I had to try 3+ times (especially the charge up the hallway toward the end). On the same note my roommate beat the game with limited FPS skills because the AI toned things down for him.

Perhaps it comes down to what kind of FPS gamer you are. I get really tired of runs through games where the "challenging" objectives are to beat it in a faster time, beat it without getting hit once, beat it with a certain low level weapon, etc. I'd much rather have the entire mission in doubt with the knowledge that if I don't make my first shot count and get behind some cover fast I'm going to get hit pretty hard.

Re:I don't know what I want to see

[JungleBoy]

Here here! 'Catch-Up' AI drives me nuts. It's a total programming shortcut. It's especially bad in Double Dash. You're 10 yards from the finish line, everyone else is half a lap back and out of the blue you get pounded with tons of items. MORE THEN THERE ARE PLAYERS! It drives me nuts, it even does this in multi-player sometimes.

A game shouldn't violate it's own rules just to 'add suspense'

-JungleBoy

How abou this?

[Random Utinni]

Maybe some intelligence would be a good start?

RTS games

[Xolotl]

My personal preference is for RTS type games -- StarCraft, Age of ... series and the like -- and to be honest the AI in most of them is pretty poor (actually StarCraft is one of the better ones, despite it's age).

So poor in fact, that on higher difficulty levels most of them resort to cheating in one way or another -- most commonly by upping the resource gathering/production rates (I've studied this by using cheat codes to show everything). Also, in many such games the AI plays 'perfectly' -- no mistakes as to when to develop which technologies, no problems controlling large numbers of units, and uncannily sending units just where I happened to be weakest for no reason (Age of Mythology seemed to be particularly bad at these sort of tricks). Having the computer cheat was no more fun than playing against a human who was using cheat codes, it ceases to be a battle of wits, which to me is what an RTS should be about.

Re:RTS games

[Webz]

From what I understand about StarCraft (and probably WarCraft) is that the AI scripts lack any sort of scouting abilities or just the notion of trying to look for players, in the way that human players do. Instead, what they have is "strength" detection. A computer controlled player in StarCraft would continually guage the "strength" of your base and your army at any given time. If the conditions were right, and the AI felt up to it, it would rush you. This is how the AI sometimes seems so good at taking expansions (they are physically much weaker than the starting base).

There was also a condition coded into StarCraft that after a while of silence (i.e. no script triggered attacks), the AI would just attack anyway, no matter what it was facing, just to keep the player engaged at least every 20 min or so.

A More Conceptual AI

[Jerf]

I have to guess how AIs are implemented in current software for this answer, and perhaps these already exist in some games. But I'm sure I haven't seen many.

What I'd like to see is more AIs that are fed some concepts a priori, and then are able to manipulate those concepts in some way. Let me take a Civilization-type game as my example, since it is there I really wish I could see this.

I'd like to see an AI that has some concept of "launching an attack", beyond just massing troops and attacking in force. I'd like to see it know about "pincer", and "distraction", and other such concepts, and then use those to simplify its planning phases. Traditional AI techniques are still a long way from being fed the rules of Civ and deriving such things de novo, so I'd like to see some cheating, not with resources but by not starting almost from scratch with planning.

I'd like to see the computer have concepts of building a city vs. exploiting it, with corresponding "cheating" done in the computations, so the computer does a better job of building cities rather than just being handed resources to cheat. (Disclaimer: The last Civ game I played was Alpha Centauri, so maybe Civ 4 has addressed this, and cities actually get built; in AC it is not uncommon to take cities near the end of the game that have the AC equivalent of a Granary, and nothing else. However, I still bet human cities are still significantly better than computer-built cities. Feedback welcomed.)

I'd like extensive simulations to be run by the game author to adjust the weights of these concepts by playing tournaments. I'd like to see the AI guys have some time to actually refine the AI post-release because there's just no other way you're going to get a good, balanced AI.

While I've used Civ as an example, this generally extends to other genres reasonable well. In the FPS examples everyone is citing, I'm not sure about the exact concepts I'd choose, but what I'd like to see is the AI guys having time to load up some set of concepts and then firing a Genetic-Programming tournament at the concepts to see what happens, then iteratively refine the set of operators and concepts based on feedback from the results. (Most likely collapsing some obivously-useful trees into single nodes in some cases, breaking other concepts out, and adding new ones as needed.)

GP could be really interesting here because that is known to produce some interesting interactions with differentiated participants; you could evolve an entire squad with specialized members relatively easily, if your program nodes were rich enough.

Note: I'm not saying to run this on the client machine; GP techiniques would be inappropriate in general on a single, isolated client. On the other hand, for something like an XBox360 game played over live, that would be large enough to run a GP-based tournament against human players as "just another AI".

I'm not sure that game AI needs much more than more respect and more resources allocated to it so that you can do something other than The Simplest Thing That Could Possibly Work. The required ideas don't seem like they're that hard, it's just that they pretty much all involve having some time to work with them and not just slamming out code and shipping it.

Re:A More Conceptual AI

[Jerf]

From this post:

Why can't the computer in Warcraft III figure out it needs to chop down some trees in order to reach me (for the few maps that isolate players in forests)?

That's another good example concept: When considering how to get from here to here, consider whether you can "terraform" your way there successfully. This was another major AI weakness in Alpha Centauri, too; if you built a ring around a continent, the computer was effectively unable to get to you until they can build an air force, and they still aren't very effective with the air force without the land support to back it up.

It's going to be a long time before an AI can derive that concept from just the bare information about its terraforming options, so cheat. Watche the smart beta testers and steal their tricks.

Re:A More Conceptual AI

[vaporakula]

FEAR uses this to really good effect. They called it Goal Orientated Action Planning - you can look up their presentations on gamasutra. It's a super-interesting system - your AI is given a Goal (kill the player being an obvious one) and has several actions that can achieve that goal - including things like Call for Backup, Change Weapon to Something Stronger, Throw Grenade, etc etc.

Re:A More Conceptual AI

[Anonymous Coward]

Galactic Civilizations 2 actually has a fair amount of this. It's limited by processor time and the imagination/capability of the designer, but he's documented lots of what he's done or will do.

Re:A More Conceptual AI

[bughunter]

If you want to see examples of how a good stratgic AI does things, play some of the fire-and-movement games from Avalon Hill: Panzer General, Allied General, and the one that still has an active ladder, Fantasy General.

Creatures

[Xolotl]

Nobody has meantioned Creatures yet ... simple neural net brains, genetics, learning (even from each other!) and all sorts of other behaviour, all done some ten years ago.

Unfortunately, all the current FPS/RTS etc etc games have ignored this completely.

Re:Creatures

[mcvos]

Nobody has meantioned Creatures

Creatures was great, and that sort of thing works fine in a sim, but it isn't easily translated to tactical/strategy games.

Besides, Creatures made some short-cuts too. As far as I know, it wasn't just NN+genetics, it was NN+genetics to tune known factors in an unchanging algorithm. Still, cute idea.

wishlist

[9mm Censor]

Here is what I want from AI in a game: Variation: not all opponents are alike. AI shouldn't be the same either. Bad guy 1 and bad guy 2 should not behave the same. Alike, yes, the same no. Scaling: AI skill should scale to the players ability. As I get better, I would like the AI to get better as well. Provide level adjustment for user adjustment. EG: Player skill = 5 AI skill = +/- X (set by user)

Easy to define, hard to achieve.

[deque_alpha]

Make it so that it feels like I am playing against / with other other humans. Some of them are good, some are bad. Some are utterly clueless, some strategize well. Some have very good aim, some are pretty poor. Everything should have some random element to it so that even very good AI's can botch, and even poor ones can make that one in a million shot. Make most of them "average" in skill. Make them adaptive, so that they learn my habits and change tactics to counter / complement them. DON'T allow them to do (seemingly) impossible things with a high level of regularity, and don't have one NPC's skill level vary widely within a session, like the "comeback king" drivers mentioned in another post.

It's really not that hard to define.

All that being said, make sure whatever is done is appropriate for the game. Games where one is pitted against a clock, or has an "obstacle course" sort of feel would likely not benefit a lot from having a real "human" AI. The repeatability of the challenge is core to the replayability of the game.

stop these tactics

[cbc1920]

For single player FPS games, I usually have a lot of fun. When things get tough, I invariably fall back on the following tactics. They depend only on the stupidity of the AI and its inability to learn:

1. Choke point: stupid AI's eventually run through a door, even if there are 20 of them and even if the last 5 guys to walk through the door are laying in a pile suffering from headshots. Can they figure out the pattern here?

2. Distract to higher firepower: you think the AI would wonder why the player is leading 10 enemies towards the gun turret? Probably not.

3. Shoot their legs: some of the slightly smarter AI will hide behind things, but usually leave a foot or hand sticking out. Sure, it takes 20 shots to kill them but they never seem to retract it.

4. Exploit routes: duh. Nearly every NPC follows a path until the player is noticed. Easy picking.

I'm sure there are many more of these, but it comes down to predictability, inability to learn, and inability to understand strong and weak positions within the level. One of my favorite game AI's is FarCry. Sure, they were still vounerable to the above tactics, but it would take a heck of a lot more coercing. Plus, NPCs would work together, always find cover, and try not to attack you from the front.

bad news

[prurientknave]

I know you don't want to hear this, but there's nothing wrong with game AI. Game AI always has to be a tease for the user, just a little hard to stress test some new skill the character needs, but never impossible. Obviously this is a contrivance, game AI's are nothing more than a little resistance to make everyone feel that they earned their way into the next step of the story. It's a little like the hazing one undergoes to join a club and this plot device is very effective in all the conflict category of games. The real fun of these games is to play multiplayer after you have acheived a reasonable level of proficiency with weapons, strategy and techniques. Ultimately, the point of conquering a foe in a conflict game is to show complete mastery over it, and for the majority of us there's no thrill to be had in mastering a game console with a limited storyline. Heck i could just chuck the console out the window if I really wanted to beat the reptilian invaders from gonzor :D.

When people whine about not having good opponents they're really whining about not having the ability to show off their talents in mastering something worthwhile; something difficult that would thrust them into the rarefied world of mythic heroes. This is why tetris, space invaders and games of that ilk were successes. The game would keep getting harder and faster and your talent was showcased by the point score you could attain and the initials you could leave at the arcade. The real competition has always been between people, either indirectly in the case of arcade games like space invaders or directly in the modern multiplayer arena. With franchises like the CPL rewarding this type of skill there is now a reason to hone one's skill in this arena well past your teens.(this holds true for professional sports as well)
The point is people don't really want amazingly hard advanced AI they just want something all of their loser friends can't beat but they handily can. This is also why people cry foul at gold farming and ebaying in rpg's. There's no point in striving to get a lvl 5000 sword of the beyondness if any shmuck can just buy it. The whole point is to lord it around the water cooler like a new BMW or new shoes. It's fairly immature high school stuff but that's really the key point in this issue. People are looking to become elites by their proficiency at mass market games. It's idiotic but that's the way the market goes.

The point is AI in these games are just fine. All you need to do is create standard levels with more and more obstacles, that filter the giants from the peons. We need levels that allow people to say 'HEY, I beat the crazy level with a troop disparity of a 1000:1 and there was a great ending but I can't tell you just how awesome it was.' Now this annoys the low-end gamer who seems to think he paid to play the complete story so he gets turned off games and that's another sale that won't happen for the next iteration of the game. So the important aspect of this model is balance. The GTA franchise did this pretty handily with the side missions idea. You can play the main thread of the story easily but you'll never get a 100% completion rate without doing the side missions which in the end makes no difference to the ending but you miss other cut scenes the hardcore elite player would see, and miss out on bonus powerups and the sheer elitism factor of finding all the hidden packages.... WHEEEEEEEE.

The point is to find a method to create elites in games who will crow about their abilities and actively get involved in selling more games than just your advertising alone can. It works out for them because it reinforces their elite status by increasing the morass of worthless average players and it also improves your bottom line by making more sales. Remember the point isn't the AI, the point is to play human psychology for all the ka-ching it's worth.

Re:bad news

[Xolotl]

When people whine about not having good opponents they're really whining about not having the ability to show off their talents in mastering something worthwhile; ... The real competition has always been between people, either indirectly in the case of arcade games like space invaders or directly in the modern multiplayer arena.

There is some truth in what you say, but not entirely. I play against humans and enjoy it, but more often play on my own, against the computer. Its like watching TV or reading a book; I do it to relax, to while away some time because I can't be bothered to go to sleep, or just because I've had a hard day and want to kick some ass. I may just want to play a favourite map/level, or there may not be anyone available to play against, or I may be in the mood for a different game than anyone else. In these cases my scores matter only to me. There are many reasons to play games and not all of them are directly competitive, and in these cases a good AI makes the gameplay much more enjoyable and less frustrating (and thus more effective for its purpose).

Re:bad news

[prurientknave]

When I play games to relax I've always rather enjoyed predictability of replaying an old mission. It's a great feeling to zip around thinking 'foolish AI! trix are for kids' :D I guess it's different for different people

Re:bad news

[Xolotl]

Yeah, I've had that too :) It rather depends on the kind of game, though. For example (showing my age) Descent was great like that, while Spaceward Ho! didn't have missions as such, just acheiving galactic domination. Different types of games, different kinds of pleasure.

Re:bad news

[prurientknave]

if you liked descent make sure you aren't missing out on the freespace 2 source code project. It's free and people keep making excellent new mods. There's a great babylon project mod but there's an even better battle star galactica mod in the works =)

Re:bad news

[Eideewt]

He's not talking about Freespace, he's talking about real Descent, which is an entirely different game.

Re:bad news

[Xolotl]

Well, true :) but Freespace looks cool too (I have the free source on disk, juzt no time to do anything with it).

Re:bad news

[Eideewt]

Yeah, it is pretty cool.

Re:bad news

[prurientknave]

I know.

Why do you care if it cheats?

[AuMatar]

I've never really understood this. When I play a game of Civ or AoK or StarCraft, I want the AI to be a challenge. I really don't care how it does that- extra resources, quicker unit building, whatever. SO long as it gets harder when I increase the difficulty slider, why does it really matter how it does it? Its going to be just as much a challenge either way, so in the end it doesn't really matter to me.

Re:Why do you care if it cheats?

[Jerf]

Because the easy way to beat this sort of AI is generally to find a trick or three that works, then ram the tricks down the AI's throat. The latter takes a bit of practice, admittedly, but compared to the level of skill necessary to beat a dedicated human opponent, the level of skill to defeat even a pumped up and cheating AI is much lower.

What happens in Alpha Centauri is that as you pump up the AI's cheat factor, the only practical difference is that it affects the odds of AI stomping on me in the first hundred turns or not. Once I get established, the AI loses almost no matter what.

I strongly suspect that this is a general pattern; as you throw resources at a bad AI, it gets linearly stronger, but at least in a Civ game, the human is getting geometrically stronger as the game progresses. The game at high cheat levels becomes a matter of luck, not skill.

Moreover, it tends to limit the experiences the human can have, vs. the full range of what is possible. I know the AI will come at me with the screaming hordes in a single mass, because while it has resources, it can't do anything with them. I never get to experience setting up a proper, in-depth defense, because I only have to defend against the relatively small subset of the possible set of attacks that the computer knows how to launch, that sort of thing.

I agree with you at low human skill levels, but a better AI really should provide a better gaming experience for moderate skill levels and above.

Re:Why do you care if it cheats?

[Xolotl]

Well, yes and no. If its done well, then fine. But often it is done badly, and then is simply frustrating. Like when the computer overwhelms you with a fleet of [whatever] which you know is simply impossible for it to have built without cheating in the time available, based on rates of production and technology development (for simpler games this can be calculated easily enough).

Worse, in some games the jump in difficulty due to this kind of cheating means that on one difficulty setting you defeat the computer easily, but on the next, same AI but with cheating, there is no way you can defeat it no matter how good you are.

The game then ceases to be one of strategy and becomes a race, or worse, a game of chance (who hits who first).

Re:Why do you care if it cheats?

[DeafByBeheading]

Its going to be just as much a challenge either way, so in the end it doesn't really matter to me.

So do you think an acceptable way to implement a higher level of difficulty would be to randomly discard some of your mouse clicks? To have some of your units spontaneously combust?

Part of the appeal of winning a game is facing an opponent on even terms. A somewhat spurious proposition to begin with when facing a computer, but I find it more interesting to fight "smarter" enemies rather than just "more" or "tougher".

Re:Why do you care if it cheats?

[julesh]

What happens in Alpha Centauri is that as you pump up the AI's cheat factor, the only practical difference is that it affects the odds of AI stomping on me in the first hundred turns or not. Once I get established, the AI loses almost no matter what.

I strongly suspect that this is a general pattern; as you throw resources at a bad AI, it gets linearly stronger, but at least in a Civ game, the human is getting geometrically stronger as the game progresses.

I used to see that a lot in the original Civ games, but I think in Civ3 they largely solved it by the introduction of new game rules that make it increasingly harder to grow a civilization beyond a certain size (the farther cities are from your capital, the higher the percentage of production and income is lost to 'corruption') -- this has the effect of pushing the growth of your power back towards linear. And I hate the rule, I think it's one of the worst aspects of Civ3.

I want AI to be obvious

[silvermorph]

AI isn't currently fun. It's used mostly as a difficulty slider. And when it's doing its most impressive things, I don't see it happening, so I don't care. So I want two things from AI: I want to know what it's doing, and I want to be the cause of what it's doing.

1) If I've got a bunch of bots on my team, and I rush out like a madman, I want them to be completely suicidal as well, or maybe cover me. And I want them to say "covering you" as they do it, so I don't go sit behind a rock as soon as I hear gunfire.
2) If I'm sneaking up on some AI enemy, I want the AI to recognize that I'm sneaking up on it, and then LET ME SNEAK UP ON IT. Maybe it notices that I'm right next to something I can duck behind, so it shuffles its feet and slowly looks around back towards me, giving me just enough time to respond.
3) If I'm in a deathmatch, I want clever, flawed, fair bots. Bots that know good spots to fight from, but get up and move after a kill or two. Bots that aim poorly when I get the jump on them they just picked up their mouse in panic.
4) If I'm fighting a squad of enemies, I want to overhear their communication somehow so that I don't think they're looking through walls waiting for me. And I want them to react slowly to what I'm doing, not instantaneously.

I want my AI to be fun, and I want it to be pretty obvious it's being fun.

Re:I want AI to be obvious

[prurientknave]

maybe they're trying to teach you something. There's nothing fun about shooting an RPG at someone? o_O

Re:I want AI to be obvious

[Spaceman40]

Somehow this is missing from a lot of games lately: we want them to be fun, remember? Not just all simulation?

Re:I want AI to be obvious

[crazydumbek]

1) If I've got a bunch of bots on my team, and I rush out like a madman, I want them to be completely suicidal as well, or maybe cover me.

It would be cooler if the bots began to alternate between saying "Oh crap!", "Damnit Leeroy!" and other curses, while futilly trying to cover you.

Straight forward

[Sigma 7]

"As someone working in new (bio-inspired) AI research with an eye to applications in games, but within an academic setting, I often hear that game developers are not incorporating cutting-edge academic AI into their projects because it's too "risky" (they can't really predict how gamers would react), and because they don't see the point in it.

That is not the correct reason - in reality, they don't know how to implement it within the ultra-short time frame.

Gamers will react favourably to properly implemented AI, even if there are a few rough points. However, if they want an AI that is just like the last game, they would just play the last game (and simply use the newer one only if it has interface improvements.)

However, maybe the problem is that no-one really knows what we want from game AI.

As an example of combat AI mechanics, I'll bring up a current game: Age of Empires III. On Moderate difficulty, I've managed to defeat 3 AI players in a skirmish by myself. While I do admit that this is on a lower difficulty setting, players don't want a 2-vs-1 match to still be a pushover on the highest non-cheating difficulty (with some exceptions).

A straight forward combat AI deficiency is in C&C: Red Alert 2 - three prism tanks flatten an enemy base, but human players actually attempt to send a few tanks before serious damage is done.

The Role-Playing mechanics for AI players is another story - they are much harder to implement. Even though this was the topic at hand, it is easily faked by using a highly detailed script (required for most plot points anyway.), as done in Arcanum. Otherwise, you're on your own.

Teamwork and "human-like" mistakes

[LoveMe2Times]

I'll admit I haven't played the latest-and-greatest games, so maybe teamwork AI has improved, but it always seemed to me that some fairly simple rulesets could dramatically improve enemy AI. Getting the bad guys to coordinate, lay down covering fire, simulate communication through phones or walkie talkies, try and cut you off, flank you, try and trap you into a corner, sneak up behind you, all kinds of basic stuff that could be randomly selected from a weighted ruleset based on character type (ie rank, weapon type, human/alien, etc) or intelligence. Also, if you take out the higher ranked officer, teamwork and communication can degrade. Perhaps other posters are correct that developers avoid this due to difficulty, but giving adjustable difficulty settings could cure that I'm sure.

Next, I base these comments on my experience playing ChessMaster. Chess might not be considered "real" AI anymore, however, there is one area that really is AI and I don't think is persued very much: trying to get the computer to play like a human. Let me explain, in case you haven't played a recent chessmaster. Newer versions provide a whole bunch of virtual opponents of different skill levels for you to practice against. Most of us mere mortals are never going to beat a chess program, so it's good to have easier opponents to practice against. The problem is, these weaker opponents don't screw up in anything remotely resembling the way a human screws up. They have these little dials that let you control look-ahead, time-to-think, randomness, and other factors. So you'll get an opponent nominally rated as a 1200 level player who will play with the strategic and tactical adeptness of a 2100 level player, and then suddenly throw away the queen by making a *random* move. Also, the computer is able to make moves instantly, say in 1/100 of a second. It is literally impossible to beat the computer on time, and the computer doesn't play differently when time is running out. Against a 1200 level human who only has 15 seconds left on the clock, you can almost certainly pressure them into mistakes by playing quickly and aggressively yourself. These same ideas apply to all computer controlled opponents. People always complained that Street Fighter AI could do things impossible for a human pull off, or that Civ AI got secret production bonuses, etc. But this would require a careful study of *poor* human play, to get a sense of what people thought they were doing and why. Hair-brained strategies and a lack of tactical awareness, doing things "just because I like to" or social reasons (being nice to/going easy on a friend IRL etc), unawareness of your opponent's strategy/options/impending attacks, unawareness of strongpoints/weakpoints for either side, not knowing how to captalize on them even if you are aware, and the list goes on. When I play chess at the local coffee shop, I see all kinds of weird reasons for people making stupid moves. If you could simulate that, at various human skill levels, it would change everything.

AI for My Character

[Anonymous Coward]

... so he/she doesn't get his/her a$$ reamed by a level 1 squirrel when I get up to take a piss. I'd like to be directing a creature that has enough common sense to run like heck when a 7 headed dragon falls from the sky. At least one that actively defends against attacks my grandmother could avoid.

I know this sounds too Sims for most people, but I like the idea of my MMORPGWTF character having a life of its own. A character that learns as I direct it in Puppet Mode and can then kick booty on its own when I need to eat dinner is totally cool. And because I taught it, it could fight in MY style, not some default wannabe style.

RTS games-Earth 2160

[Anonymous Coward]

Earth 2150 and 60 was pretty good, even if the voice acting was cheesy.

A Game Designer's Perspective

[vaporakula]

I'm a game designer, working on a next gen title, specifically focussing on AI. (yes, it's at e3, and yes, i'm exhausted as a result! excuse any incoherence...)

Here's my perspective on things.

First, you really need to focus on a genre. The gameplay desires and thus AI desires are completely divergent between, say, a Poker AI and an Racing game AI. There are some commanalities though, which I think lay in an interesting area to be studying...

The point is that your opponent should be human.

All too often, game AIs are purely robotic. You can't surprise them. You can't outwit them. They don't show cunning, they don't show fear, they don't interact with each other - they lack emotions.

Having a bot in a FPS is all well and good, but if at the first sound of a footstep behind it, it whips around with perfect precision and starts firing - it is next to useless. It won't satisfy a player - it doesn't give the same thrills as sneaking up behind a human player.

Study the behavior of humans in your chosen genre: see how they react, the ways they interact, the ways they try to play the game against each other - and use this as a foundation for building an AI.

I think there is a severe lack of emotional response in current game AI. Building an opponent with emotional model, that directly affects their chosen actions, would lead to some interesting results.

Of course, good AI is nothing without an appropriate means of displaying the results... for instance, back to the FPS - if you can't tell the difference between a bored opponent and an alert one, the player will never appreciate your efforts!

Re:A Game Designer's Perspective

[LordNightwalker]

Study the behavior of humans in your chosen genre: see how they react, the ways they interact, the ways they try to play the game against each other - and use this as a foundation for building an AI.

Depends on the kind of game, really. For a game of chess or poker, you're definitely right. For tournament style FPS your assumption still holds. However, when I play Quake 4, I expect the enemies to behave like believable strogg soldiers, while for a game like Unreal Tournament I want them to behave like believable human players sitting behind their monitors, using keyboard and mouse (and limited reflexes + aim accuracy).

Actually, that last bit is really high up on my list of AI features I want. Right under giving each NPC an agenda, an emotional state and restrictions imposed on him by the environment or society as portrayed in the game (like having him perform a daily routine that consists of food, sleep and work). But right under that on the same list, lies one of my biggest pet peeves with most AI: the AI can do things no human opponent can match. Their reflex rates are through the roof (FPS), they have perfect awareness of their visible part of the battlefield while we need to scroll around to see what's going on (RTS), they can handle multiple tasks "simultaneously" while we're restricted by the speed at which we can pan over the battlefield to get the relevant units in sight and move the mouse to select units and issue them orders (again RTS). Plus we get tired or miss some detail because it's small and doesn't stand out against the background, thus occasionaly we mess up; the AI is always at peak efficiency and sees everything, even a black player skin against a black background.

Realistic Animation

[grumbel]

This isn't AI in the normal sense, but neither is it a problem that you can solve by simply adding more frames of animation. Animation in video games still looks rather awful, you have some motion capture here, some hand animated there, but most of the time you see in the game is blending of all that stuff together and it never really fits, characters feet don't land on stairs but they always glide up the stairs like it would be normal ground, when running against a wall the feeds often continue to move and such. Flashback had some pretty good animations back in 1992, it was 2D and grid based which is why they could make the animation blend together in a static, but very seamless way, I would like to see that type of quality animation done in a dynamic fashion in a 3d environment. I don't care much about polygoncount and texture resolutions when characters still move the same ugly way like they back on the PS1. And in the end I think this is an AI issue, since the characters have to actually have some kind of intelligence to control their walking and make it look realistically, since no amount of pregenerated animation can ever handle the dynamic environment of a gameworld.

Re:Realistic Animation

[vaporakula]

Very true. Without sophisticated display, sophisticated algos just end up looking like junk...

Ultimately it doesn't matter if your character is thinking a million things, if it can only show you one.

Re:Realistic Animation

[Anonymous Coward]

Things have been done on this front. The term you are looking for is "inverse kinematics" The 3d Prince of Persia games make very good use of this. However I suspect that is can be very computationally intensive currently. I expect we will be much deeper in the uncanny valley in 3-5 years as processors get more powerful and the techniques and tools get more refined. As someone else noted it isn't really AI, but the best AI in the world isn't that great if it can't display the depth to the player.

For FPS-Type Games

[Undefined Parameter]

Allow me to preface my post with the following: I am not any of the following: game developer, computer scientist, psychologist.

That being said, in regards to First-Person Shooters, and other games of similar interactivity levels, I believe the limitations of AI to be tied more to the NPC (and player-controlled character's!) mobility options and to information theory than anything else.

In an example of the former, I give you the bipeds of Halo 2. Only two AI types will ever go prone -- the Grunt and the Jackal -- and even then, they immediately stand back up. In reality, all characters should have had the option to go prone and stay prone, so long as their physiology (and in-game psychology) allows for it. Yet due to this artificial limit on mobility options, the AI seems thin and very much false.

The latter I have a difficult time finding a good example of, so I'll refer to the problems encountered in WWII for the Allies (descibed in more length and detail, for those interested, by Neal Stephenson in his novel Cryptonomicon). Basically, it comes down to having the AI programmed such that the over-arching system can inform it selectively based on limitations set by the game designers as to what any given NPC could and could not detect and know. (As to what they could know, I mean whether an NPC would be programmed to act as if it were informed that there was [unit type X] in the area, or whether it would be a surprise to that NPC -- thus dictating how that NPC would be able to react; if known, that NPC might, for instance, open fire immediately, while it might not open fire at all if it were programmed such that the NPC did not know that there was [unit type X] in the area.)

Those are the two most common difficulties I see with AI in FPS-type games; too often will a completely new NPC (or the player's character, for that matter) jolt another NPC into immediate action. Too often, also, will an NPC be unable to react to an event or situation realistically due to improbable constraints on mobility options.

~UP

Keep in mind smart AI can make some games too hard

[TheLink]

Uh never forget: the games have to at least be somewhat _enjoyable_.

Enjoyable != super opponent.

If you really have a "smart" computer opponent, it will make many popular genres (FPS, hack and slash RPGs, RTS) much harder. It would be hard for most human players to defeat many computer opponents.

Making a really "smart" game opponent does not need major advances in AI (IMO the AI field has not made significant progress anyway - rather dismal lot). Given the narrow boundaries of most games, game makers can "hard code" in various assumptions, heuristics and behaviours which could be pretty _devastating_ against human or lesser opponents.

It's fairly trivial to make say Doom3 so much harder that you can't get past the first few levels, just by making the enemies behave more intelligently.

Same for those WoW stuff. If the enemies just ran away from you until they figure they have more than enough allies to overcome your group, you'd either be wiped out, or have an enemy-free (but no xp, no loot) journey to wherever you were heading ;).

When I see all this talk about game AI, I figure most people have no idea what they really want.

You want a human like opponent? Play against humans. While some human players are so good the average player would be wiped out, usually the average player is more likely to be playing the average player.

What would be interesting is if you had a game system where you could play against/with your pet or something. Dogs etc understand games. Rats learn...

AFAIK, AI researchers have only come up with AIs that are magnitudes less impressive than what you can get from your local petstore.

I mean these people are already so happy to get something about as smart as a cockroach. So far from being able to understanding stuff like "grass is to cow like plankton is to baleen whale" and then gradually deducing/associating other similar relationships. Most seem to be still stuck at the stage of: "Ah! the word grass tends to appear close to where the word cow does- this means they could be related".

Quite pathetic.

A smart AI would be able to guess some possible "absolutes" or new categories based on a number "relative vectors/relationships". A sudden realisation that a bunch of vectors/relationships can be arranged in a more "compact/efficient" way or even suddenly massively linked, could probably be considered a "flash of insight". An "unusual" but "true" arrangement could be what results from "getting the punchline".

But I'm no AI researcher so what do I know...

Minimum requirements...

[Jah-Wren Ryel]

What Would You Like to See from Game AI?

What I tell it to open the pod bay door, it had damn well better open the effing door!

bad news-Turing beats slashdot.

[Anonymous Coward]

Hmmm. An insightful and interesting post on slashdot. Are you sure you're human.

AI has been low priority for a decade

[Dracos]

And will continue to be low priority until the result of graphics and sound is indistinguishable from reality (another 5 to 10 years). Once all the eye- and ear-candy has been exhausted, then the game designers will have nothing left to do but enhance gameplay; one element of which is AI.

FPS and RTS games don't need highly sophisticated AI compared to RPG's. If developers wanted to (read: saw the business), sufficient AI from for FPS/RTS games could be achieved in 2-3 years.

An RPG needs AI capable of sustaining social interactions (at personal, econoomic, and political levels) between any 2 or more agents, in both speech and in writing. When a virtual population can sustain its society (then grow/adapt/react to stimuli) without the presence of players, then a true RPG is possible.

How cool would it be if you could walk through town in [insert fantasy franchise] and overhear the conversation between the butcher and the blacksmith's wife? Think of the ways quest hooks could be handled with such a mechanism. Your character is now the commander of legions, and it all started with eavesdropping in a tavern. Who hasn't played a D&D game that involved such a scenario?

(Psst... levels don't work at this scale... stop relying on them)

Eye candy sells games. Game play keeps the players interested. The more dynamic (and balanced) the game play, the better the game and the longer the player will (wait for it)... play. One player gets bored? Publisher just lost $10 per month. 10,000 players get pissed off because the mechanics get over-tweaked by a patch? Publisher lost $100,000 a month.

Re:AI has been low priority for a decade

[ScaryMonkey]

Um, have you played Oblivion yet? You can walk through towns, listen to people's converesations with each other, and start quests based on them...

Re:AI has been low priority for a decade

[mshurpik]

FPS and RTS games don't need highly sophisticated AI

Clueless...the "S" in RTS stands for "simulation."

How many trolls in this thread so far, a hundred?

Re:AI has been low priority for a decade

[Eideewt]

And to think I thought it stood for strategy all this time.

Re:AI has been low priority for a decade

[mshurpik]

Hahaha...good point. Well I guess I was defining RTS upwards. Strategy games are missing a lot of strategy when it comes to memorizing tech-trees. Currently I am working on learning Tech Wars 1.9d (war3) which is a small fraction of the overall war3 melee experience, and I'm well aware that once I learn it, the map will be useless (as I will be able to own anyone regardless of circumstance).

For example, air fleets in Tech Wars harass tanks, which absorb damage but ultimately fail. What stops air fleets? Nothing. Air hits fast and it hits hard. It's all memorization, not so much different from Super Mario Bros so long ago.

The real strength of RTS is simulating the interactions of different components, and I think this has been ignored. Most of the improvements of RTS "pathing" involve the behavior of single-unit movement, not communication.

Re:AI has been low priority for a decade

[Eideewt]

Huh. I had never approached RTS games from that angle. Maybe it's why I've never gotten into them.

Re:AI has been low priority for a decade

[Azathoth_lca]

I have, and the NPC conversations boil down into two categories:
- quest starting conversations that are 100% scripted with no AI involved
- random, AI initiated conversations. These generally go along the lines of:

NPC1: "The fighter's guild is looking for new recruits."
NPC2: "I heard that if someone wanted work, the fighter's guild would be a good place to start."

That's Not Quite What I Meant

[eldavojohn]

So what you're saying is, you'd prefer that Ragnaros emerge, kill all of the healers in the raid, then begin picking off the DPS one at a time as they die from a lack of heals? Because he keeps a log of what has happened in the past and he realizes that it's the healers keeping the others alive, so logically he would kill them first? Or, he sees the same 40 people as last week, and decides not to emerge at all, thus denying your guild the loot from another kill?

That sounds like fun to you?

Yes, that's precisely what I said ... to the letter.

What if Ragnaros changed his strategy (not so that he constantly targeted the healers--because there are rules of aggro in effect in WoW) but what if he followed a markov model built off of spell/counterspell? What if Ragnaros had a little bit of unpredictability built into him?

What if he slowly remembers which characters (yes, by name) are spec'ed and drinks resistance potions according to their specialty class of magic? This would stop people from camping and grind-looting him (I know camping isn't possible in instances but guilds do the same instance run over and over with the difficulty the same everytime).

I don't know who Ragnaros is but if he has the ability to move around, why don't they have him become more frantic or hostile if he's seen the same people and remembers them?

You'll noticed that I said you will find heuristics that make it too hard (like targeting healers first or not emerging) and you have to purposely dumb it down to meet your user's needs.

I said use your imagination and make things interesting--not figure out how to make NPCs afraid of conflict ... just more wise in their choice of engagement or more wise in their choice of moves.

That would be my idea of fun.

Re:That's Not Quite What I Meant

[ottothecow]

Well, in most games when you encounter someone, you kill them.

It doesnt make much sense for them to learn from you since they will be dead before they can apply that knowledge. If other characters get to apply their knowledge (unless we are talking about a hive mind here), then they are just cheating since they dont actually have the experiance. When you kill people outside the building, the unsuspecting guys inside shouldnt have already learned the tactics that you used to kill the outside people.

It could work for longer fights with single enemies but in the time frame of most games its just not plausible that people would have analyized your behavior and retrained their troops to combat you in such a short time.

Re:That's Not Quite What I Meant

[ranton]

It doesnt make much sense for them to learn from you since they will be dead before they can apply that knowledge

They do not necessarily need to learn from a specific set of players, but they can learn from what most players do against them. If Ragnoros is supposed to be a very powerful NPC, then it must have killed quite a few PCs along the way. That means he would know alot of the standard tactics, such as the standard Warriors aggro / Mages blast away / Healers heal.

I honestly do not understand how players in those MMORPGs can have fun doing the exact same thing over and over. If the NPCs would actually fight intelligently then PvE could actually be interesting.

--

Re:That's Not Quite What I Meant

[Meagermanx]

The players get to resurrect, so why shouldn't the monsters? What if the monsters only learned a tactic if they survived combat (run or fled), and then made it back to their base and could teach it to their teammates?

Re:That's Not Quite What I Meant

[txsvxn]

Somehow the thought of hearing the orcs in Everquest screaming "TRAIN TO ZONE!" as they high-tail it for the entrance to the zone comes to mind.

Re:That's Not Quite What I Meant

[Anonymous Coward]

You have to separate the concept of Ragnaros from the specific instance of Ragnaros you're fighting. (Think OOP). To train the Ragnaros concept up to an acceptable skill level you throw an instance of it against players again and again and then have it learn. (Some algorithm beyond my grasp). When the instance reaches the desired level you make it the model for the Ragnaros concept, and any new instances are spawned from it instead.

These new instances of Ragnaros can learn individually (adapt to the current opponents), but do not contribute their experience (data) to the concept of Ragnaros. Their previous collective experience can be seen as instinct instead of cheating. And the PC would probably not be the first thing Ragnaros has fought if he's such a hardass, so he has previous experience too.

That's what I'd do, but I'm just a first year CS student, so I have no idea what I'm talking about.

Re:That's Not Quite What I Meant

[smallguy78]

What amuses me about World of Warcraft is guilds that are so proud of themselves being the "server first" or even haven just killed an endgame boss in WoW. The AI in the game is so scripted and predictable, it really doesn't take a genius to work out what the boss does. Infact this is so true that CTRA (a mod for WoW) actually has a Boss Mod part that warns you when each part of the highly scripted fight comes. Admittedly Blackwing Lair fairly successfully breaks this mould, but on the whole the game has inredibly simplistic scripted AI that is geared purely towards drops than anything challenging, like you might find in say Half Life 2.

Re:That's Not Quite What I Meant

[lazyl]

You'll noticed that I said you will find heuristics that make it too hard (like targeting healers first or not emerging) and you have to purposely dumb it down to meet your user's needs.

Here's the trick. Balancing the encounter. You dismiss it in one sentence like it's easy, but it's by far the hardest part of the entire exercise; far harder than simply developing these ai heuristics you're suggesting.

These larger 40 player raid encounters have to be carfully balanced so that it's not too hard and not too easy and there's no silly 'ai exploits'. And it has to be fun. Trying to write a fancy ai algorithim that was properly balanced and fun would be a nightmare. There's all sorts of problems, including complaints from guilds who discover that Rag is harder for them then for some other guild because of some wierd ai learning code.

Ragnaros is the original end boss of WoW. That developers can't screw that encounter up. Scripting the fight is by far the most reliable for the developers to guarantee that the players have fun. Anything else is playing with fire.

Re:That's Not Quite What I Meant

[esper]

Scripting the fight is by far the most reliable for the developers to guarantee that the players have fun.

That depends on the players.

I read "scripting the fight" and I (having gotten bored of WoW's repetitive grinding and abandoned it at level 47, well before even hearing of Ragnaros) hear "making sure that if 100 groups of players attack him 5 times per group, that he will behave the same way in all 500 fights". BO-RING! If he follows the same script every time, then that implies that the players can also follow the same (presumably optimized, given enough replays of the battle) script to beat him every time - at which point you may as well remove the players from the equation and just have your computer run the script for you. That's supposed to be fun?

I'm with the original "Turing" poster. Give me NPCs who don't do the same thing every time. Make them smart. Make them unpredictable. Make them change from one time to the next - changed strategies, changed stats, changed drops, you name it. Eternal sameness is boring.

Re:That's Not Quite What I Meant

[lazyl]

Scripting the fight is by far the most reliable for the developers to guarantee that the players have fun.

That depends on the players.

Only partially. Yes, the players have to be open to the fact that the fight isn't the same every time and they can't just download the strategy guide. But making that system fun depends a lot more on the ability of the devs to design and balance the encounter. That I guarantee. That task is not trivial. You can't just hand everything over to some fancy AI algorithim and expect it to result in a fun experience.

he will behave the same way in all 500 fights". BO-RING! ... That's supposed to be fun?

It only get's boring when you starting beating him over and over. No argument there; I agree. However the fight is fun the first couple times you do it. And it's very fun if you don't download the strategy but instead try to develop it yourself (or if your guild is the first to try). If they switch to some advanced AI they run the risk of losing even that level of fun if it's not balanced properly. BTW, I think this concept of guilds running instances over and over again is stupid. These games need better endgame content that doesn't involve repeating the same fight 500 times. Even if the AI was adaptave that would still be boring.

Now, don't get me wrong. I think smarter, more adaptive enemies are a great idea. But at this point only in some places. Other less important encounters. And the devs will have to realize that they will be spending a *lot* of time (probably post release) tweaking those encounters. It will take a lot of experince at balancing those types of encounters before they should try using that sort of a system for a major boss fight.

I definitely agree that encounters with fancy adaptive AI would be better, but my point is that learning how to design them is significantly more work then you seem to think. Far more then the current, fully scripted encounters. That also makes them very risky. If they get it wrong then their game tanks. That's why I said that Ragnaros, the end boss of WoW, would not be a good place to experiment with this type of a system. They need to learn how to do it first.

Re:That's Not Quite What I Meant

[Anonymous Coward]

The war/mage/healer trio is the rock/paper/scissors of the game. Your AI strategy seems to be for Ragnaros to just break the rules and have rock beat scissors i.e. go straight for the healers.

What actually happens in the encounter is that he spawns Sons of Flame that do alot of damage to casters including healers.

Re:That's Not Quite What I Meant

[Anonymous Coward]

I like this post because I have thought about this a lot. MOBs would not have to know everything all the time. With each spawn they could adapt towards different behaviors as the frequency of a particular strategy was used until it is no longer easily viable. Players would have to use different strategies based on observing what the current strategy the MOB is using is. I would guess that for any given MOB type there would only need to be half a dozen strategies each with a set of tactical weights. This should apply to all MOBs (not just bosses) and perhaps groups of MOBs as well (different group strategies based on numbers of comrades versus numbers of opponents).

If the player doesn't see it, it doesn't matter

[robson]

As a game developer, here's something I've seen several times: A gifted AI programmer puts together a complex, ambitious system to govern AI behaviors. The system is elegant and probably involves lots of interesting problems.

However, because the focus was on the system rather than the output, the behaviors on-screen don't actually reflect any of that. A character could be dynamically evaluating its environment, weighing its tactical options, and interpreting a complex set of stimuli, but all the player sees is the character standing there.

It's my feeling that the problem is better approached from the other direction -- what are all the possible outputs defined by the design? Is this a shooter in which these enemies see, hear, talk to each other, shoot, evade, and retreat? Then come up with behaviors that utilize those building blocks in an interesting and engaging way.

When we eventually have characters capable of MUCH more complex behavior (when I'm talking to this character, his or her subtle facial movements suggest they're lying, etc.), I think elaborate procedural systems will become far more valuable. Until then, however, effort spent on the underlying systems (without regard for the outputs) seems misguided.

Re:If the player doesn't see it, it doesn't matter

[vaporakula]

So very true!

At one point, one of my coworkers was engineering a complex driving mechanism for the tanks in our game. They had gears, and everything - so that when they went uphill, they would slow down a bit, downshift, gain speed again (to a lower maximum) etc...

It was a shame the player was flying and had no chance to notice such detail!

Re:If the player doesn't see it, it doesn't matter

[Eideewt]

That's an example of a gifted programmer making a bad AI. A good AI isn't one that uses fancy algorithms under the hood. A good AI is one that (as you describe) makes effective use of the actions available to it. In my opinion, the mark of a good AI is that you don't know what it's going to do next. In so many games enemies just come at you with no variation. I'd rather have just a few that would take cover, try to draw me out, and sneak up behind me than have a horde of morons.

Re:If the player doesn't see it, it doesn't matter

[robson]

That's an example of a gifted programmer making a bad AI. A good AI isn't one that uses fancy algorithms under the hood. A good AI is one that (as you describe) makes effective use of the actions available to it. In my opinion, the mark of a good AI is that you don't know what it's going to do next. In so many games enemies just come at you with no variation. I'd rather have just a few that would take cover, try to draw me out, and sneak up behind me than have a horde of morons.

You know, that makes me think of Tremors. Yeah, the monster movie. That featured several excellent examples of tactic and counter-tactic between protagonists and antagonists. I think that's what we want in the long run -- we want antagonists that we can use a tactic to overcome, and as soon as that tactic becomes boring to execute, we want the antagonist to adapt and find a way around it, forcing us to use our creativity and ingenuity to in turn adapt to their adaptation.

Put another way, I've learned more about game AI from playing with my cat than from anywhere else. Why? Because my cat has zero tolerance for repetitive "AI" (in this case, a little jangly ball on a string, attached to a stick). If I just swing the thing around in front of my cat, I get zero response. Why? Because there's nothing fun about that. There's no challenge to catching prey that's right in front of my face. However, drag that jangly ball slowly along the floor and around the corner, now we're in business. But that doesn't stay interesting for long, either. I have to switch it out. Oh, look now it's hopping through the air. Now it's flying. Now it's crawling up the wall.

In short, though AI is usually considered an engineering problem, I think good AI is probably more of a design problem. The implementation involves a lot of engineering problems, but the stuff that keeps players engaged? Good old-fashioned balls to the wall design.

Artificial worlds need to feel self-sustaining

[Dr. Spork]

I know first hand that it's possible to write excellent combat AI as an ordinary (though long) nested list of conditionals. I'd call that hard-coded intelligence. This is the sort of thing I'm pretty good at (for example, the smart Irenicus in the Ascension mod for BG2, who's deadly without cheating). I have a feeling that most gamers are satisfied by this sort of AI if it's really done well, so that smart NPC's really fight smart.

But real AI has to involve some sort of learning, which is to say, letting game events "write" your behavior script. When would this be useful? The best example I can think of: Entirely stable environments that are "alive". Current games give you staged non-equilibrium situations that get triggered when your PC enters the scene. This sort of thing is just very obvious and unsatisfying if the goal is immersion. What good AI might do is this: before the game is released, the various separate settings might be populated with a bunch of artificial-life characters with specific motivations, needs, preferences, etc. (Maybe like the Sims, except more complex psychologically.) Then the game authors would let this initial system reach an equilibrium their big server. If they don't like the equilibrium that was produced, they tweak the initial AI and try again. Eventually, this will produce in a "natural" way something like a small, functioning village. When a PC enters the village, it will have been in an equilibrium which the actions of the PC will disrupt, almost certainly in unpredictable ways. That is how you give the player a true sense of freedom, like their actions really matter. Somebody like me might wonder: What would happen if I steathily killed the village miller, or gave him a gigantic horde of treasure, etc? That sort of scenario is impossible to play out in current games. And that sucks.

Now granted, writing Artificial Life that reaches an equilibrium similar to a real village, and still manages to react believably when a PC shows up may be a tall order, but I absolutely think it's a goal worth shooting for. For one thing, since many of the A-life interactions will happen in mutual isolation, the processing could be easily broken up into separate threads. Also, it's worth mentioning that this is not an all-or-nothing affair. If the equilibrium state produced at the end of several A-life generations is not exactly what you wanted, it's OK to slightly tweak the end result. The effect will still be much more convincing than the "village/dungeon/colony/factory eternally frozen in a moment until a PC triggers it."

One last bit: If you want to make AI characters seem realistic, maybe a good place to start is with Maslow's hierarchy of needs. Start there and build on that foundation. Everybody puts self-preservation ahead of other priorities, and so should AI. Ditto for all the other stuff on the hierarchy. So for example, if a fire I started destroys your hut, you will interrupt your routine to seek shelter, because this is your highest priority. For this, you may need to interact with other A-life, who might offer you a place to stay, financial help, or help with constructing a new hut - all for their own reasons, that depend on how much they like you, how likely they think you are to reciprocate, etc. If you saw me kill the miller, you will tell the other villagers, who may decide to ambush and capture you. How does a game produce that sort of behavior? Well, for one thing, planning requires some sort of awareness of expected consequences. But this should not be hard for a computer to do. It would guess that the PC would resist any attempt at capture, and it can (in the background) play out several "what if" scenarios compare their outcome to the goals of the villagers. This should show that a haphazard attack is a bad idea. Now it may look like I'm asking for a crazy amount of processing power. Maybe, but remember, we'll all have many CPU's to work with in the near future, so the ones that aren't running the game can be computing these "what if" scenarios. Also

Re:Artificial worlds need to feel self-sustaining

[Shai-kun]

Very very interesting post. Your ideas are rather straight-forward, and I think you've got a good grasp on what would make good AI in games. One thing you can be certain about: don't be afraid to use 'too much' processing power on this, because we'll have plenty of that even in the near future.

Academic interest in games

[Black Parrot]

The academic community has developed an intense interest in games beyond board games as a platform for AI research over the past few years. There are a couple of dedicated conferences (AIIDE,CIG) and special sessions at others (e.g., CEC).

DARPA was briefly interested in funding that kind of stuff a couple of years back, but someone suddenly decided not to go there. Academics are betting that they'll be back due to the obvious utility in training tools, as well as basic AI research.

3 Things

[MeanderingMind]

1) Learn - When someone plays any game these days with AI, it's only a matter of time before the weaknesses are found and exploited. Whether it's Super Smash Bros, Unreal, or WarCraft you always have a good idea what the AI is going to do, and how you can hurt it for doing that.

2) Vary - Super Smash Bros Melee had a major flaw, excellent as it was, in that every single character played the same way. Some were faster, some were slower, but they all acted exactly alike. Give us options like we had in Perfect Dark, simple as those were.

3) Suck - Computers should not be able to do with supreme accuracy anything a skilled player can't. The number of people who could accurately reflect, super block, and meteor smash recover in SSB:M was very small, but every level 9 computer AI in the game could constantly do those moves. Don't be that precise, it's annoying.

Lacking illogical behaviour

[4D6963]

Which are the most obvious intelligence deficiencies of current NPCs that need to be fixed?

Maybe surprisingly, one of their deficiencies is their lack of stupidity. I don't know if it could improve the game play at all, but making them more illogical would make them more human. AI's always seem to know what to do, although they don't do it perfectly, intendedly for not being too strong or unintendedly by bad design.

I'm not saying that we should have TKing AI's in online coop games for example, or having AI's standing on the deck of an aircraft carrier and shooting in the air or making bunny hops while waiting for an airplane to get in to spawn, but if we're looking for more human-like AI's, I think we should make them adopt a satistically defined illogical behaviour based either on statistics obtained in "laboratory" or by statistics obtain "on the fly" during the game. it would involve quatifying and qualifying the human players illogicol behaviour and being able to replicate it, I guess.

And while it's nice to have an AI be about at your level, it's annoying when an AI dynamically changes it's skill level during the game (as in Unreal Tournament for example), so I don't know what would be a solution to avoid having to do that, but I think it's something to absolutly avoid, simply because human players won't become any better as you start playing better during the game.

Oh and while I'm hear, one thing that is annoying is having an AI to cheat to match your level, simply because they don't "know" the techniques you know. Maybe the AI could learn from the techniques you use by "observing" to get better?

I think that the main reason why it's not fun to play against AI is that you can't measure yourself to an AI because it never is any close to an average human. I hardly see that ever changing, but I'd like to tho.

Randomness?

[Cybert8]

How about putting a hardware random number generators in PCs? All the consoles have them. The PC just uses a lousy LCG with the time as seed.

Re:Randomness?

[Surt]

Lots of games these days are using Mersenne Twister for random number generation, and that RNG has essentially no flaws that would allow you, as the gamer, to tell the difference between it and a hardware generator.

Check out Galactic Civilizations 2

[StillAnonymous]

Best AI I've seen in a strategy game was Galactic Civilizations 2. It adheres to all the same game rules the player does. It doesn't cheat unless you put it at one of the higher difficulty levels where the game tells you exactly what advantages the AI will get that you won't.

The computer opponents won't sniff out that you're the human and immediately gang up on you, you're treated as an equal. It's a very refreshing change from the usual "AI was an afterthought" you find in most games.

That's my idea of a good AI: Opponents that all play by the same rules I do, can put up a decent fight, and as a bonus, have some variety in the opponent play styles.

Grammar police.

[Anonymous Coward]

"but what we should we use it for?" We should we use it for English classes!

Interaction and higher thinking

[Dr. Eggman]

Better interaction and cooperation with other ai and the player. AI with goals that may make a traitor out of it is a despirate situation. AI that would give your position away because it doesn't want to die. AI that will bargain with its enemies to gain the upperhand, whether its rooting the player out or surviving that encounter. AI that is afraid to die; not just opposed to dieing. Perhaps its too much to ask for AI with a higher set of goal-oriented mental processes, but that's what I want to see.

Re:Interaction and higher thinking

[Garrett Fox]

Heh, appropriate name! Actually certain fantasy scenarios are good test cases for designing an AI. Say that a certain doctor sends a robotic minion to seize a bird from a girl in a prison cell. Robot: Give me the bird. Girl: No! Robot: Resistance is futile. Give me the bird. Girl: I wanna know why you want it! Robot: ...Data unavailable. Solve this problem and you're on your way to good AI. For another test case, think of the guards in Swamp Castle in "Monty Python and the Holy Grail"! (I used a "tailed_frog" as a test object for early AI characters to fetch.)

Half Life 1

[Ekhymosis]

I'm talking about the original HL1, not the rehashed one with fancier engine. That game had pretty badass AI that at times pissed me off because the marines appeared to work together and route me out. Also those fem-assassins were really good (maybe because their reaction timing was tweaked?). The rehashed version just showed how retarded the motions of the marines became compared to the original.

I also remember talking to one of my compsci teacher about this problem back in 2000. He and I were discussing the orders of magnitude of computing power, etc. it would take to achieve realistic AI into games. I think, with the way hardware and processing power is developing, this is only a matter of time. I hope within 10 years I can play a badass RPG and really feel like I'm part of the world, like the classic DnD meetings I had way back when.

Re:Half Life 1

[LD gspot]

I am sorry, but have you played half-life 2? do you play video games? have you seen a computer? do you know the best engine out there when you see it? do you know what AI stands for? have you ever seen squad tactics actually carried out with efficency? do you know what it means to flank someone? holy sweet moly man get yourself together.

Artificial CHARACTERS

[Anis_a]

What's really needed are NPC's with character, emotion, flaws, quirks, and the ability to grow/learn. Something to make you believe you're dealing with a real being, not some super-smart computer. De-emphasize the word intelligence in Artificial Intelligence, and you'll get a lot further towards your goal.

Watch the translation pitfalls

[Jonah Hex]

Let's travel back to the heady first days of Ultima Online, when they had a spawn system that approximated real world birth-life-reproduction-death cycles. Unfortunately the player population couldn't FIND any mobs to fight, the spawn was too realistic and couldn't cope with the artificial nature of players screwing up their routines. Old spawn system trashed, new spawn system based on timers goes into place, problem solved.

The way things work in the real world is not nessesarily a good basis for a game.
Jonah HEX

Re:Watch the translation pitfalls

[Hannah E. Davis]

I tried to design something like that for a friend's MUD -- unfortunately, though, it never went anywhere, and I crashed his computer a couple of times trying to get my scripts to work. My plan was to have a little ecosystem going on where the monster population would always be replenished if necessary from some other source. I'm not talking about "spawns", per se, merely a naturally reproducing population that is in some way safe from the players.

The idea comes from basic ecology: if you have a closed system with predators that can feasibly kill the prey faster than they can reproduce, the prey go extinct... unless they have somewhere to hide. If at least some of the prey can go into their dens at the first sign of danger, the population has a chance of stablizing, and similarly, if you always have fresh waves of immigration from some external "safe" place, you can get the same effect. You could even make that safe place visible to players to add realism -- for example, it could be at the top or bottom of a cliff that's too steep to scale, but the players might still see the baby monsters peering down at them and occasionally making their way to the unsafe areas en masse.

Another alternative is just to link the reproductive/spawn rate to the rate at which the players kill the monsters -- if they're killing 'em fast, each couple should spawn many baby monsters, but if the players are neglecting that area, the breeding pairs can be a lot less *ahem* prolific.

My point is that you can use real world ideas in games; you just have to do them right. Simplification is ok (and indeed necessary in many cases), but oversimplification (eg. ignoring current ecological theories on population control when designing a "realistic" spawn system) is often even worse than just having monsters on timers appearing in midair.

Re:Watch the translation pitfalls

[ShecoDu]

The demand for fighting monsters is too high, I personally don't like out-of-nowhere-spawns, where there's always a supply of enemies (and loot), it encourages camping and/or triggering the respawn (changing zones, guild wars).

How about breeding monsters?
How about alternate activities (interesting, challenging) while the monsters grow (why would you go on a puppy killing spree?)

Besides of that, I think content should be created by the user (kind of like second life, but creating usable/equipable items with bonuses and characteristics)

It's too difficult because people often want to play just to kill and kill some more.

I think I think too much and it doesn't work.

meh

Depends on the genre

[dosboot]

I find it interesting that some people think that games don't need better AI (in the computer science sense), it just needs to be more realistic, more human, more unpredictable. I agree 100% for some genres, like mmorpgs or FPS where making the AI difficult is already solvable. I think the AI for TBS games, and possibly RTS games, are suitable for academic discussion. These games haven't been 'solved' from an AI standpoint. Take civilization and heroes of might and magic games. Both are very old series, so you'd think the AI would be better by now. Yet even with huge production bonuses it is still a trivial matter to beat the AI once you know what you are doing. (For HOMM I'm only talking about the world AI. The combat AI, while far from perfect, is still typically good enough to be a fun opponent. This is because due to the fact that HOMM combat has a low possibility space.) It isn't like the AI programmers are in the dilemma "Well if we make the AI any harder it won't be any fun to play against". The problem is that they haven't figured out how to make good AI at these games. I think the obstacle is that: (I) You can't approach a modern TBS with tree searches. You are forced into using heurstics. This takes time because it requires humans becoming good at the game and figuring out what they've learned codifies into the AI. (II) A good AI for these games has to be able to think on many different orders of magnitude. Humans are good at this, AI's are not. What I mean by (II) is that in order to be good at the game you not only have micromanage the small scale units very well, but you also have be smart in how you play the broad based, global aspect of the game (as well as all the levels in between two). Now, it appears to be doable to make a good, heurstics based AI for each *individual* order of magnitude. However if you try to make an AI to handle it all at once it doesn't seem to work at all - a purely hierarchical AI is exploitable. To avoid exploits the small scale AI sometimes has to talk to and occasionally influence the higher up AI, and that's when the complexity of the code seems to be intractable. For example in civ, I might exploit a bad AI by declaring war, stay defensive until the AI throws himself on my fortified positions, then go on the offense and conquer his country against little resistance. On one hand the "higher up" AI wants to attack because I haven't offered any offensive threat (this is often the right reaction, e.g. if I'm not attacking him because I'm incapable of doing so then the AI should attack aggresively). The "lower level" AI doesn't want to attack because it can compare the strength of unit stacks and see that it will lose. You can't let the lower level AI make all the decisions though, because then I can exploit the AI by baiting him intentionally putting weak units out in the open and crush him with overwhelming force when he runs after the bait. I'm just an amateur programmer but this is the only time my thoughts on "game AI" ever became intersting from an academic standpoint.

Re:Depends on the genre

[mshurpik]

I think when the AI throws himself on your fortified positions, his units will send a high-casualty message, resulting in a switch to a long-term strategy.

What I want...

[QuantumPion]

I want more ninjas and pirates, and fewer robots and monkeys.

Re:What I want...

[Anonymous Coward]

what about a ninja robot?

Re:What I want...

[ockegheim]

Ah, then try some of this.

You *really* underestimate checkers

[edremy]

Let us take the simple game of tic-tac-toe (three by three) as our game space. Not too hard. Tic-tac-toe, checkers

Jonathan Schaffer also thought that checkers was simple. Years of effort building Chinook fixed that. He just announced after ~17 years of effort that three of the ~150 3-move ballot openings are draws, and I'm just amazed that he managed that. It's *not* a simple game- it only looks like it.

What I'd like to see in games with NPC's

[Filik]

- Protagonists that actually do stuff when the PC (player character) isn't around.

- Rumors spreading in a natural way

- Instead of totally scriptet plots, make a bunch of protagonists with different goals and abilities, who themselves have to figure out how to reach their goals.

- The ability to build "gang's" / followers in a natural way (for both PC's and NPC's).

- A living world where every NPC is unique, and constantly living their life.

- NPC's able to understand some rudimentary language, so they can get input both from PC's and other NPC's in both vocal and written form.

- NPC's have to be able to cooperate toward common goals with both other NPC's and other PC's.

-Filik

Wants and Needs

[KermodeBear]

I want an AI that will replace a human with believeable behavior (except for calling me a "gay fag", I can live without that, thanks). Something that not just learns, but creates ideas and acts on them. I need something that will entertain me. As long as it's fun I'm okay with that. That one things I hate most about most "AI" systems is that they cheat and perform at inhuman levels. Super Duper Never Miss accuracy. 200% production points. Knowing where all your troops are. That's not AI, that's just cheating and it sucks. For RPG-like games, I always thought that a "Needs" system similar to The Sims (Yeah, I know... But most people know what I'm referring to) would work to an extent. NPCs have needs and they have various ways of fulfilling them. I'm hungry... Go to the tavern? Shoot some game? Go home and tell my woman to make me a sandwich? Starve because I have no money, and beg for some from NPCs and The Player? Or should I steal? If I look at my environment, what's the best way for me to sate my hunger? This would be better than a simple scripted time-based action (like Oblivion, from what I have heard), easier to extend, and probably a better implementation overall. But what do I know. I'm just a consumer. Ultimately, keep it fun and you get my money.

Behaviours, not AI

[Anonymous Coward]

We don't need better AI, we need better "behaviours". Computers are inherently "better" than humans, when it comes to games. Well, in theory they can play games perfectly. They can even cheat if permitted. So really, the key to better AI in games is not "smarter" AI, but the right balance of "dumbed down" AI.

In the end, since games require dumb AI, what we need to refine is better BEHAVIOURS.

For example, a bot in a first person shooter could be programmed to find the shortest path to you and also shoot with 100% accuracy. So instead, we need some detailed and elaborate "behaviours" that mimick human tendencies. Humans can't see through walls. Humans can't always calculate the shortet path if missing key information (ie: if you haven't explored a location already). Humans have motivations, and judgement. Humans can be cautious, or bold. Humans have different thought patterns that influence decisions. There are so many parameters...

a common misconception

[Surt]

It's a common misperception that game AI developers are not up on the state of the art in AI research. Most game AI developers are well read, and read not just the books, but also various journals related to the topic. The problem is that there just isn't enough CPU time or memory available for games to make interesting use of any techniques that are not already being used. There are games built using long term development of neural nets, but there's not the computation speed needed to train the neural nets further during gameplay. Genetic algorithms are used similarly, but once you've shipped the game the opportunities to evolve further are basically dead ... by the time you learn anything from the player, the player has become bored. Expert systems are common as well, but inevitably have weaknesses that the player finds.

The core problem is quite simply that players are really smart, so much smarter than the AI we have now that a thousand fold increase in the available CPU time for AI wouldn't put a dent in meaningful improvement. Consider playing chess: the best minds in AI have applied themselves to this one game, for decades, and have only recently exceeded the capacity of human players. That's one game, with the most trivial state space you can imagine, with minutes (or hours) of CPU time on enormously parallel hardware available. Compare that to making a decision, in milliseconds, in an FPS or RTS with a state space millions of times larger than chess, and you'll begin to understand why game developers are resorting to methods other than AI for most of their projects, sticking to scripting and other simpler techniques, which deliver better performance than AI because they derive from human design.

The bottom line is that you just won't see a big jump in AI quality for games until the hardware catches up.

I've worked in eco-AI in academia

[CrankyOldBastard]

I have to say that I'm rather bemused by your posting:

As someone working in new (bio-inspired) AI research with an eye to applications in games, but within an academic setting, I often hear that game developers are not incorporating cutting-edge academic AI into their projects because it's too "risky" (they can't really predict how gamers would react), and because they don't see the point in it.

Ecologically inspired AI is in no way new. I suggest you look at Dana Eckhart's work (rutgers) or Dave Green's (Charles Sturt). If you've actually done much reading then you'll realise that the "ai" solutions are at best quasi-stable where "stable" solutions seem to be arbitrary, and rarely comprise parameter sets which are at all useful. One reason why commercial developers arn't terribly impressed by the eco-models is that few game companies want to invest in an IBM SP machine or a Connection Machine just to find parameter sets that are not fun!.

As a gamer, and as someone who cares what gamers think, I am often surprised by the sorry state of current commercial game AI - it has hardly moved since the 1980s. However, maybe the problem is that no-one really knows what we want from game AI.

No, it's pretty obvious what's wanted from game AI - something that will extend shelf life whilst reducing the amount of difficult to tune error prone tweaking of parameters. Unfortunately AI research is much more interested in the problems of machine learning and complexity theory, which have little if any connection to games marketing.

Academics keep coming up with innovative AI technologies, but what we should we use it for? What do you think? What sort of intelligent behavior would you like to see in games, but don't at present? Which are the most obvious intelligence deficiencies of current NPCs that need to be fixed?

The most likely positive applications of AI research would be machine learning to get NPCs to try not to die. The catch with that is that players will be very annoyed with NPCs that own them and keep getting better.

In particular, the AI methodologies that are used by the ecological modelling community could be used for world design profitably, but IMHO (as someone who has worked in this stuff) trying to get it to work in a fun game that's in the RPG or FPS genres are very very unlikely to work terribly well OR to impress the people who write the cheques if they are being used to try to provide strategic behaviour for the mobs.

The AI research should be used for such things as image pre-processing, experimental design and the other things they're being designed for. Trying to shoehorn these technologies into games just to be able to claim "cutting edge AI" will be an expensive lesson in what makes a game successfull - FUN. Regardless of what the Press may say, no game has ever been a success due to technology, rather the successes have all been that way because they were fun and people played them. AI != Fun except for us mathematical types.

A good example of what I'm taking about are the Creatures games - excellent applications of machine learning AI and sim-LIFE. Fun to some people, but never likely to outsell Doom or Quake.

AI has already been perfected...

[Winlin]

One word...Daikatana! /me feigns death.

GalCiv 2

[77Punker]

Play Galactic Civilizations 2 and you will understand what game AI should be.

STARCRAFT

[mshurpik]

I want RTS units that have morale, so they know when to stick, know when to run, know how to pick high-value targets, or switch targets (say if marines get rushed by firebats). It could be learnable, programmable, or hardwired, anything to make RTS less of a clickfest.

I want macro orders like "storm base," "harass," "defend," instead of the simple back-and-forth patrol that results in eventual slaughter. All units place too much priority on completing their movement, and right-clicking the soil is the worst mistake you can make in heavy combat.

Two games later, the units are exactly as dumb as they always were, just with more spells and a deeper techtree. But I wanted - expected - all this to happen in 1996 before Starcraft even came out!

Re:STARCRAFT

[Superfarstucker]

By your own admission there is a lot more than 'mindless clicking' occuring in a modern RTS and I would argue these are the functions that make the game competitive, if you want ai to do all this for you play civilization, as that is essentially what that would do to a rts. Further, games like civilization do have enormous tech trees to make up for the fact there really isn't any other substance to the game. Of course, the end result is not as cinematic. If battles did not become entirely deterministic from this change then you would simply see a push for more units and the management would become meta-management. i.e. macro-micro. Morale is a trivial thing to implement but it would tend to favor outcomes which are already likely (i.e. overwhelming force -> you're pwned) and hence doesn't add anything to the game. Unit pathing could always use improvement and it has improved, just look at w2->sc->w3, there is a big difference in the interface and pathing capabilities between these games

Re:STARCRAFT

[mshurpik]

By your own admission there is a lot more than 'mindless clicking' occuring in a modern RTS and I would argue these are the functions that make the game competitive, if you want ai to do all this for you play civilization, as that is essentially what that would do to a rts.

Yep, War3 has a lot more than mindless clicking but it mostly falls under the category of memorization...which can be interesting as long as you care...Civ is turn-based, I have a hard time with turn based games these days.

If battles did not become entirely deterministic from this change then you would simply see a push for more units and the management would become meta-management

Meta-management is what I'm looking for yes. I thought hard and decided AI would not make games deterministic, otherwise AI would be, by definition, perfect and I don't see this happening soon.

Morale is a trivial thing to implement but it would tend to favor outcomes which are already likely (i.e. overwhelming force -> you're pwned) and hence doesn't add anything to the game.

Exactly, AI would rule out the most obvious outcomes and make subtle play more interesting.

Unit pathing could always use improvement and it has improved, just look at w2->sc->w3

Granted, War3 has a number of improvements that make it compelling, but I tend to prefer the custom maps.

The never ending suicide run

[Deth Rot]

Stop being a tank and heal ME or the ones who need it. Stop crawling up my back while I am trying to agro to draw some away from a massive horde we can never take out unless we use gorrilla tatics. Be able to think for the group or at least yourself and run when the rest of the party is running away. Or at least only leave one martyr instead of everyone of the AI's because they cant seem to break off the attack from the enemy on a cliff shooting down onto us which we can not retaliate from.

Current gaming concepts do not support real AI

[dscowboy]

Game designers "don't see the point" of real AI in games? Not surprising. Let me summarize the basic fallacy of the whole gaming industry: They think games are like every other medium (movies, music, etc); a static experience produced in a studio and then handed down to consumers to passively enjoy. But the real potential of the gaming medium is giving players the ability to write their own experiences.

Because game designers insist on dictating the sequence of events, there isn't ROOM for real AI. You want this NPC to make meaningful, autonomous decisions about game events and communicate them to you? No wait, that would conflict with all the scripted crap we paid some actress to recite in a studio. Want to try winning that 'bad guy' over to your side? Sorry, but we spent sooo much time on this elaborate boss fight coming up... just sit back, trust us, we've got a GREAT idea for what you should do. And everyone else who plays this game. Every time they play it. Just keep going through the ONLY door that opens and keep killing EVERYTHING you see. You're going to feel like a real rebel cowboy hero after you perform the exact sequence of events we require of you and the 10 million others who played this game.

AI is essentially dynamic decision making. What's the point of dynamic decision making if everything that matters in the game is already pre-determined?

non-character AI

[MikeFM]

One of my favorite uses of AI in MUDs I made in the good old text days was using it for non-character intelligence such as making tools for generating quests in a way that responded to how the players behaved, weather patterns, etc. Current RPGs and online games are fricken boring. Quests and tasks are very canned and they don't really behave like a real living world in which magic is alive and good and evil clash.

One of my favorite toys was an oracle ball that when users used it would predict future events in the game. Nothing really fancier than predicting the stock market or anything else. Just analyze events as they happen and use statistics and AI to predict what is going to happen later. Put into a game this kind of thing can be a fun sort of psuedo-quest. Gamers can go to find the oracle in order to get tips as the best thing to do later in the game and those tips will vary in reliability.

Teamwork

[SanityInAnarchy]

I like my games easy. Or, even if they're hard, I like to completely overwhelm my enemies, even if it's a challenge to do so. So I don't really see the need for better enemy AI, though I'm sure many of the more competitive types will.

I want bots that work better together, especially on my team.

This makes it incredibly hard to adjust difficulty, of course. I remember hearing about a similar problem with the Lord of the Rings movies: in the Battle for Helm's Deep, the problem was in adjusting the relative intelligence/skill of the riders and of the orc army (or are they Uruk Hai?) -- too far in one direction, and the riders kind of splat against the wall of enemies. Too far the other way and they're untouchable, completely decimating the entire opposing army. So how do you balance it properly? In a movie, lots and lots of trial and error.

It would be incredibly hard to do this properly in a game, because it's far too easy to make the ally AI too powerful, thus making the player feel unimportant, and making the game entirely too easy -- just let your friends do all the work, keep yourself alive. Because of this, most games err to the other side: the AI of your allies are usually crippled even compared to your enemies, and if the enemies are ever hard, it's because of an unfair advantage.

Look at the Halo series. Consider why the first game begins with you as the sole surviving Spartan, and there will never be two Spartans in any Halo game. It really helps that the Chief is, plot-wise, the single most badass creature in the Universe, because on Easy, you'll always get far more kills than your allies unless you're extremely lazy or this is your first FPS ever, even though you could conceivably do very little and let your allies handle things. On Legendary, you cannot rely on your allies, because they'll be anihillated by the enemies every time -- if anything, you'll have to save them often, not the other way around, with the dim hope that sometime they'll prove useful, maybe to draw enemies around them so you can take a break, then chuck a grenade...

In other words, the fundamental design of the game is such that you're as much disconnected from your team as if they were overpowered and you could hide behind them and let them do all the work. Now, Halo 2 does a better job of this than most games, but it's still often hard to feel you're part of a team.

Maybe what I really want is multiplayer, perhaps a game that allows real co-op, local and/or networked/Internet, throughout an entire game, with an actual plot for why there's (say) two Spartans instead of just one. A game whose very plot requires that it be played co-operatively -- much like an MMO needs human players to be fun, even if there's no pvp, such a game would need at least a roommate to be fun.

Of course, for anyone to buy the game, there'd have to be a single-player, even if it put a bot in the second player slot.

Or, in other words, I want the co-operative system of Time Crisis, only with good friendly AI and some gameplay that's designed for truly co-operative play, not just "Oops, you died, more quarters please?"

As for the monsters, I'd actually prefer playing against obviously-robotic enemies. The fun of bots is finding and exploiting the weak points, both in their "physical" armor (pistol to the back of a Hunter in Halo) and their logic (sticky grenade on a Grunt below Heroic level, and he'll run back to his friends before he blows up.)

I've had this conversation before...

[PhoenixOne]

"I often hear that game developers are not incorporating cutting-edge academic AI into their projects because it's too "risky" (they can't really predict how gamers would react), and because they don't see the point in it."

One huge reason why cutting-edge academic AI isn't popular in today's games is QA. Debugging learning AI is a nightmare.

Another reason is not enough time, development or CPU. When you have only 10% of the CPU and 6 months, a FuSM may be the best you can do.

Add to this the fact that, for most players, predictable behavior is actual more fun. Being able to "outsmart" your AI opponents by recognizing patterns and taking advantage of them makes the player feel good about themselves (e.g. "If I toss a grenade over here, the AI will run away from it and I can shoot them. Damn I rule!"). ;)

Anyhow, there are some good game AI groups out there that love talking to academics. Check out gameai.com to start. Also pick up the three "AI Wisdom" books.

depends on the game, of course

[Khashishi]

I can point out two categories of games for which the goals of AI are dramatically different: strategy games and role playing games.

In strategy games, challenge is the most important aspect to be desired in AI. All other aspects like realism or flavor should be sacrificed if they weaken the playing strength of the AI. In a strategy game, the AI is just another challenger in a battle of wits, perhaps taking the place of another human. The AI needs to be perfectly ruthless and clever. Because, frankly, in most strategy games, a seasoned human can wipe the floor with AI even when the AI cheats heavily.

In role-playing games or action games, it's typically player versus world. The two sides are completely asymmetric, and the goal of the AI is not to beat down on the player (after all, this should be easy when it's a million to one), but to immerse the player. Different creatures should exhibit different behavior. NPCs should be interesting and interactive, having there own goals and not merely waiting on the human. Characters should remember past interactions and adjust their responses accordingly. Your hirelings should turn on you if you are acting against their interests.

Don't mix these two categories up. It's no fun if you are playing a shooter and the enemies all head-shot you as soon as you get into range, to provide difficulty. And it's stupid if the opposing civ makes peace with you because the AI likes your civics.

History and Dialogue

[Garrett Fox]

History: I've been thinking about this and related concepts from some experiments of my own. One way to build a more realistic setting includes large-scale "evolution" of societies. Create models of tribes that move around and interact, semi-randomly developing different cultural traits, eg. cultures that spend time in Steppe terrain sometimes get horses (Chocobos?) as domestic animals and are more likely to invent chariots and compound bows. Run the simulation for a while and you have a virtual guidebook to the world's cultures. Then, use that data combined with something like "Medieval Demographics Made Easy" to design and place some random cities and other structures, with data on the kinds of people present there. Then you can randomly generate specific locations within an area, specific quests etc. A really crude old demo of the kind of thing I mean is here, wrapping a dinky game around the world-viewer program.

Dialogue: From talking with fellow Loebner Prize Contest entrants, I get the impression that dialogue is very important to people's impression of a game world. In most RPGs, dialogue is severaly limited ("Welcome to our town!"), and even in "Morrowind" (haven't yet played "Oblivion") it all has to be written by a human. What if, instead, dialogue were generated by the AI, based on its own desires and experiences? If you used the "History" system above, that could also affect dialogue. "Long ago, it's said our people lived to the east..."

In general, what I want to see are characters that have their own motivations and are capable of learning and reasoning. This is more likely to be noticable among ally characters whom the hero can get to know over time, rather than enemies whose function is to die after a minute's struggle. One approach would involve lots of specialized "codelets" as in Hofstadter's Metacat program.

Ah, marketting bullshit

[Moraelin]

The fact is, a mouse allows perfectly good aiming to snipe a player circle strafing and ducking behind cover, and some of them have damn good reflexes. In my CS and UT days I had no problems turning and accurately headshotting a player crouched behind some crates in a split second, and I'm nowhere near the best player out there. Or I haven't heard of anyone having problems with the bots in Noone Lives Forever, which _do_ take cover. At any rate, the mouse was perfectly fast and accurate for that.

So I'm supposed to believe that only against bots is a mouse so slow that the bot has to stand still in the open to be hittable at all? Heh.

Want a more natural controller? How about a lightgun? Those existed for ages, and some of them (e.g., Namco's) had nearly pixel accuracy. They can be moved around with a flick of the wrist, just like the Wii controller. And if holding it like that had any kind of advantage, noone's keeping you from holding some lightguns by the barrel and using the button on the side to shoot.

So, please. The Wii controller may have its advantages, but (like anything else even vaguely computer- or console-related) it also has a bunch of shameless marketroids behind it. Tho'll cheerfully claim any absurdity if it helps sell their snake oil. Use your own brains when reading such patently absurd statements.

The limit was _never_ how fast or accurate the mouse is, but how fast and accurate the player is. And if the AI was dumbed down in a lot of games, I can also tell you why:

1. Because of player skill constraints. It's not that the mouse isn't fast enough, it's that the average gamer in your target market segment _isn't_ Thresh or Fatality. You can't throw 20 perfectly skilled opponents at them and expect the average player to track them all and handily dispatch them all. More importantly, the gameplay isn't a deathmatch where you just respawn when shot either: in most single-player FPS, if you got killed, it's game over and you need to reload. You don't want the casual gamer to _need_ to save before each corner, and reload 5 times before he manages to shoot before the bot headshots him. Because that's just no fun.

So even if you do program the bot to take cover, you still must make it easy enough to kill by a casual gamer. You have to just give the impression that the bot is playing well, not actually have it play well.

Contrary to uneducated belief, making a bot unpredictable and have 100% accurate aim takes no skill at all. The first is just a matter of using a random number generator, and the second is elementary maths: take the bot's position, take the target position, set the gun to point exactly along that vector. There you go: a deadly aim that never misses. That's not what's hard to program. But what you really want it to stand out of cover at regular intervals that the player can learn, and spend several seconds missing, so Joe Average has the time to aim and dispatch him.

2. Even more importantly, because of budget constraints. _The_ reason FPS exploded in the '90 wasn't because everyone wanted to play only that, it was because they were _cheap_ to produce. You could make a profit even if you sold less copies. You could license a 3D graphics engine, hack together a few levels and a couple of skins, and call it a game. Unlike RPGs and adventures which needed lots of scripting, animations, and occasionally AI, a FPS was cheap to produce precisely _because_ it didn't bother with those.

So noone was going to spend more money on one. Definitely not on ellaborate AI and character interactions, and not on the myriad of animations needed for the bots to take cover behind corners, crates, pillars, or upturn tables when nothing else is available.

3. Because of CPU power constraints. Let me tell you how those "minimum requirements" on a game box are born: the marketting guy, yours or the publisher's, comes and tells you: "studies say that 10% of people still have a 286, so our game must be 16 bit and run on a 12 MHz 286." That's an actual q

Re:Ah, marketting bullshit

[Lave]

Great points, but with one (tiny) flaw in your argument. Your talking about Mouse vs Wii, and the original poster was talking about wii versus joypad. The reason Halo was "Meh" on the PC was because it was designed with the slower joypad in mind. With the mouse it's a lot easier to have enemies above and below you, which isn't the case with a joypad.

People are excited by the Wii because the remote offers the ability to almost match, match or supass the speed, accuracy and crucially fun that a mouse and keyboard offers - whilst still allowing you to still sit on your sofa.

Thats a big deal to people - but obviously till we play the games (and maybe not till the 2nd generation games) we wont know how well they've done at this.

But as snake oil goes, at least this one is different - and thats always good.

Re:Ah, marketting bullshit

[Gulthek]

Wow. That's the longest rant against game consoles having superior gaming that I've never read.

Surprise me

[wolf369T]

I don't have to told the game producers what I would like to see in a AI character. Surprise me. This is an intellegince all about, after all.

What I would like to see?

[SmallFurryCreature]

First off, something very very simple. An AI that doesn't cheat in obvious ways. The recent most blatant example of AI cheating I saw was in Advance Wars : Dual Strike (DS game). It is a turnbased strategy game and you earn cash to spend on units based on the number of bases you control. 1000 per base. In one game the AI owned 8 bases and indeed the money counter for him showed an income of 8000. Nonetheless for several rounds he bought a tank + artillery + recon. A bill off about 15.000

A very obvious cheat. Even worse then those other strategy games that simply gave the AI an undisclosed amount or infinite money. (Command & Conquer?). In AW:DS I can see a bit of intel, adopt a strategy based on it (starve the enemy of the means to fight a war) and then get it ruined by a cheating AI. (Or maybe he was sponsored by axis of evil nations?)

The sad thing is that the AI is resoanably capable in the game without cheating. It usually buys the right units for the moment even the right counter units. In a map with a sea between my bases and it it will buy air transport and when you put up fighters it will increase its own defences. Better then most AI I seen.

What I would however like to see in all games is AI that stops doing the same thing over and over. Again AW:DS is a bit guilty here. It send 4 loaded airtransports to their death against the exact same air defences over several turns. In another game it keps producing lander crafts and putting units on them. It didn't move them into the killing field of my artillery but instead blocked up its own space with a dozen units. It should have bought battleships and pounded my artillery to scrap and open the beaches.

The same thing is true in FPS. Even in the best you can still shoot a guard and the guy next to him will eventually go back to business as usual usually standing guards in his mates exploded brains.

I don't want the AI so much to learn as to not forget. Splinter Cell does attempt this but the alert stages decrease again in a few minutes. Yeah right because a mysterious assasin is going to sneak all the way in, kill a guard and then just disappear?

We keep reading about FPS designers and telling us how they got some really cool AI that did things they didn't expect and then in the game itself we see the same old bullshit again and again.

Oh well, if I can't have learning remembring AI can I have one thing? Friendly AI that keeps the fuck out of my line of fire. Sure sure having the look before you shoot it one thing but if I am emptying a machine gun into an enemy I can sorta expect my mates not to fling themselves into the stream of bullets right?

Oblivion is a very bad example of this. It ain't AI. It is game design. Wich fucker at Oblivion decided to include friendly fire in a hack and slash? 99% of the time in a struggle you end up hitting your ally. WHY?

Forget artificial intelligence in game. How about some intelligence in game designers or is that like asking for journalistic ethics in slashdot editors?

Re:What I would like to see?

[BenjyD]

I don't think I've ever noticed the AI cheat in AW:DS. Are you sure it wasn't using some other method to get money, like combining units? I think some of the COs have cheaper unit costs or powers that give them more cash, as well. It would be a shame if it did cheat, as otherwise the AI is pretty good.

Opponent AI and Neutral AI

[Flyboy Connor]

There are basically two different kinds of AI: opponent AI and neutral AI. Opponent AI decides on fighting moves, neutral AI decides on everyday behaviour. Both are hard, but in a different way.

Opponent AI is hard because the AI is matched against the rational skills of a human player, who wants an opponent that does not cheat and that is of about the same level as he is, but that still poses a challenge. Well, that is nigh impossible to create because humans are inventive and game AI is not. It is not that it cannot be inventive, but if you augment game AI with inventiveness (i.e., learning skills), there is a huge risk that the AI will learn stupid behaviour. Game developers and publishers do not want to run this risk. Besides that, effective machine learning usually takes far too many learning trials to work in a game. And finally, learning AI severely increases the length of the testing process, which is already considered far too long. All in all, it will be a while before we see opponent AI that learns. Incidentally, machine learning techniques might be used to design good opponent AI offline, and I know a few game developers are looking into this.

Neutral AI is hard because we basically expect the imitation of human behaviour, or, we want AI that can beat the Turing test, at least within the confines of a game. Oblivion's Radiant AI is hailed as a great step in that direction, but when you play Oblivion you find it is still a long way off from what we want to see. For instance, in Oblivion NPCs have schedules - they get up, have breakfast, go to their shop, go to the tavern at night, etc. That is all nice, but what are NPCs doing when they are at home? They are staring at a wall. What are they doing when they are in a tavern? They are lifting a jug to their mouth, completely synchronised with the other tavern visitors. What are they doing when they are with their spouse at home? They greet them politely, make a remark about mud crabs they saw, say goodbye, and stare in another direction, until they repeat the same process a minute later. It is absolutely ridiculous - even more so because there has been made an attempt to make them more realistic. The funny thing is that almost two decades ago we already had NPCs with a schedule. Look at Ultima V. NPCs get up, have breakfast, go to work, stroll around town, go to the tavern, etc. And the net result was far more realistic than Oblivion provides. Why is that? Because game worlds were much simpler then. It was easy to create NPCs that behave realistically within the confines of that particular game world. The only thing they did not well was having a conversation - but, guess what, the conversation system of Ultima V is even more advanced than that of Oblivion. At least you can provide your own subjects of discussion. What we need to learn from this all is that the complexity of the game world determines how realistic NPCs can be with the current state of the art. And since game world complexity increases all the time, game developers need to be innovating all the time only to keep the level of the AI at the same quality as last year. And frankly, they are losing that battle.

Yes, academic AI provides solutions for some of the problems faced by game developers. Most of the game deveopers are not interested in academic AI, for several reasons. But a few of them are. My guess is that these will be the ones that bring us the next generation of game AI. And hopefully they'll become pretty rich in the process.

You can't have your cake and eat it

[NetRAVEN5000]

On the one hand, the game programmer can't have "the perfect AI" which can predict a player's every move and beats him every time - that's not much fun, it gets old real quick. Not that the AI should be stupid, but it shouldn't be unbeatable, and the first AIs the player sees should be weaker.

On the other hand, he can't just make a "dumb" AI that can't do anything realistic.

So, what would I like to see?

* AI teamwork where AIs can work together to formulate a game plan against me.

* AIs that project probabilities based on my previous behavior and other things (kinda like in "Terminator 3" where John Connor threatens to kill himself but Arnold knows there's a low probability he'll do it).

In other words, I want them to "notice" things. I want them to see that 64% of the time, when I throw the dagger, I hit dead-on; that before I enter a dark alley I look around with my flashlight; that I'll pistol-whip them if they come close; that I'm not the best sniper; that I'm not good at finding hiding spots. . .

. . .and then make predictions about what I will and won't do.

* AIs that make sacrifices. If they see that I'm sniping behind a boulder, and there's only a 12% chance they can kill me, they'll go for it if their team is in trouble.

Just a very badly designed game

[SmallFurryCreature]

There is an idiot at Betsheda and he is the designer. Oh and the person who includes a nipple in artwork in todays america.

Who on earth could have thought that including friendly fire was a smart move? Especially in a hack and slash. In a ranged combat game where you aim your weapon it is possible if you make sure no friendlies are about to make a headlong rush into your line of fire. But how do you stop yourselve from hitting a friendly in a melee when you can't control your swing and everyone dances about?

Please do note that this was a very real problem as well and massed melee combat was like walking into a grinder. The best armies were those who could control the amount of friendly stab. There is a reason all soldiers who fought side by side in thight formation did NOT wield large two handed swords desinged to be swung around.

No, this is very bad design. It is an attempt to introduce realism that is to real for the limited enviroment. If they wanted friendly fire they should have made sure you had far better control over who you hit. Yes those games where you can lob a firenuke at an enemy to fry him yet your mates inches away are unscathed are unrealistic. But they work and don't have endless rants on the net about how stupid it is when the guard that comes to your aid either buys the farm OR ends up attacking you because of an uncontrollable swing.

Bad Bethseda.

That doesn't necessarily mean AI, though

[Moraelin]

Most worlds are unsustainable and utterly unrealistic because... they've been designed that way. Not because some imperfect AI decided to put his fortress half-way into a hill, where people can walk over the wall. Most worlds are designed as just "levels" for the player to conquer, and no thought is given to what would that town do when the player isn't around.

But here's my take: it's not actually necessary for the NPCs to actually _do_ that stuff while the player isn't around. I don't really care if an NPC has been sleeping, sweeping his floor, and restocking his shop when I'm not around, or if he just got spawned in the right point for that time of day when I entered the level.

Games are just about creating an illusion, and an illusion is perfectly good for me. What's missing is any attempt to even bother about that illusion.

E.g., if a tribe is presented as having fought for grazing land (e.g., the Tauren in WoW), I want them to either be able to graze themselves or have cattle. I want their culture, their myths, etc, to reflect that. There are (not-so-subtle) differences between a hunter-gatherer society and an agriculture/animal-husbandry society, so a tribe that fought for grazing land for their cattle damn better not act like they're hunters-gatherers. I want their towns or settlements to have stables. I want their economy and their quests to reflect that. (E.g., while mystical and ancestor worship concernns are good and fine, surely a lot of their peasants would worry about more mundane stuff like cattle thieves or wolves or land conflicts, because they're a more immediate threat to their livelyhood.)

E.g., in Morrowind if a town has built city walls, then I want someone to sit and think what did the citizens try to solve with that. Building and maintaining a city wall was a _huge_ expense, so noone would do it just for decoration sake. So I want it to be a functional defense. Maybe it would actually need city gates too, not just some wide arches that either enemies or wild animals can just walk through. Maybe it would need ramparts, towers, or other suitable places for the defenders on that wall. (An undefended wall, anyone can just prop a ladder against and climb over.)

E.g., if the Imperials there built a fortress, what did they try to achieve with it? Maybe placing it on _top_ of a hill would server that goal better than placing it between two hills and half-way into one of them. If anyone can climb an easy slope to the top of the hill and rain arrows _downwards_ into the fortress, then wth purpose do those walls serve? Doubly so if continuing to walk over the hill gets one neatly over the walls and into the fortress.

Or what are the logistics of such a place? If you read for example Sun Tzu, armies used to cost a _lot_ and could even break an economy. Maybe they'd need to be regularly supplied, for example? There'd be a whole economy supplying a nearby fortress with wood, weapons, food, ore and coal if they have their own smith, etc. Which is why IRL even a small castrum often evolved into a major city. A whole economy basically evolved around it and in turn depended on it for protection. Important or wealthy merchants would eventually be granted or bought the right to have a house _in_ the fortress, either for the services they provided or just because they were willing to pay for it.

And so on. Basically, again, I don't care if the NPCs actually do anything when I'm not around. All I want is an illusion that those places _could_ realistically function as places for the NPCs to live in, and not just as levels for me to conquer.

Well that is easy

[SmallFurryCreature]

if(player.settings.geek == true) echo "no"; else echo "yeah right who are you trying to kid";

Don't let the player mega-powergame.

[WWWWolf]

"The computer will always retreat when I do this."

"If I go just a little bit here, it won't attack me!"

I've always thought that AI is bad when you can clearly figure out a pattern to exploit the AI's weaknesses. This especially in games like RPGs where you aren't supposed to pay attention to the weaknesses of game mechanics; you're supposed to sit back and do strategic moves and situational choices, not think of what flaws of the system to exploit.

At some point, we might grow tired of beating human players who are dead predictable - same goes for computer opponents too, though they're even more predictable.

Re:Artificial worlds need Distributed Intelligence

[Anonymous Coward]

So why not have a "distributed computing" like system setup for game AI? Use the whole community (with their permission of course) to help evolve the "village". Even when the client is not running the game, it could be helping with the computations with spare cycles. Let me choose if/when to participate in bettering the community. Get rid of my einstein (BOINC) and SETI. Let me use my cycles for something that I'll benefit from ... FUN!

I cannot believe I'm the first to think of this. Why haven't MMOs been using this idea? Distributed Intelligence.

Ehm Choke Point is very human

[SmallFurryCreature]

I was in a beta for a FPS from novalogic. It was fun until the kiddies got in but even at the best of times you had wonder if you were playing bots or real humans.

The game had vehicles including the very desirable choppers.

I like being an evil sniper. No I don't camp because that gets you killed (well only if there is a enemy player with an IQ above room temperature) so I move between shots. Or if the enemy is from america, between clips.

What happened?

Well spawns and runs to the airport to obviously catch a chopper. I don't bother to shoot at him running. I know my enemy and what I know is where he is going to be for a few seconds with his head held still. That is right. When he is in the chopper starting her up. Bam. Headshot.

Check my bearings but no enemy seems to be near. Probably to long a walk and you can't see me from a chopper.

Ooh an other victim just appeared. Yes he does get in the chopper, since obviously the sniper the shot the dead guy in the pilot seat has now gone away. Bam. Headshot.

Another arrives. Again shot on the pilot seat. It is now getting insane, the game has to drop the body out of the seat before a player can enter and that combined with the sometimes dodgy detection of you wanting to enter a vehicle means that now three people are trying to get in the chopper littered with corpses.

Bam, Bam, Bam.

This kept up untill I ran out of ammo.

I killed some people multiple times doing the exact same thing. Is this intelligence?

Not I was not spawn camping. The spawn area was out of my line of fire. I was simply denying an area wich is the role of a sniper.

Anyone of them could have gone on foot to root me out. None did.

Even in other games it was usually a handfull of players who did the sniping/counter sniping and the rest was just fodder.

The above is the choke point you describe but at least the AI has the excuse that it is a physical chokepoint. Nothing forced these players to take that chopper. There were boats and off course infantry can always just walk.

I regurly scored top rank and only got myself killed when I hopped aboard a chopper for a lift only to get my own team to kill me by doing something insane. Like flying a chopper straight in at a enemy base. Kamikazes had nothing on these guys for sheer desire to commit suicide.

SO if that is the state of human intelligence I take AI anyday. At least they can't vote.

RTS AI Based on Military Command Structure

[carcosa30]

I'd like to see an RTS AI based on military command structure.

The AI would be shaped like a tree.

At the top, there would be a strategic AI controlling resource management and making high-level strategic decisions like whether to rush, guns/butter, etc.

It would have a certain number of sub-generals, each of which might have a different sector of the map or some other division method.

Each general would control a certain number of companies. Now, these generals might be just AI objects, or they might be connected to command-vehicle objects within their areas of operations. The generals would make decisions within their AOO, keeping track of chokepoints, the large-scale enemy shape, etc.

Each company would be run by a lieutenant, which would control a certain number of platoons, and would conduct recon and other operatonal-level tasks.

You get the idea. Basically, AI would percolate down through a tree, from the general all the way down to the troops in the field who would have little more than basic insect intelligence (much like the American military.)

I think this type of AI could be made absolutely brutal.

Furthermore, if the command units did indeed map to troops and vehicles in the field, then command structure breakdown and operational paralysis would occur in a very realistic way...

FPS AI

[payndz]

In an FPS, I'd like to see (whether as campaign NPCs or multiplayer bots):

• AIs that actively hunt for me once they know I'm there, but do so within the limits of their 'senses', rather than just homing in on my co-ordinates.
• AIs that don't 'forget' I'm there just because I evaded them for 30 seconds, even though I slaughtered 20 of their friends right in front of them.
• AIs that communicate with each other - if I'm causing trouble in Area A of a map, the NPCs will call for help from Area E - and when I reach Area E, the NPCs there won't just still be mindlessly patrolling their default routes, but will have reacted to the alert.
• If I've sneakily killed a guard on patrol, his comrades get suspicious when they realise he's missing and react accordingly.
• Different kinds of bot behaviour in deathmatches. Rather than just following paths and shooting at anyone they see, have enemies who assess their surroundings. A 'charging berzerker' type would ignore the narrow gap overlooking a large hall, but a 'camping sniper' would set right up there.
• Bots who have the same knowledge as the player. If it's the first time I've ever played a level, it could be fun to have the bots also having to learn their way around rather than just running straight for the nearest power-up. Next time I play, the bots also 'remember' the layout.
• If I shoot one of a pair of guards in the head, then dammit, I want his buddy standing six feet away to notice!
• Enemies who use cover intelligently rather than, again, just following a path. As physics engines advance (vehicles move, walls collapse, etc) NPCs should take advantage of this. If a helicopter gets shot down and crashes in the middle of the street, it would be extremely cool if the bad guys registered the change in the environment and took cover behind the new obstacle.

The feeling of actively being hunted in a game, rather than just 'homed in upon', would be a hell of a lot more tense and involving than the standard fare at the moment. And fewer but smarter enemies would make for much better replay value than having large numbers of dullards who follow the same paths every game and then swarm you on sight.

So true!

[mwvdlee]

Years ago I developed a simple game prototype for a tennis/pong/sport style game. I had planned to do a number of basic AI tricks then dumb them down afterwards to make it fair. The thing is that the very first bit of code I put into the AI, which did nothing more than calculate direction to a moving target (ignorant of distance, time or other players blocking the path) completely blew me off the field; I had no chance of winning with the computer having even the simplest of AI's.
Imagine a basketball game where the AI could only calculate the perfect path towards the basket; that's all it would need to win the game every single time.

WoW AIs to play 'Dungeon Keeper' against players

[Anonymous Coward]

In World of Warcraft, most instances (dungeons, if you will) have some final boss living there with its minions.

There once was a game named Dungeon Keeper where instead of playing the hero conquering dungeons and monsters, player was the boss whose task was to keep his dungeon safe from heroes' attacks.

Now we have these wonderful MMORPGs with thousands of wannabe heroes roaming around but with static dungeons? Let instance boss AIs build and manage their dungeons with limited resources and dumb minions who he would punish occasionally for being so dumb.. :D

I hope you realize that...

[Moraelin]

I hope you realize that:

1. You don't actually need an eco-system to create the illusion you describe. If all the reproduction happens behind the scenes anyway, and the most a user ever witnesses is the baby badgers crawling out of a hole in the ground, then you don't actually need parent badgers in that hole. The hole itself can be just a monster generator (a concept that existed for two decades) which spawns badgers crawling out of it. The player is free to imagine that some parent badgers must exist in that hole, but they don't actually have to exist as entities in the game world.

It's even used in a lot of games. E.g., in City Of Heroes a lot of the enemies come out of some door or elevator, and the rescued victims disappear by running to the nearest door or elevator. Problem solved, and it offers enough answer to "where do they come from?" or "where do they go to?" to keep the suspension of disbelief going. As a player you're free to use your own imagination (e.g., maybe that NPC actually lives in that building) and that's a very powerful tool in maintaining suspension of disbelief.

The moral of the story is: never go for something horribly complicated when a simple illusion works just as well. Or to put it a lot less nicely, that's the difference between newbies and pros. The newbies get stuck trying to immitate the real world... badly.

2. Origin did try hard to make their idiotic "realism" ideas work, and they had the funding, the motivation and the programmers to do so. It tried really hard to have a "realistic" world, not only including the creature spawns, but also stuff like ore distribution, or trying to force players to create their own in-character justice to deal with griefers. They tried for _years_ and tweaked each such concept in dozens of different ways, even if it pissed off their player base. (In fact, much of UO's decline and losing the crown in the genre it created, can be traced back to its never giving a damn about what their players wanted, and being stuck on trying "realistic" stuff that noone wanted.)

Basically if any of those ideas had an obvious way to make it work, I do believe they'd have eventually stumbled upon it. Don't assume that they missed the blatantly obvious, and only you can tell them how it's done. Or not unless you're prepared to show a comparable game where that works.

The problem of simulating RL isn't trivial when you have 10 to 100 players per acre of virtual world, and they're all hell-bent on exterminating everything in sight. And more importantly, where some will deliberately try to break the game, just because they can.

E.g., players or whole guilds can decide to exterminate every single badger in your world. If those badgers actually have a simulated biology, it doesn't matter if you gave them holes to hide in: the players _will_ be willing to camp day and night in front of the hole, until the mama and papa badger get hungry and come out. So you want to put them on top of a cliff? So player mages will bombard that cliff with AOE (area of effect) spells like fireballs, and player rangers will rain arrows upon it, until they finally manage to exterminate the last breeders.

So you want to regulate reproduction rates? Players will only take it as a challenge in the attempt to ruin the world.

E.g., this is an actual UO example, Origin tried to have a self-regulating economy by limiting the amount of available ore. To prevent the market from being flooded by 10 times as many weapons and armours as players are, their idea was that there'd be a fixed quantity of iron in the world. New ore would only be spawned when existing iron items broke down or were otherwise destroyed. Sounds smart, right?

Well, as soon as the players understood this, enough of them started working hard to take metal items out of the game. They'd stash hundreds of swords and breastplates in their bank, or in the banks of characters they didn't even play actively, until mining and smithing did grind to a painful halt.

And just to qualify it again: there was no plan to corner the market to make a profit selling them later. The whole plan was to _destroy_ the economy, just because they could.

Re:I hope you realize that...

[Fuzzie Viking]

I understand (and agree with) just about all of your points. But no one ever thinks of the overall picture when something becomes unbalanced like that. There are other ways to address world issues when they get skewed, and they may have nothing to do with the problem at hand.

Using your UO metal example, were I to tackle that problem my answer would be to either

• drop a way for extremely smart talented players to "hit" the banks and take the stuff.
  (Admittedly this would create as many problems as it solves.)
• Have some NPC group start hitting the banks, as the payoff was way too high.

in either case the stuff would end up released into the world again and that problem is solved.

While it is unfortunate, a game really needs to be designed to minimize the impact of the 0.1% of the assholes that will play it.

Re:I hope you realize that...

[JollyFinn]

While it is unfortunate, a game really needs to be designed to minimize the impact of the 0.1% of the assholes that will play it.

No not really, just have scripts to make automatic fix ups, and announce that the economy&ecology while is algorithm based and stable and that you can screw it to the limits, it has realistic fixes that those who manage to hit those limits will not like. It might screw up the balance
In soviet MUD the ecology and economy screws YOU!

AI necessary for strategy, not FPS

[mcvos]

You really don't want artificial intelligence. You want the game to behave in a way that seem plausably intelligent but are actually beatably stupid.

Depends on the game, ofcourse. AI isn't all that interesting to FPSs and standard CRPGs, but I like strategy games, and there, even the best AI is just painfully stupid. I want strategy AI that doesn't need to cheat, get special bonuses or gang up on me in order to beat me. When I play Stars! against real humans, I have to work hard in order to outsmart them, which is a lot of fun against a really good opponent. But there's simply no turn-based strategy game that provides even a halfway decent challenge in a single player game.

Re:AI necessary for strategy, not FPS

[lurvdrum]

except chess. and Go.

Re:AI necessary for strategy, not FPS

[mcvos]

except chess. and Go.

I'll grant you chess, but go? I admit I've been out of it for a couple of years, but last time I checked, I could easily beat the best go program out there. Maybe it can beat me now, but I seriously doubt it could beat a good go player (which i'm not).

Informal poll

[The-Bus]

A while back there was a similar question posed in a gaming forum; rather, the question was "What bothers you most in a game?" and the answer was AI related. One of the main gripes in gaming period was bad AI on NPCs you need to protect. For example, in the military shooter, you need to escort one of the generals to safety. Or in the RPG, you need to escort the squire back to the castle. This was the most frustrating for a great number of people.

I would like to see more interaction with NPCs, especially in RPGs. I'm particularly miffed when playing (for example, Oblivion), and I can't simply ask someone to come over here, or stay put. Sure, if you're a complete stranger the NPC will probably ignore you, and that's fine, because it's realistic. But if you're well-known (for better or for worse) and you break into someone's house at 2AM and you're carrying a bloodied sword and your chainmail has some neural matter on it, I think the NPC may want to pay attention to what you have to say. In this way, more choices make the experience a bit better.

But there's a caveat. There's always a caveat. I've never much enjoyed playing the Tom Clancy games because they relied so much on co-combatant AI and the AI wasn't very good. On top of that, because many times you had to make decisions in the heat of battle, it was frustrating to have to pull up a submenu of eight different commands to give to your team. (At least this is how I remember it). Well, in GR:AW, you're limited to a matrix of four options: make your team aggressive or passive, move/attack or stop moving/attacking. The AI, in general, will figure out the rest, knowing if they need to be quiet or bust down a door, which targets to take out first, etc. In this way, a simpler control kept all your choices open and made your experience better. But that was only because the AI was pretty good.

What about sports?

[spyrochaete]

I'm surprised nobody has mentioned sports games yet. I used to be an avid fan of EA's NHL series but now I can't stomach it because of the AI. I understand that intuition is a big part of being an effective hockey player, but it doesn't take a rocket scientist to figure out that if you and a teammate are charging the net in a 2 on 1, the other player should try to get in scoring position. EA's NHL AI is absolutely atrocious - it does what you wish it would do perhaps 10% of the time when you're in the opponent's zone.

I'm not even sure AI is required to solve this problem. Seems to me it could be resolved with a couple of IF statements. Apparently that's too much effort for EA's annual crap factory.

Extrapolating from experience.

[master_p]

I would like computer AI-driven characters to learn and extrapolate from their experience. It would make a heck of a difference in sports games, FPS, even pacman (where ghosts learn to choose better routes than the ones preset).

Particularly obnoxious in sports titles

[ianscot]

Amen.

The 2004-5 generation of EA sports titles was absolutely obnoxious with that trait. It came enabled by default, and made for countless vein-bulging moments as your NBA team, out ahead by 30 points in the fourth quarter, suddenly couldn't get a layup to go down. Meanwhile the other team's oafish third-string center is inexplicably draining deep three-pointers to close the gap. Not only was it frustrating, it just plain didn't look or feel like the game it was meant to imitate.

EA games have extreme play balance problems in maybe two of three years, partly because sports in general are so familiar that we expect pacing and so on to feel just so and when they tweak some lame thing to give us new features things get out-of-whack. The answer, though, is not to cheat the player. If your football title is routinely resulting in scores of 100 to 86, you need to fix something more fundamental.

One thing I've always wanted.

[Anonymous Coward]

Make AI allies or enemies that you can talk to. Currently any NPC you ever interact with is essentially a menu (and usually an extremely small menu). NPCs will feel like they're people NOT when they fight me in an intelligent way and win, they'll seem like people when they talk like people.

First Person Shooters / War Simulations

[FunnyLookinHat]

I have yet to see a first person shooter with a war-setting be able to coordinate computer players with a human player efficiently. Games like Brothers in Arms and Ghost Recon - Advanced Warfighter require you to micromanage just about every single movement of your squad, when you would rather have them think for themselves (at least enough to shoot back at enemies).

Example: I was playing Ghost Recon and my "squad" was following me across and alley when we came under fire from a guy in plain sight. My squad just bunkered down in the open alley and didn't even try to save themselves. After they all died off I turned the corner, and easily took out the guy thinking to myself "What idiots would just sit there waiting for an order to shoot back when being killed off?"

Dawn of Skirmish: Dawn of War AI Mod

[sesshomaru]

There's a mod for Dawn of War that's just supposed to improve the AI. It's called Dawn of Skirmish (skirmish versus the computer is a single player mode in Dawn of War.

Currently, Dawn of War has some single player problems, I'll admit. Which is a pity, because graphically and thematically it is a great game. I haven't had a chance to play with the AI mod lately so I'm not sure how much improved it is from a previous build I played.

Conversations

[AlaricLuden]

The responses to this thread seem to be about tactical AI, but I thought of something quite different after reading the question. My main concern in game design is conversational AI, NPCs who can talk like real people. In most RPGs I've played, conversation seems to be limited to walking up to people and reading bits of scripted dialogue. If you're lucky, maybe you get a menu of three choices.

NPCs on the level I'd like to see are feasible: they have been done, by Emily Short. If a major publishing house produced a full-length RPG populated with Shortian NPCs, the result would be as significant as Doom, The Sims, or GTA.

Have AT, foul skeleton!

[Anonymous Coward]

clicky

GTA SA

[EcoPark]

One of the reasons why I like GTA SA so much is for its AI.(Or rather lack of it!) Well, I've liked all the GTA's except II and I suppose you can't really call it AI, but it sure makes for some hilarious moments. There's always some dude on the run from the cops running a old lady over with some dude walking past giving some quirky comment while looking at the body.Or one gang shooting wildly at another gang while a cop strolls past them without a care in the world. Or the peds that blindly dive right into your path of travel in front of your car just as you're about to pass them. It's all these flaws and really bad AI that just makes the game so much more enjoyable. Oh, you're talking about REAL AI? Sorry, dunno.

Teh best AI ever

[LD gspot]

The kind of AI improvement needed depends on the game. In FPS's, the best AI is for Half-Life 2 and I dare some one to counter that with some crappy Call of Duty or Medal of Honor AI comment (here is a hint, you are wrong). The AI uses complex squad tactics, takes cover behind obstcles based on the environment, provides cover, uses grenades in the fasion they were meant to be used in urban combat (to flush the player out of the players cover into the AI lines of fire), and attempts to flank the player and responds efficiently to players flank attempts. However, if one simply runs directly at an enemy with shotguns a'blazin, the AI does not adjust. It continues to take cover and fire in reactive burst. Now all FPS games have great AI and now they know that all they have to do to make it better is to make the AI hold the trigger down while being stupidly charged by the player. Problem solved. As for MMORPG stuff, you must have different kinds for different types of things. If there are wolves in the environment that are a dime a dozen and the player hacks throught them like a firey sword through wolf flesh...um...yeah, these wolves should not be able to constantly react and learn from every single wolf demise. However, if the wolves slowly evolve convergently to the players and adapt to their tactics (such as traveling more in numbers, or creating ambushes for players) then the players must still be on their toes to make sure they are not outwitted by a seemingly dumber beast. This forces the player to change the way the player fights in a realistic manner (I don't care if there are wizards and dragons, realism helps). If after a player slaughters 200 wolves in 70 encounters, and then dies because the player went for a seemingly easy kill and then got pwned on all sides by 20 wolves, that AI has succeeded. As for facing NPC characters rather than creatures, these characters should adapt mid battle to the players tactics. If the player does something extremely damaging to the AI, the first reaction by the NPC should be, "okay, this cannot happen again." Make the AI personality based too. If the NPC is a minion dedicated to die for it's evil leader, make it a berserker that will die with one strike. But if te NPC are supposed to be cunning, make them cunning. Make them counter spells and kill heroes. I want to lose to AI, and not for campy reasons. I want to lose because I need to level up and/or find another way to defeat him if I don't need to level up. If the player is expected to outsmart an NPC, the NPC must have a brain to outsmart. The AI I would like to see get better is for the Natural Selection mod of Half-Life 1 (if you haven't played it, check it out. You will thank me). If said game had AI that would win 50% of the time without boosts on the AI or nerfs on the players, then all other AI everywhere have no reason to not be perfect other than laziness and/or lack of imagination (i.e. Halo, most games for the PS2)

AI gripe list

[British]

1. Racing games: Most racing games seem to have the default AI where computer opponents drive PERFECTLY. They never make an accident. They take turns perfectly. They never drift. They drive as if they were almost bored(that is, if they had feelings). IIRC Gran Turismo 4 if you just leave your car standing in the middle of the road, they don't go around you. They plow right through. I only rarely saw AI cars in GT4 take a turn badly and wipe out in the grass.

2. GTA and True Crime: NYC

a. Cops are lemmings. They have an AI route of beelining for you. You can have some fun with this, like gaining ground, or making them drive off a cliff. Sadly, they don't know what to do in case of a car accident. I witnessed a cop car flip over, so the cop got out and stood along side it. Car becomes car-b-q and he perishes. i also once drove for 3 miles with a cop walking ON my car like it was nothing.

b. Triads in GT3 will attack you on sight(later in the game). They will do this in front of a cop. The cops look the other way. You do it, and suddenly you are in trouble.

c. Civilians. I think you can tweak the AI of the pedestrians in the PC version of the GTA games. Sadly, if there is gunfire and they are in a car, they will suddenly go into Carmageddon mode. This is 100 times worse in True Crime since a hit & run takes off half your health. makes gunfights REAL hard in a congested area.

d. "homies" in GTA: San Andreas. See c. Your 'homies' are almost magnetized to be in front of an about-to-be-speeding car and thus you can lose 3-4 of your brothers in arms right when the gang battles get started. You have to be your own Rambo.

Witht that being said, the "perp" AI in True Crime: NYC is actually a bit better. I've had perps hide(crouch) behind cars on occasion, and during a melee this one guy kept blocking over and over(later in the game). Slightly better AI than the one time I saw an ambulance try to drive through a house in SA. In True Crime, peds will back up & go forward several times in their car to get the heck out of harm's way.

Lack of Creative Thinking.

[the dark hero]

This is something I notice in some PCs as much as i do in NPCs. Most gamers would rather find the easiest way to acheive their goal rather than showing any play style. For instance in a game like Fable I see a lot of people go for the exact same skill set. While I was pleased with the amount of creativity I found in some players, I did not find many who played for fun rather than for power. Same goes for the Multiplayer in Halo 2. The "n00b combo" was very popular at on point (and may still be) and it was what most players would go for. Hardly a change in their strategies. I may as well have been playing bots. If i can see some real change in behaviour rather than static AI sequences that would greatly please me. The way it seems now is that the majority of players seek out X amount of challenge and Y amount of power to defeat the challenge (where Y = X + 50). Why can't we have an AI that will change it's strategy in order to avoid defeat at a disadvantage. That's what we as humans would do.

Cheating AIs: the annoyance thereof

[meringuoid]

I really don't care how it does that- extra resources, quicker unit building, whatever.

I don't mind the player having a handicap, but I don't like AI cheating. For instance, if it says in the manual

'At Ultra Hard level, all the computer players gain +1 of each resource from each square, and will be much more likely to trade resources and technologies with each other than with you'

- then fair enough. You know the rules, you know they're stacked against you, but you can plan for it. You can say to yourself 'That city will be producing about x shields and so can produce one spearman per y turns, and therefore my attacks must succeed m times, which will necessitate a strike force of n units...' Or some such. Only the hardcore will go so far as to compute the numbers, but anyone might try their hand at weakening the enemy by targeting their resource infrastructure. Mines, roads, farms, solar collecters, whatever.

If, however, the AI is cheating - pulling free units out of its ass whenever necessary, for instance - then you can't make plans of this kind, and the game loses a lot of interest. Why target the AI's resources when for all you know it'll just conjure up extra units anyway? Might as well just send in the horde and butcher them all, brutal and ignorant though that strategy is.

But that wasn't what really pissed me off. Nope. What pissed me off was when, in Civ 2, I'd exchange world maps with an AI player I knew only had triremes (which sink if sent into open ocean) and see their shipping lanes, where their triremes had sailed across the ocean in a perfectly straight line towards land the AI had clearly already known was there...

Warning Forever: A Learning AI

[kthejoker]

Has anyone here played Warning Forever? It is a fairly simple game, you are a spaceship with some guns, and there is a boss ship you must defeat.

But the neat thing about it is that after you beat it, it learns from your patterns and then a new ship is spawned with ways to flummox you. Does the bosses' shock wave weapon seem to kill you most often? The next boss will have twice as many shock wave weapons. Like to stay closer to the boss to avoid the targeting? It grows slicer arms which slice out to destroy you. And so on. And ultimately the boss created knows all your tricks, and you have to either adapt to new tricks or die because you have no tricks left.

All in all, this is a great game, and it really challenges you to play an old stand-by (Asteroids/Galaga) in a new and different way. I think this is definitely one AI model to consider ...

Re:Warning Forever: A Learning AI

[planetoid]

I always suspected the trick to how the boss ship evolves is not based on principles from the AI field but merely the program gathering statistics, for lack of a better word, and using them against you. Of course, the program is closed-source, so as far as I know he could have a Prolog interpreter hidden under there ;-)

Well, that's exactly the thing

[Moraelin]

Well, that's exactly the thing: it creates more problems than it solves.

If I'm allowed an awfully long metaphor: There's one thing you get to learn early in Go (and presumably most other game boards) and that is not to lose good pieces trying to save bad pieces. At some point you have to look at a group of your pieces on the board and recognize that they're dead. That's it. It may take another 1 or 100 moves to actually get them encircled and removed from the board, but that's it, there's no way to save them. The newbie mistake, leading to some spectacular defeats, is to try to anyway, and lose more "good pieces" trying to save those who can't possibly be saved. The only good course of action is to abandon them and resign to the idea that they're already lost.

The same applies here IMHO: adding even more bad (if "realistic") ideas to save previous bad ideas, is IMHO the wrong way to go about it. At some point you have to look at it and admit that it was a dead end, and prolonging the suffering with even more bad ideas won't save it.

E.g., your solutions not only create more problems than they solve (as you yourself have correctly noted), but also don't solve the real problem. If stashing the iron swords in the banks won't break the economy, then players will stash them somewhere else and still break the economy. E.g., create 20 mule characters each, which exist only to store a few dozen swords each in their backpacks.

What next? Have NPCs rob random characters which aren't even logged on? Surely that's a way to piss off most of the player base, who didn't take part in the scam in the first place.

It can easily become an arm race, or an equivalent of Go's "ladder" formation, that leads nowhere. You end up piling even more "fixes" to the problems created by the previous "fix", until it breaks or it loses most players' interest.

So while I'll aggree that "a game really needs to be designed to minimize the impact of the 0.1% of the assholes that will play it", as you aptly put it, I'll disaggree that such an arms race is the way. Sometimes the right solution is to discard a previous bad idea, not to keep "fixing" the wrong "fix".

In UO's case, the solution was to simply get rid of that iron-tracking silliness. It removed both the problem _and_ the consequences of the assholes' actions in one fell swoop. The game works perfectly well without it, and a hundred or so other MMOs never needed it either. There are other, better, ways of regulating an economy, or even better: of letting it regulate itself.

Re:Well, that's exactly the thing

[Fuzzie Viking]

I agree with you, but the problem is that you cannot design a system that is everything to everyone. If you design something that is exploit proof and with no way to abuse, I would dare say you have made yourself an extremely boring game. It is all about trade offs. Granted I am so far from an expert I should shut up now, but in for a penny...

For a large percentage of cases I think you are correct. The flaw is in the design. But even the best of designs will have paths of abuse. And if the design is as correct as you can get it (or want to get it) then an arms race is a perfectly acceptable solution. All you have to do is manage the reward / punishment scale. Then again I'm from the camp that people who live to exploit like that should simply be thrown out.

BTW, What next? Have NPCs rob random characters which aren't even logged on?
Depends on the type of offense we are talking about. If it is an easily provable example? (Lets say its the UO example and 10 players hold 90% of the iron. - Again I think this is a bad example as a better design was the way to go.) Hell I'd do it to their face. Nothing like spawning "unkillable" thieves who completely strip a character down. But I wouldn't just take the iron. I would take *everything* they own. Being the vindictive type I would probably either boot them afterwards or make them start from scratch. Probably boot'em though.

AI to suit the game

[Anonymous Coward]

With a game that has linear progression it's better to have scripted actions than learned action. Even in a vast space like WoW, you still have linear progression of dungeons. You start at the entrance and end at the main boss, typically. In this type of gameplay, scripted events are much better for storytelling. For FPS there little that an AI can do that can't already be scripted. To the NPCs in these games all that matters is the current state of the player. Nothing from the environment really matters except as an obsticle between them and the player.

The best game genres for AI would be more of the sandbox/social type where the AI not only interacts with the player with they also interact with each other. evolutionary/sim games would be the best place to see AI in action. It would be like a Sim City game in progress, but instead of the player playing god, he/she would be playing as a citizen.

What AI will never become is a tool to prevent players from exploiting the behavior of NPCs, especially since games are designed so that an enemy has a weakness or several weaknesses for the player to take advantage of.

Comments from an RTS programmer

[The Optimizer]

I worked on the "Age of" games through Mythology, though not on the AI, and perhaps I can shed some light on what you were seeing. I know much less about the system used in Mythology, so things that were true for Age1 and Age2 may not be so for it.

In each of those games, there were actually multiple 'AI systems' that comprised the "AI". This allowed for separation of short term goals from long term strategy and computer player personality, as well as for exploration, expansion and reaction to events such as being attacked.

In all of the Age games, AI players could be put on your team as allies, or grouped in teams against other teams of AI players. You could reveal the map, and sit back and watch them duke it out against each other.

The AI players always had an advantage over humans in terms of issuing order to units. In one 1/20 of a second turn, it could locate every unit that was idle, determine a task for it, and issue the task command.

The AI players 'Age of' games I worked on did not ever cheat by creating resources out of thin air or get different gathering or building speeds. You could starve a computer opponent out of a resource. On the Harderst difficulty level, all AI players would get bonus resources at the start of the game (and possibly at the start of each age).

Multiple AI players would work together in various ways; coordinating attacks, trading with each other, and even acting as a 'feeder' player.. going all economy and then tributing other AI players to boost them up quickly.

In Age of Empires 1 & 2, the AI players did not look at your list of units of and buildings to determine what you had, where you were, and how to counter it. Same for the locations of resources. An AI Player had to see it first, before it was allowed to 'know' about and act on it. In the other big RTS games, this is defiantly not true as they just look at the global object list. For this reason, the 'Age of' AI sent out units specifically to scout the world map. Once noticed, the info would be shared with allied AI players.

You could exploit this: if you were on an island, and never let any units go close enough to the water to be seen by AI ships, the AI players would never send any units to attack you, as it did not 'know' where any of your stuff was. Another way was to build in an area that the AI's had previously explored but were unlikely to wander back into. The line-of-sight distances in Age2 were increased to help keep this from handicapping the AI's too badly.

One area where 'cheating' was necessary was path finding. If you took a unit, and told it to move to an unexplored location, a valid path would have to be computed, requiring that it know about unexplored obstacles. This could be exploited in some ways. If a wonder was surrounded by walls, yet there was hidden break in the wall somewhere, just pathing a unit from outside to the wonder, would make the unit take the break, even if it was in an unexplored area as it was moving along a validly computed path.

I felt that we put much, *much* more work into making the 'Age of' AI's not cheat than most RTS games. What I learned from that though was that the public usually notice enough to care. A non cheating, general purpose AI often will not seem as 'intelligent' as an AI that cheats and has highly scripted or triggered events. The upside is that it allows the AI to play reasonably well on randomly generated maps; something most RTS games don't even bother to try.

Learning AI is the ultimate in MMO

[meregistered]

This post is likely to never be read by the poster or others.
However I think its an interesting note (and again I obviously haven't read all the other 220 posts at this point).

IDEAL AI:
AI that learns from every player encountered in the same way flickr photo categories are selected.
This should be possible to be accomplished by using a basic dynamic structure for the AI which would know about the abilities of each NPC it runs by spawning off a version of its self for each NPC with those capability attributes. The next step could be made by utilizing a genetic algorithm (point based system that allows effective AI routines[usually specific choices made] to continue and become increasingly more likely to reproduce and less effective AI routines to be less likely to be reproduced ). Finally the addition of random reaction elements to seed the genetic algorithm and to be produced by the genetic algorithm along with cross breeding of more affective routines.

JUSTIFICATION:
Second Life, Flickr, and Myspace are all user driven and therefore very effective at what they do. When thousands (or millions) of users determine the direction a program can take the effectiveness of that program increases. MMO games would be wise to take advantage of that model.

DRAWBACKS (potential):
Some interface that allows users to provide feedback regarding the AI they are fighting that makes preferences clear. It would be a little lame to have the program say [in a pleasent female voice] "How did you like your combat experience, please rate it on a scale from 1 to 10".
If there was some clear way to rate the user experience without having to bug the user this method would become extremely effective. Could be as simple as how frequently the user comes back to that spot... pro-rated on how easy it was for the user to defeat it's oponents. Might need to instance combat for each player in order to make it work...

-ME®

A Young Korean Woman who...

[Anonymous Coward]

likes me, knows football statistics, can bring me a beer and ...um, is uh..., well, you know, correct in all aspects!

Lots of firepower...

[PhYrE2k2]

Jim Hope: Since Christmas is coming soon, I thought we could talk about our
                    favorite toys. Milhouse, what have you got there?
Milhouse: My Busy Box! It's got everything! [turning steering wheel] Vroom!
                    Vroom, vroom! [dialing phone] I'm calling Daddy!
Jim Hope: Good for you, not being bound by the recommended age.
Milhouse: What are you talking about? [reads "ages 2-4"] Oh, jeez!
Jim Hope: How 'bout the rest of you? What do you like about those toys of
                    yours?
Sherri: [holding doll] They're special.
Nelson: They're challenging. [cranks jack-in-the-box for a while]

Jim Hope: Very good. Now I want you all to imagine the perfect toy. What
                    would it be like?
Terri: [holding stuffed animal] It should be soft and cuddly.
Bart: Yeah, with lots of firepower.
Milhouse: Its eyes should be telescopes! No, periscopes! No, microscopes!
                    Can you come back to me?
Nelson: It should be full of surprises.
Milhouse: It should never stop dancing.
Martin: It should need accessories.

Understanding Human Behavior

[az_zel]

I think that one of the things which I would really like to see is the ability of the game to simulate a person's behavior. This would include such things as natural language processing and generation, emotion, goal-oriented behavior, etc. One of the most deficient part in games is that it is incredibly obvious that NPC's are NPC's. They are really just talking heads with a little dialogue to spout at the first sign that someone is paying attention. I guess what I'm really talking about is believable interaction between computer characters and the player. Of course, this is an "AI Complete" problem :-/.

What *I* would like to see..

[revlayle]

... is a re-designed Ian that will QUIT SHOOTING ME IN THE BACK!!!!

An example of _good_ AI response...

[meringuoid]

... Sorry to reply twice, but you know how it is...

Rereading that rant I wrote earlier about rotten AI behaviour, I remembered one case in which I was unpleasantly surprised by a very, very good AI action. I've since been told that this was actually a scripted encounter, but it still stands as an example of the way enemies should behave. Even if it was rigged, it had to say 'if player tries this sort of thing, respond like this...'

It's the original Half-Life. I'm trying to escape from Black Mesa in the face of a rather large military task-force dedicated to trying to kill me.

Around a corner I come, out of a narrow corridor into the open, and there are _loads_ of these soldiers, some dug in, some standing around. Yikes. Don't fancy this much. So, I play it a bit cunning: I spray the area with random fire, enough to alert all of them to my presence, and then duck back round the corner and select a nice heavy weapon. The plan: bloody stupid AI will charge straight into the corridor and I can mow them down in a narrow, confined space.

So, I'm sitting there waiting for the enemy to appear. What do I get instead?

* clink! clink! clink! *

... and onto the floor in front of me there rolls a grenade. Perfect throw, spot on target.

Oh bugger.

Enough Spatial Reasoning.....Add Social Reasoning

[Boxcarwilli]

How about some "social reasoning" instead of purley "spatial reasoning". Too much is focused on what we as players see/hear vs. what may or may not happen based on communications with NPC's. If I can get a tip that beating the big dragon is as simple as shining light from a diamond into his eyes vs. beating him death then great, give me the opportunity to intellegently dig up that information from an NPC.

Not simply conversation trees mind you, but use "mood" engines, "memory" engines (as in I hurt an NPC's friend and the NPC isnt likely to help me out), persuasion, diplomacy, bluff and any other SOCIAL metagames/engines that you can think of.

Also, not necessarily AI on this next piece, but If a branching system does exist, make each branch have a significant outcome vs. all brances leading back to the same tree. Example, if my team is given notice that the right fork in the road leads to a man who gives me a potion to make the dragon talk and tell me where other dragons can be found, vs. the left fork ending with killing the dragon and getting his loot.

No "S" in RTS AI

[bender647]

Playing against most RTS AIs right now is like a matador leading a bull.

Put some defensive structures up where you want the AI to attack. He'll spend his entire economy banging his army against it, even if it doesn't guard anything useful.

Lure small groups of AI armies to their death. They'll never walk away from a fight, no matter how mismatched.

On the flip side, the AI never seems to have a SENSE for when its winning. If it whips me to within a flicker of my death, but has to weaken its economy below some limit to do so, it'll wait to recover before coming at me again. By then, I've recovered too. Can't it ever take a chance?

The current design strategy to make AIs competitive seems to be to let them cheat economically. Which typically means the player needs to turtle until resources run dry. Pretty boring after you've done it a few times.

Nope

[SmallFurryCreature]

It was the fat sailor and he wasn't listed as having any skills.

It was very blatant because I had him pinned down but was unable to get enough units in fast enough to wipe him completly out in a turn. It then becomes very apparent when he produces way more units then he has cash for turn after turn.

Even if you ad all the costs reductions in the game you don't reach over 50% reduction (he had to pay for repairs as well). I see no way to generate 1 tank 1 artillery 1 something else + repair from 8000. Perhaps it was a bug as I never noticed it obviously cheating before but usually I end the game differently so perhaps I just don't notice.

While we're mentioning it...

[fluppeteer]

I've had thoughts for a while of producing a mud-type game with an entirely simulated AI. Not allowing language helps a lot, which may make multi-player interaction a problem, but it would be nice to pass a Turing test in a silent environment. It would only work in the right kind of game, obviously (so this has been an interesting discussion thread to read).

All I really had to contribute was: I thoroughly recommend "Inferno", a short film from 2001 about a FPS based on a learning AI, with the film told from the point of view of one of the AIs (who falls in love with the player).

For a comedy, it's actually quite informative about how this kind of thing could turn out if the AI was intelligent enough.

use the `net

[Incy]

So that the AI engine could learn from experiences on each players machine to get better. I.E. I want the AI to learn from me, and upload those results to a server, where it gets merged with data from other players.. and then redistributed back down. Oh okay.. maybe it won't be "fun" for the buyers of the game.. but it would be *neat*..:)