Complex Mob AI: Sweet or Sour?
There’s a great thread over at Quarter to Three called AI in MMORPG Games, in which the community discusses mob AI and how many players desire more intelligent encounters in MMOs. Lum made some great points on the second page, which can basically be summarized as: testing and data suggest that players don’t really want complicated artificial intelligence. Lum’s post and my thoughts after the fold…
Lum says:
“Gamers say they want better AI in MMOs.Data mined statistics show that the most popular monsters in any MMO are the ones with the most brain dead AI, preferably bugged into virtual unconsciousness.
CPU load is another issue, but not as much. Many MMOs have very intricate AI encounters – but players tend to hate them because good AI adds a concurrent risk. While saying “hey, it’d be cool if guards responded when their buddies are attacked” while discussing the subject, it’s entirely different when you attack a monster and are suddenly fighting off 10.
It’s a classic game design challenge – make the player *feel* as though they have completed a difficult challenge, while keeping it easy enough for them to actually accomplish.”
Lum hit the nail on the head here. There still are technological restrictions here, but it’s actually more of a design reason that there isn’t intricate AI in MMOs (or, at least, it isn’t prevalent).
It appears that players don’t want inconsistent experiences, and complex AI can create inconsistent experiences. Many players actually prefer the mobs who just sit there with their sole purpose in life to get slaughtered by players.
That’s not to say there can’t be more complex behaviors in MMOs, but they need to remain consistent for players to truly desire and enjoy them… which means that you aren’t really making complicated AI so much as utilizing intricate predetermined behaviors.
Orc Warriors always attack mages first. Wilding Mages always blow themselves up at 10% health unless they are stunned. Venturian Archers always keep a distance until they are hit with a melee attack. Etc.
Experiences that are consistent and predictable after they’ve been experienced a few times, but must be learned and appear complex and random when first encountered.
I do believe that players would like to see more complex AI when it comes to NPC behaviors for NPCs they can’t (or aren’t intended to) fight… to a point. Quest givers who went to sleep at night in a house you couldn’t get in would be confusing and annoying. Flavor NPCs in towns who behave like real people would add atmosphere and immersion, however.
Anyway, this is a case in which I think players believe they want complex AI but what they really want is additional challenge and consistent but intricate behaviors. I know this post is going to be controversial, but there’s no need for me to pussyfoot around. Plus, it feels good to be controversial sometimes. 
I’ve said the same thing (sort of) in a post on my greenmonster dedicated boards. While not really intelligent, I’d prefer to see mobs with a simple “brain”. In other words, they fight in almost a predictable, simplistic way, yet these simple behaviours add a depth to the game.
I discussed at length with one person in that I’d like to see a game in which mobs actually “make sense”. In other words, mob creature A dislikes mob creature B. They will fight if close together. This really isn’t all that complicated as far as AI goes (indeed, I’ve run tests of my own to see if it’s feasable).
Simple comparison of “if A is in the vicinity of B, then A attacks B” (obviously, you can start adding in percentages, and see if players are near, etc. etc.), but to be perfectly honest, I’m tired of playing games in which I walk through an area and see 20 different creatures standing dumbly waiting for some even more dumb player to plop himself in the middle of the group. Some simple behaviours (as opposed to AI) are all that is really called for IMO.
But, then again, I’ve said some other things about certain designs that’ve gotten me laughed at
[...] March 19, 2007 06:00 Complex Mob AI: Sweet or Sour? There’s a great thread over at Quarter to Three called AI in MMORPG Games, in which the community [...]
Of course, this likely comes from data mining and experiences based on existing games, all of which are leveling grindfests. Its natural for people to prefer brain dead predictable encounters when the goal is loot and exp which only come at the end of the fight. We live in a world with games who’s design amounts to making the journey a hindrance to the destination, and the destination is what matters most.
Before people appreciate complex AI, you have to make them actually care about the fighting itself over the rewards obtained after victory.
[...] Complex Mob AI: Sweet or Sour? [...]
points on the second page, which can basically be summarized as: testing and data suggest that players don’t really want complicated artificial intelligence. Lum’s post and my thoughts after the fold… (more…) posted by Ryan Shwayder @ 22:00 Comments Off
I think Lum isn’t interpreting his statistics correctly. How much time does someone spend grinding versus working for loot/flags/etc.? Conservatively, I’d it’s probably 95+% grinding and 5-% other. So to me, Lum’s stats are saying is that players want a consistent “stupid” mob for grinding. Know I know this doesn’t state anything about questing for loot but it illustrates my point – it doesn’t state anything about what players want beyond grinding – and I think players (for the most part) want to earn their uber loot and having “smart” mobs as a challenge can accomplish this. That being said, challenges are great, but as Ryan said, it is really just making the player think it was hard – how many players will honestly rethink their strategy and try again? I bet most (average players) will just pound on the problem using the same techniques over and over again until they succeed. If that doesn’t work they’ll just whine on the forums until the quest/mob/whatever is nerfed.
Jason: Yes. This is all speaking in reference to a traditional type MMO (EverQuest, World of Warcraft).
And a point of clarification based on one of Abalieno’s responses at Q23:
Yes. I think my assumption of what is referred to as AI is different from yours (and seemingly the majority). I don’t see predictable behavior as AI so much as simple response-based events (which is really what AI is, there’s just some point where I don’t consider it AI because it becomes too predictable).
MOB AI that is nearly as intelligent as players would be frustrating (to me) because it would be too unpredictable. MOBs that react in a particular way any time Criterion (or Criteria) x is met, insofar as I am able to learn these behaviors and predict/plan for/counteract them, are awesome.
I still hold that people won’t really enjoy, won’t really appreciate that until the focus comes off the destination and lands squarely on the journey. Yes, I, and it sounds like you, would enjoy it immensely if there were hundreds of combinations of programmed responses to experience and learn… but “most people” tend toward paths of least resistance, and as long as the goal is levels and loot and the path is grinding, people will learn a couple or three mob types and focus on them in order to achieve the goal.
I don’t really want true AI… that would actually stink because a computer is always going to outthink me in the long run. It has to be limited in its choices, there must be a finite number of reactions in order for me (or any player) to have a chance of consistant survival.
Getting levels in most games is mindnumbingly boring. Exploring the game is fun, but it stunts character growth, and if you wish to participate in high end gaming or PvP it hurts your ability to do what you want.
I have to agree that truely intelligent AI would turn a game not-so-fun very quickly if improperly used.
I don’t want a pack of wolves planning a complex ambush. But I also don’t want AI so dumb that you have a wolf that wanders around next to all kinds of furry critters that should be it’s snack.
For me, it comes down to: it has to be real enough that I’m not shaking my head at it, but it also can’t be so real that little ol’ me is getting owned by a smart set of NPCs.
What particularly comes to mind is any kind of group/city interaction.
Yeah, its kinda fun in the game to pick off inhabitants one by one in front of each other… but it’s also stupid. Thats one reason I have a problem with “aggro ranges.” Come on. Dragon A isn’t going to sit there and look at this raid force waiting to kill it and say to itself “nope… 100m. nope, 90m. Oo, Dumb_Guy at 50m, killin’ time!”
But then comes the logistics side of it… it takes work to get a group or a raid together, and it’s no fun if you’re trying to form up and you can’t because the dragon keeps seeing you…
So I think in order to balance it out, you have to give a little bit of each. But also create with this balance in mind.
If you’re attacking some orc city, they should notice that. And if they do, you know theres going to be more orcs than you can handle unless you do something. So let the player do something… anti-aggro spells. A silence spell that prevents the mobs from calling out for help. An illusion spell that prevents the fight from being seen.
Or, maybe also add in some more NPC fun. A group of PCs joins a larger group of NPCs… the NPCs handle much of the fighting, but leaving enough for the group. That way the group gets to do their thing… but it’s also not the notion of slaughtering a city bit by bit in plain sight.
“I discussed at length with one person in that I’d like to see a game in which mobs actually “make sense”. In other words, mob creature A dislikes mob creature B. They will fight if close together. This really isn’t all that complicated as far as AI goes (indeed, I’ve run tests of my own to see if it’s feasable).”
EverQuest did this. You’d see Sarnaks fighting Goblins all the time in LoIO, and even in Old World you would see NPCs of different factions fighting each other.
But anyway, I think a lot of the posters have hit the nail on the head. Under the current model, I don’t want to have to out-think my opponents. That’s what PvP is for. As long as PvE centers on the XP and the loot, and the battles themselves are essentially meaningless, I want pretty stupid enemies with moderately interesting, predictable, and varied behavior.
Of course, I’m still not really tired of that model. It’s fun. I haven’t yet seen a graphical MMO that really convinces me it’s more fun to do something else than call my friends and gang up on hapless monsters with the IQ of a rock for fun and profit.
I think you can also make the MOB interact with the environ more to create better game play (you could call this AI I guess, but I would venture to say it might be more of level design/mob design interaction).
One of the most memorable mobs off the top of my head in the last year or for me has been The Berzerk/Ram type monster at the end of Stage 1 of Gears of War. The mob in GoW was stellar just because of the room he was in (and how it played into strategy) and the way the Devs forced you to flow from one part of the encounter to the next. Flow = baiting the thing to punch holes in the wall for you.
This may be more related to the level/encounter design than the actual AI of the mob, but I think it still suffices for creating an encounter that has a little more depth than the norm, and really the mob itself doesn’t have to be “smarter”.
I would have to agree with the general consensus, but would like to question other areas that could be related to mob AI… such as Faction. If I kill a Coyote in the desert (and don’t happen to collect it’s pelt to turn in) how will the Road Runners know I did it? If there wasn’t someone friendly to the Coyote faction nearby to realize it was me killing the mob and blog it so the PETA members out there could aggro me on sight, would my faction with them really go down?
Or look at it from the serial killer perspective. The neighbors have no problem inviting the quiet guy next door over for bbq’s… it’s only AFTER the police dig up his backyard that he’s shunned by everyone except cell-mate Bubba.
Does this fall into the same AI bucket? Do players want mobs/NPCs that are smart enough to handle faction in a more ‘realistic’ manner, or do players want faction that is predictable and automatic?
And… what about post kill? Or AI that may not make it more difficult to grind, but livens up the play experience. Sure, we might not want to fight a Bear that’s smarter than we are and once we know what the Bear’s routine is, we know pretty much how to consistently beat him. But what if the mob-to-environment was a little more unpredictable? Is the next Bear over a little hungry? Does he come wandering by to check out the yummy smelling Bear corpse, or is he too full to be bothered? Is he normally hostile, and attacks me without provocation, or is he sleepy and will just look at me from under his big bushy browridge and let me wander by with a warning growl?
Here’s my counter-example: do people want simplistic AIs for fighting games such as Virtua Fighter, Soul Caliber and so on? Or do they want complicated and challenging AIs that make them work for their effort and feels like there’s a person behind the controls on the other end.
Another example: PvP. Admittedly, PvP tends to be a minority experience for the vast majority of gamers, but there’s a question as to why this is so, usually dealing with matters of balance and so on. If people weren’t interested in better AI, they just want to wipe out mindless mobs, then why do people go for player versus player experience?
When you get down to it, the whole issue about AI is an attempt to make the PvE experience more closely resemble the PvP experience. In fighting games, they definitely go for that approach. The question is, will that work in MMOs?
Complex AI?
The sour:
Back to balance again…damnit.
5 years ago, when you spoke about AI for a game or in general you were automaticly drawn into a “How challengable this could become” debate, more or less.
Now add the powerfull MMO effect to the word AI and the question suddenly take one hell of a turn.
What was once a question of challenge is now a question of comfort.
AI should serve the purpose of challenge(fun) before comfort. And that is definatly not happening, for alot of good reasons, sadly
—
The Sweet:
Instanced zones milking the complex AI experience way up.
And;
It’s what I call PvEP
Basicly the game offers the grounds , tools and persona for both challenged parties. In otherwords the best AI you may find might just be another real player within good quality limitations. Just enough to keep your players away from SecondLifing their experience in a poop fest.
A quick basic comparaison starter is Capture the Barrack (not to mention Flag) , giving the challenge a medieval taste ( if your game is medieval based).
LOTRO has developped interesting concepts around that kind of interaction with their players as mobs.
—
I am all for inciting the players to focus on the journey instead of the grail but perhaps AI interactions is a base not flexible enough for that?
Smelling a mix of PVEP and AI, in the same zones, same non instanced world.
You?
Cyanbane wrote:
That’s the OTHER reason we don’t have complex AI in many MMOs. Scripting AI responses to triggers in the environment is actually what passes for AI in most games. Sure, it’s some fixed amount of extra programmer time and some CPU time on the servers, but it’s a TON of extra level designer time to build smart levels that the AIs can be smart in.
I’m pretty much with Ryan though. Once you kill a hundred of something all that fancy AI is going to be reduced to a pattern anyway, so you might as well just cut out the middle man and aim for a diverse set of consistent pattern. I loved going into the first mission or two with a new villian group in CoH because I never knew how to deal with the bad guys. The mix of powers would usually be very different, and it was an interesting puzzle. Of course the game usually made me play 50 missions against each set of villains so by the time I moved on I was sick to death of them. Diversifying them on another axis based on their role in the instance would have helped fight that fatigue.
I think this is a case of over-engineered interpretation.
The players don’t have the same concept of “AI” as we do; we developers make the mistake of reading our own definition and understanding of AI into the term when they use it.
What the player is looking for in AI improvements is relatively simple: a relatively simple finite state machine with some elements of randomness.
When player’s are talking about “better” AI they are often talking about improving the simple rudimentary behaviors: pathing, quest interactions where an “AI” is perceived as “dumb” because it doesn’t accept the items you’re trying to turn in or because it spawns 50ft up a wall and the game insists on telling you it is “too far away”.
Keep your AI simple, and build complexity through encounters.
I have two kinds of fun available to me: a dynamic encounter or pushing the lever to get a pellet of xp/gold. Either of these will provide some enjoyment to me this evening.
If I push the lever, I know what I am going to get. Goblins are not terribly difficult and I can consistently solo them; input 5 minutes to get 10 satisfying goblin deaths, 50 gold pieces, and 100 experience points. This will help me if I want to kill more goblins or try the dynamic encounter next time.
If I try the dynamic encounter, we have two problems. First, it is probably less rewarding per unit time. I must give up my fun shinies to have a more fun play experience, which seems like really poor game design but it seems like all the interesting things to fight are harder and far less rewarding for the effort than dumb mobs. So there is the trade-off involved.
Second, the experience is inconsistent. It might take a minute or ten minutes. I swear, if the enemy panics at 5% hit points and runs away from my melee-only character, I will strangle someone after I spend five minutes chasing it down, assuming it does not run through eight other enemies or jump over a cliff (oh, he doesn’t take falling damage? Lucky). I need to bring excess resources because it is unpredictable; do I have enough healing potions on me for this fight? My win is not guaranteed, so I might fight for a while and end up in a worse position (level- or loot-wise) than I started the evening. The encounter might be bugged and therefore not interesting or not possible, and I will not know that until I am in the middle of it. Tell me that you have not engaged in a complex, multi-step battle, only to have one enemy fall through the map, fail to spawn, be un-attackable, hide somewhere you could not find, etc.?
So I am giving up my shinies and slowing my dinging in return for the chance of having an interesting encounter. And how interesting is that encounter really?
How often do you try new restaurants? How often do you get fast food? You do not go to Burger King for fine cuisine. You go because it is quick and reliable. Who wants to try the new Thai/Lebanese/Hungarian place, only for it to take an hour, cost five times as much as fast food, and turn out to be completely unpalatable?
If you want me to take a risk on something, the payoff needs to be worth my while, because I have been burned before, a lot. Despite that, we keep trying new things. Sometimes. People will try more if the most interesting encounters are the most rewarding ones.
If you had a game where the AI wasn’t for opponent’s nessecarily, but just social entities, and you had constrained multi-player instances backdropped against that simulated culture, then you’d have a very different story. Of course, there aren’t games quite like that yet, so I can’t test that hypothesis.
I think some of the people here are missing the point of upping the AI: if you make for a more challenging AI, then you need fewer mobs. Saying that fighting one smart mob is going to be less rewarding than fighting a dozen stupid mobs is an example of poor game design. The idea is to come up with a good rate of satisfying progress for players. If upping the AI makes the fights harder and fewer kills per unit time, then you up the reward to restore the desired rate of reward per unit time.
I don’t think they are missing that point, mutant, rather the premise here is that the players don’t see it that way which is [or was] a bit of a shocker to some devs. Players are always asking for smarter AI, but when presented with AI NPCs that win awards or land new beachheads in the field of artificial intelligence, players don’t react positively to it.
The average player is no Kasparov, they’re not looking to fight big blue. The AI they care about is the part that makes the NPC a believable actor within the world – pathing, quest text, etc. I think you will find that the average player actually thinks of the dialog given by an NPC as “ai”.
Again, I think the lesson here is that we’ve gotten a little caught up in the possibilities and allowed ourselves to over-think some aspects of the technology. Quest text is a perfectly valid form of artificial intelligence in this use case.
I think an interesting way to simulate intelligence while lessening some of the issues mentioned above would be to let the mobs ‘learn’. This could range all the way from inidivual creatures to the collective conciousness of an entire species.
And within each section there could be a wide range of implementation methods. It could be something as simple as letting mobs level up. IE, a mob is involved in the deaths of XX number of players, the mob ‘levels’ and gains increased stats, making it slightly harder to kill than its brethren. And/or you could add new abilities as they level up. Perhaps if a generic orc reaches level 5, it becomes an orc chieftain and learns how to snare its opponents and call in additional help when threatened.
On a larger scale, you could implement something where a mob camp collectively learns based on its experiences. Are heroes coming along and slaughtering them individually day after day? Well then maybe the mobs start walking around in groups of 3, or 5, or 10 depending on how bad the situation is! To prevent an over-escalation of power, I would have it work in reverse as well. Has that same camp become so strong that they never get attacked? Well then they might become complacent and wonder out alone more often. Or perhaps extend their wandering patterns to cover greater areas. Or maybe they tend to be drunk a lot of the time, resulting in a decrease of their abilities.
You don’t really want the NPCs to think for themselves too much. But you want to give them a semblance of organized thought, which is actually just a complex interaction of scripts resulting in varied, but somewhat predictable behaviour. Of course, mapping out these behaviours would be up to the playerbase, and I would throw in a few changes on a regular basis just to keep them on their toes.
Oliver, I would take that in a different direction: in practice, the reward is rarely increased commensurately with the difficulty of the encounter. As Mutant For Hire says, “If upping the AI makes the fights harder and fewer kills per unit time, then you up the reward to restore the desired rate of reward per unit time.” Many games add the improved AI, but the rate of reward per unit time is much higher in the areas with weak AI.
I cannot tell whether players want improved AI from any game I have played. I can tell that players are not willing to give up advancement for improved AI, which is the choice they have. If anyone can point to a game where the good-AI encounters have as high a reward per unit time as fighting dumb rocks, we have something to discuss.
The best case scenario seems to be this: (1) new encounter is introduced with improved AI and a good reward; (2) seeing fun and reward, players flock there; (3) seeing this flocking, the developers assume the reward is too good and lower it; (4) repeat 2 and 3 as long as the “fun premium” holds out on the encounter. Also, the fun encounter is abandoned last by the most competent players, since they are best able to keep a high rate of return; they have a comparative advantage against good AI, because almost everyone gets the same return per input against dumb-as-rock encounters. The reward for the good-AI encounter will end up set so that it is worthwhile only for the people who can complete the encounter most efficiently, who can then get bored with repeating it and move on. The rest of the players will only visit the good-AI encounter when they think that the chance of improved fun is worth the loss in reward they get.
This need not be the case, but my cynicism tells me it often is. I know that when I have tried more dynamic encounters, my risk has increased much more quickly than my reward. My risk of a bugged encounter seems to be one of the fastest rising problems.
[...] Blog · MMOs · Links · Forum · Login · Register Nerfbat – Complex Mob AI: Sweet or Sour? http://www.nerfbat.com/2007/03/19/complex-mob-ai-sweet-or-sour/ "There’s a great thread over [...]
A lot of what the players want to see in an AI is believability, a realistic response, something normal. We get used to it after a bit in these games, but when you step back and look at pulling npc after npc out of a camp….wouldn’t someone notice?? Adding a bit more realism to these instances would go quite a ways towards quieting the better AI brigade.
A truly smart AI isn’t something people will ever want unless they have counters to the actions. A smart npc will take out anything in a dress first because they are easy to kill and will reduce the damage output against them quite a lot. Smart npc’s will play like we do..and in most of these game there aren’t counters to stop them.
For it to be viable you would need to have full collision so an npc could be blocked from the healer/caster. You would have to get away from the current aggro dynamics and build something truly realistic before you could create smart npc’s. You cannot just increase a mobs intelligence and call it good, you have to build the game in such a way that any move that a mob might make can be countered.
When many of us are doing is to look at current games and encounters and adding an additional level of intelligence onto mobs. That won’t work because the players don’t have enough options to either stop the mobs or survive the mobs attack. Our current crop of games and designs are built around tanks, aggro, and controlling the mob. But if you build that intelligence into the game from the start and give people options…then you have something truly fun.
One thing I’m curious about,
(And hello, btw, I’m a real estate developer in Boston looking into developing some game ideas I’ve had for years, great forum)
is what folks think about the place that pc-controlled or gm-controlled events or mobs have in MMO’s. It’s been a long while since I’ve seen a gm-event in WoW, and then only in screenshots. Player or gm-controlled mobs seems like an ideal way to test how an entire player base would respond to a true challenge. I wonder if any developer blogs note this sort of interaction?
I know from personal experience in pvp that given any sort of reasonable intelligence, the tried-and-true “meatshield” phenomenon doesn’t exist in a player-level challenge situation (i.e. healers and damage-dealers die first, no exceptions, contrary to how every single AI-run mob is forced by script or player-utilized aggro ability to prioritize its’ kill order from least threatening to most threatening)
For me, as a player, this is a true “head-shaker” when in teamspeak on a daily occurrence players joke about the inability of a God-level being to not take heed of the 10 or so casters keeping one tank standing, or the 20-odd beings actually doing it all the damage.
I think in the case of some of the current MMO’s, a tradition and expectation that such roles should exist (meatshield, back-stabber, ranged caster, etc) forces the developers to code mobs in less than ideal fashion in order for everyone to “play their role”.
I’m wondering how far a game designer might need to go in breaking those expectations in order to actually be able to code mobs with semi-intelligent behavior, and if there is even any market in doing so?
If you like reading about game design, game mechanics or even working in the game industry, be sure to check it out. Some highlights that I liked from this month’s carnival include reflections on intellectual property and gaming, what complex AI would do to gaming as well as an analysis of just how to define casual and hardcore.