Being an MMO gamer for years, I can't help but notice how rudimentary Artificial Intelligence is implemented in the vast majority of titles. NPC's and monsters little provide to a game to help convey the feeling of its setting. They lack anything interesting to really set them apart and make them become a part of the game world. What you see is a sad reality where NPC's live extremely static and invariable "lives". Not only that, most MMO's suffer from a constant lack of content with players rushing to see the entire game in a glimpse. I believe that working towards implementing a more lively and realistic AI could not only convey better the immersive aspects of a game, but it could potentially resolve the problem with developers having to constantly pump new content out.

There have been a couple of ideas that have sparkled in the industry that has led me into this train of thought. First, I saw a news report of a group of students that had developed an AI for Starcraft that could perform real-time decisions based on the situations at hand. This allowed for much more spontaneous gameplay, a far more involving game, and a vastly more intelligent opponent. Another was Skyrim and its bold claim of having "infinite quests" with its Radiant quest system.

What I envision is a system where NPC's can make decisions based on a list of options available to them whenever they are presented with a situation. This list of options would vary not only based on what the NPC actually does in the game world (king or a blacksmith), but would also vary based on the NPC's "personality" and the situation presented to the NPC.

Take, for instance, Bob the blacksmith (an NPC). Bob wakes up everyday, goes to work, eats his meals, talks with friends, heads back home, takes care of his children, goes to sleep, and starts his routine all over the following day. However, on the following day, Tom the thief (another NPC) decides to steal Bob's blacksmithing shop. This presents Bob with a situation and a list of possible options. Bob is configured to being a rational and calm man. His options are the following: 1.) Ask for help. 2.) Forget about the thief and fortify his shop. 3).Go after the thief. His personality, his need to provide for a family, and his profession as a blacksmith (he is no thief hunter, that's for sure) lead him to ask someone for help. A player may come by and choose to help Bob or not.

This is just a crude example, but I hope it conveys my message. In this case, the Artificial Intelligence performed a decision in real-time and spontaneouly generated the possibility of a quest. Several other interactions would be occurring at the same time, and I believe this would lead to infinite content, as content would be tied to the life of the game world itself.

most mmo count on quantity right now rather than quality that's why people seem to feel bored of the kill 10 boar quest because to attract people you need 10,000 quest.

While doing what you ask might be awsome in reality it would also bring us back even further in the realm of endless boar kill quest with some luck at get that epic mace laying around for free quest. But it's definitely something that would attract new liking to the system ... You could even expand the idea to have it tied to daily cycle around the life of the world as in ...

bob goes to sleep and get awaken by stray dog ( go kill dog)

bob just woke up and he has been robbed( go catch the thief)

bob is being harrased by brigand ( kill brigand)

and finally bob is crafting epic mace ( collect mats to gain epic mace).

All of this for a single npc multiply that by x many npc and you could even tie them around giving you endless content that seem dynamic without requiring much effort.

AI is very dry in MMOs ranging in terms of both NPCs and mobs, no doubt.  I have been going over this in my head lots for the concept I'm developing.

My question is this: will expanding the AI across the board in terms of both PvE and NPCs greatly increase server resource requirements? 

If not, it would be nice if the MMO genre would get more onboard with AI development and implementation to create a much more dynamic experience.  Maybe highly advanced AI is not cost effective to do just yet.

edit: I also highly agree that, in general, making these MMOs more dynamic and challenging will eliminate as much need for additional added content.  Problem here is: that's exactly how Blizz and others milk the **** out of sales further.

Professional AI programmers (people who've made multiple games and watched how players play them and react to various methods of AI) would point out that if the player doesn't perceive it it doesn't exist.

In this case Bob's AI gives out the same exact quest he could've just been hardcoded to do.  Most (possibly all) players won't know the difference, and if we don't know the difference the underlying logic is completely meaningless.

Time and time again when I read articles about AI the important keys to success are about giving the perception of intelligence.  The actual need for true intelligence isn't there -- in fact in many situations intelligent AI is detrimental to the fun of a game.

So while AI can certainly stand to be more game-like, less exploitable, and less perceptibly stupid, it's actually not a good idea to make it more intelligent for intelligence sake.  

Well, better AIs have been around in other genres for a while, Morrowind was more or less pioneering it for single player games where npcs sometimes got married and got kids if you were away for a while.

But it isn´t a true AI, games with those are 30 years away still, just improvement to the "always doing the same thing no matter the time" npcs most MMOs still have.

Combat AIs is another thing were MMOs truly suck.. The trinity combat kinda makes mobs dumb as doornails (or as Nobby Nobbs would put it "salt of the earth") and MMOs really are after other types of games there. 

It isn´t hard to add a little more to NPCs both in and out of combat but it wont really be a AI. While true AIs probably will show up sooner or later it just isn´t a possibility right now.

Originally posted by Axehilt

Professional AI programmers (people who've made multiple games and watched how players play them and react to various methods of AI) would point out that if the player doesn't perceive it it doesn't exist.

In this case Bob's AI gives out the same exact quest he could've just been hardcoded to do.  Most (possibly all) players won't know the difference, and if we don't know the difference the underlying logic is completely meaningless.

Time and time again when I read articles about AI the important keys to success are about giving the perception of intelligence.  The actual need for true intelligence isn't there -- in fact in many situations intelligent AI is detrimental to the fun of a game.

So while AI can certainly stand to be more game-like, less exploitable, and less perceptibly stupid, it's actually not a good idea to make it more intelligent for intelligence sake.  

ummm more "game-like?"  What?  Not sure what you mean.  I'd rather have AI be less "game-like."  Feel more immersed by seeing NPCs with routines instead of standing around in the same spot... never ever moving. 

edit: maybe I'm not speaking of AI but scripting... still a little new to proper usage of many terms.

From what i've experienced ingame the radiant story/quest is quite simple though. It's supposed to be all-encompassing, but most stuff was already there. If it's really done by an "AI" of sorts, instead of thousand of scripts and case checks, it's a nice step forward, but it's not really noticable.

And some stuff is more of a step backwards. I kill someone deep in the woods near Riften, the next morning, his relatives in Solitude not only already heard of the death, they also know who did it, hired a killer and sent him after me. And that relative might be a daugher he never mentioned when speaking about his family, and who doesn't like her father when talking directly to her.

It might be infinity quests, but if npc a wants me to go to cave b, and npc c says go to cave d, whats the difference if the caves are nearly the same? Reward same, effort same, experience (seeing similar stuff etc.) same, i don't need an infinite amount of these quests.

Now if i kill someone in the wood near a lumbermill without anyone noticing, and the villagers later ask me to find out who the murderer is after the lumberjacks found the body, whereas when i hide the body good enough, the villagers only wonder after several days, because they believed he went on his usual sales tour, and even then still suspect he simply settled there..and if i can actually get a quest to cut some wood, and "find" the corpse myself..

And now expand this to dozens, houndreds or thousands of players, someone kills a NPC in the wood, and only after either a player, or a NPC (who's inclined to tell it to the guard etc., a known thief might avoid going to the village after finding a corpse..) found it and reported it there is a quest available. And only for a hand full of players, who actually get told several different versions, because the later get told someone is already helping, and there might not be a reward left once the return..but if you take too long, they offer it again..but not if *everybody* takes too long, because they fear they might sent people to their doom..

One thing is that when doing a single quest, no matter how many steps where involved until the NPC decided to offer you exactly this quest, it could all be scripted, but the amount of different quests would tell.

It's like what ArenaNet will do with their Dynamic Events in Guild Wars 2, but taken a step further.

I would certainly like that, but not if it's the same quest over and over only replacing names and then saying "infinite quests". Obviously you got a lot of possibilities when you  take 100 npcs and 100 places and generate all possible quests, but it would most likely end up feeling like 10 different quests and the rest feels exactly the same.

Originally posted by Anireth

*Snip*

I would certainly like that, but not if it's the same quest over and over only replacing names and then saying "infinite quests". Obviously you got a lot of possibilities when you  take 100 npcs and 100 places and generate all possible quests, but it would most likely end up feeling like 10 different quests and the rest feels exactly the same.

Doesn´t all MMO quests feels like 10 different quests, or at least almost all?

I think I killed 10 of all possible critters in MMOs from rats to hill giants.

Originally posted by Anireth

And some stuff is more of a step backwards. I kill someone deep in the woods near Riften, the next morning, his relatives in Solitude not only already heard of the death, they also know who did it, hired a killer and sent him after me. And that relative might be a daugher he never mentioned when speaking about his family, and who doesn't like her father when talking directly to her.

That sounds like horribad implementation.

The problem with Bob going home to his family at times is that Players are on a completely different schedule. The game itself would have to make the options at hand, for when time changes in-game, to tell Players that "Bob is done for the day, talk to his wino brother Bill at the pub". If NPCs are no more than fluff, as in, they just wander around and have no actual use beyond aesthetics, then this works. If they hand out quests, or require *any* kind of meeting in-person, I will call it a straight-out flub. It works in Skyrim because you can manually change the time instead of breaking into their house and waking them up to turn in something. Not so much online.

Now, the thing that I do find interesting is the concepts of AI here, but in terms of 'reaction'. I was playing MW3 with a bud the other day and spent most of it grumbling my thoughts on how little effort goes into a game for all the money thrown at it. My argument was that "People are not going to act stupid enough in war to stand right in front of you while you reload without, at least, some kind of kneejerk reaction - but - I guess if the games were any more believable they would be impossible to play, let alone beat". Imagine a realistic wargame where you die just as fast as you are supposed to, basically PvE as impossible as taking on human opponents online. The entire scope of the game would have to change in order to draw it in, and nobody is willing to play a shooter without regenerating health anymore (besides TF2).

Hell, even Madden games have zero hint at player personalities, or what could be expected of them. Have any of you seen a computer-controlled member of your team get a flag for unnecessary roughness? Or getting too hasty and getting called on offsides? The games have no sense of personality, nor any kind of swing momentum like sports tend to have. It's not going to get harder in the last quarters, people aren't going to gain/lose morale, and people simply don't get tired. If you play a Madden game and just gutter-stomp the other team early on, that trend won't change, because there is no real scaling of difficulty mid-game. Sucks.

The step I *can* see, though, is simple adding more 'reaction' to NPC behavior - they all act like robots, and because of it, are extremely predictable. A method I would suggest is two-fold; that of the chance roll and pools similar to HP determining behavior.

I'll throw an example;

Typical shooter, but when firing at enemies all the bullets whizzing by will raise a pool based on 'fear', which then makes a roll after every tick past a certain threshhold, and that roll's chance is higher as the bar fills. If successful, will make that NPC flee to a new position, and that pathing route would be any one of a handful, based on the rails used to take cover on in the first place (if any), and then a simple roll of the dice could determine which route he takes.

Lots of different pools can be kept in mind, for various behaviors and outcomes, but the specific outcome itself is based on a roulette of event situations. Would add the aspect of enemies changing tactics after enough stimuli, and the outcome somewhat randomized.

I agree with Axehilt. Aside from a handful of games where the whole point is to just watch the world unfold procedurally (Dwarf Fortress immediately comes to mind), there's not much benefit to having complex NPC motivations going on behind-the-scenes.

Originally posted by Disdena

I agree with Axehilt. Aside from a handful of games where the whole point is to just watch the world unfold procedurally (Dwarf Fortress immediately comes to mind), there's not much benefit to having complex NPC motivations going on behind-the-scenes.

Yes there is. You just have to do it right. The behind the scenes stuff has to effect how the environment interacts with the players. Sure, having Bob walk around is only annoying, but having monsters whose life or death or location doesn't interfere in a specific quest line interact in a way that changes the world is actually useful.

Here's where I believe it to be more interesting than a standard quest system. If the interaction between Bob and Tom can spontaneously create a quest, what about the interaction between Bob and a real player or Bob and the game world itself? What if Bob's shop were stolen by a real player and the quest now involved killing or taking back Bob's stolen items from the real player? What if Bob's shop were damaged by a lightning storm and Bob asked players to help fix his store? 

Not only is there a large possibility for there to be a lot of content creation on-the-fly, there's also the possibility of adding real consequence to quests. If no one helped repair Bob's shop after it was damaged by the lightning storm, he would consequentially go out of business. He may then lose his home, become a beggar, and possibly even go to other cities in hopes of finding new opportunities for work. What if Bob used to supply the town garrison with armor and weapons? Now the soldiers have no one to provide them with the necessary tools for battle, and a rival village's leader decides to invade and conquer the town in its weakened state.

You can see how one thing could lead to another and the world would always be changing, always providing new opportunities for content. Not only that, the world would be immensely more believable and immersive.

Originally posted by MacroHard

ummm more "game-like?"  What?  Not sure what you mean.  I'd rather have AI be less "game-like."  Feel more immersed by seeing NPCs with routines instead of standing around in the same spot... never ever moving. 

Yeah, game-like.  Where interacting with the AI isn't just the face of a game's PVE, but actually creates gameplay completely on its own which layers on top of the other game systems.

That's why some of the best PVE in games are things like WOW bosses where you have your Personal Rotation layer, the Boss Abilities layer, the Threat (AI) layer, and possibly an Environmental Challenges layer.  Each layer adds to the overall gameplay.  Whereas if bosses acted entirely realistically the game would be as fun because that layer wouldn't be a minigame unto itself (which is what Threat-based AI is.)

Similarly you have AI like FEAR and COD where the AI's job is mostly to act as skeet shoot targets, because when the AI is realistic and flanks and never pops its head out without a hail of bullets and covering fire from teammates, well that AI just isn't as fun for players to engage in.

It's really only when you get to strategy games where the games are deep enough that the AI has to be intelligent to even be playing the same game as the player, that smart AI becomes necessary.  But even in these deeper game genres, AI benefits from being game-like.

If you make AI too smart, you expose players as dumb.  And they won't generally play to feel dumb.

It's the balance question that's been put off because of the missing tech needed for strong AIs.  Now that we are close to having the tech, the implementation is the question.  And given the current trends, AI's that are 'smart' about being dumb is the way it'll go for a while.

Personally I would not mind some unpredictable, challenging AI opponents.  But if NPC Brick Dumb takes 20 seconds to beat, and NPC Joe Cool takes 120, guess who gets picked on?  You'd really have to hang a big carrot on Joe Cool to change that.

I'm not talking about NPC difficulty levels. I'm talking about NPC's making decisions on their own from a scripted list of options which would allow the world to truly feel alive and could potentially create the possibility of infinite quests. AI difficulty is something else entirely.

You're correct, AI in MMOs is incredibly poor compared to what is possible.  It seems like most MMOs just emply pre-defined routines for the NPCs to follow.  E.i. generally just walkinging in random directions in a pre defined radius.  The combat is also predefined - it's just a big if statement where a particular action will be performed depending on what the player who is interacting with it does.

This is so primitive compared to some of the latest stuff around in AI these days.  If the game developers can't be bothered to implement a good AI system.  Why not just buy a provenly effective neural network  / support fector machine / etc driven AI that has already been coded - train it on data from real life players, and just call the correct functions that cause the AI to do the related action in game.

I think the issue, as others have said, is that some games have to have a deeper AI and some games are not built with the need for a deep AI.

There seems to be a misconception that MMO mobs are dumb because designers can't put in complex AI or are reticent to do so because it will frighten people away.

MOBs are not there to be complex or challenging. They are essentially powerups. It has been said before, by developers and players, but when players encounter more difficult or more clever mobs, it might seem novel at first but eventually they go and find mobs that aren't as hard.

Having said that, if a game developer made a big deal about how hard the mobs would be and that the game was made to be a challenge at every turn it could draw a more hardcore group of people.

Originally posted by Axehilt

Professional AI programmers (people who've made multiple games and watched how players play them and react to various methods of AI) would point out that if the player doesn't perceive it it doesn't exist.

In this case Bob's AI gives out the same exact quest he could've just been hardcoded to do.  Most (possibly all) players won't know the difference, and if we don't know the difference the underlying logic is completely meaningless.

Time and time again when I read articles about AI the important keys to success are about giving the perception of intelligence.  The actual need for true intelligence isn't there -- in fact in many situations intelligent AI is detrimental to the fun of a game.

So while AI can certainly stand to be more game-like, less exploitable, and less perceptibly stupid, it's actually not a good idea to make it more intelligent for intelligence sake.  

I remember someone uploaded a conference where they talked about different types of AI.  I wish I had the link, as it's very pertinent to this conversation.

I am REALLY paraphrasing here(it's been years since I saw it), but in summary, they talked about how most AI, particularly MMO AI is, and should be designed to lose.  Because the goal isn't to kill the player, but rather to provide a surmountable obstacle to getting to a goal.  The game or task at hand doesn't really focus around 2 sides trying to "win".

The AI for, say, chess games, or RTS's is different.  There they intend for the AI to behave like a player.  It plays to win, because unlike MMO's, winning IS the game.

I wish I could remember more.  It was pretty interesting.

Originally posted by Robsolf

I remember someone uploaded a conference where they talked about different types of AI.  I wish I had the link, as it's very pertinent to this conversation.

I am REALLY paraphrasing here(it's been years since I saw it), but in summary, they talked about how most AI, particularly MMO AI is, and should be designed to lose.  Because the goal isn't to kill the player, but rather to provide a surmountable obstacle to getting to a goal.  The game or task at hand doesn't really focus around 2 sides trying to "win".

The AI for, say, chess games, or RTS's is different.  There they intend for the AI to behave like a player.  It plays to win, because unlike MMO's, winning IS the game.

I wish I could remember more.  It was pretty interesting.

That would be Soren Johnson, a smart designer who's written a lot of design articles I've enjoyed in Game Dev magazine.

Playing to Lose: AI and Civilization