Tobold's Blog
Saturday, July 05, 2008
 
Artificial stupidity

Many people hate murlocs in World of Warcraft. But when you consider the reasons for it, you'll see that murlocs are hated because they are slightly less dumb than other mobs: many of them use ranged combat, and thus can't be pulled away that easily, and when they are hurt, they run for help. That makes them annoying to kill, but also shows how MMORPGs are designed around artifical stupidity. Even with the most intelligent races of them, a single adventurer can kill half their settlement in plain sight of the other half without anyone noticing or reacting. That is essential for soloable gameplay, but not very bright.

This explains why work on artificial intelligence in games advances so much slower than work on lets say graphics. There are graphics cards, and physics cards, but the "AI card" never took off. Because more graphics power is always good, but too much artificial intelligence is just plain bad. If the first mob in Karazhan would do the logical thing and shout for help, so that all of the mobs in the place come running to repel the intruders, even the world's toughest raid guild wouldn't have beaten the place yet.

One argument for PvP features in MMORPGs is that they would provide players with intelligent and unpredictable opponents. But being intelligent, these opponents object to getting slaughtered, which is why even PvP-centric MMORPGs need dumb PvE monsters, who don't mind getting beaten all the time, and don't log out to roll a new character on the winning side instead.

Artificial stupidity is a design challenge. It isn't actually all that hard to program mobs which would play WoW better than a human, because they have much faster reaction time, and playing WoW well doesn't actually require more brain power than a PC can provide. It is hard to program mobs who *seem* to react intelligently, and still manage to lose all the time. The standard "taunt" and aggro management mechanics are a typical example of artificial stupidity: It's stupid for the mob to hit the best armored and least dangerous member of the player group. But it plays well, because it is just hard enough for bad groups that make aggro management mistakes to fail as a result of it. That gives the players the impression that by good aggro management they outwitted the mob, and winning because you play well is fun. That works so well that nobody has come up with a better method yet. Even Age of Conan, with its superficially different combat system still relies on taunts and aggro management in group combat. If the devs wanted to program better artificial intelligence, the mobs would kill the healer first and the mage next, and MMORPG combat would be a lot different and less fun, especially for the cloth wearers.

So what the future will bring is monsters that are still stupid, and still lose all the time even if they have superior numbers. But maybe somebody will manage to make them appear more intelligent. The one field where MMORPGs could use a large dose of artificial intelligence is with NPCs: Many quest NPCs don't even recognize you any more five minutes after you saved them from death. More intelligent NPCs could act as source of information, giving you directions to quest goals for example, which would be better than looking the quest locations up on Thottbot. But even for MMORPGs game companies spend far more money to develop graphics, than to develop an artificial intelligence that would make the virtual world come alive. Artificial stupidity is here to stay.
Comments:
Your post title reminds me of this MIT course: https://hkn.mit.edu/6guide/src/fall91/6034.html

Seriously, though: While it's true that in MMOs, computer controlled characters are (by design) completely stupid, there *are* genres of games where AI is very important, and developers clearly put significant effort into making the best A.I. that they can: Take Half Life. Or Unreal Tournament. Or any RTS where you play a computer opponent.

On the other hand, I wonder if it is possible to design a fun MMO where artificially intelligent enemies are a part of the game. For example, you attack a village of foozles and they mount an organized response, rather than sit idly by until you come within a certain aggro radius...
 
Hang on there a minute Tobold.

A lot of gamers like it when the mobs display a reasonable level of Artificial intelligence. In Guild Wars for example the designers have tried to get the mobs to behave more like human players. They generally target your healer first for example. Of course they are still pretty dumb but it does make for more interesting and challenging combat. I Still remebmer the first time I played FEAR I was crouching behind cover shooting some enemies when One of their number went round the back of the building to flank me from behind. I was stunned and impressed.


Most MMOs at present use dumb enemies and just up their hitpoints and put them in large bunches to increase the challenge. If the AI improved you need to reduce the hitpoints and reduce the number of enemies but overall I think it would make for a more enjoyable game.
 
PvE gameplay is designed so that the players win 90% of the time for "normal" challenges.

The design patterns with taunt and close range aggro goes hand in hand with the class design - if you have healer/tank/damage dealer type of classes, then you end up with a certain enemy mechanic in most cases.

Tabula Rasa did things differently here, providing different venues for combat tactics and enemy behaviour. I think that was a refreshing change.
 
I think of AI as The Holy Grail in software development (in which industry i make my money btw). In contrast to what you say i think it would be very difficult to create an WOW mob who would play better than a human. Most importantly that mob would lack adaptability.

Which would lead to the human losing perhaps the first few encounters but then recognizing patterns (unavoidably with present day technology), adapting to these and using them against the mob.

Adaptable, evolving, creative and learning AI is still lightyears away i guess.

Even the most succesful implementation of AI to date, which would be in chess, only succeeds by using brute force (which isnt very intelligent to begin with) in an unchangeble world bound by a very small set of rigid rules.

With present day technology incorporating genetic algorithms in mobs are our best bet i guess. Beyond that who knows...(quantum computers?).
 
With regards to the stupid quest NPCs: In LOTRO at least they give occasional comments like: "Those robbers won't harass us any longer now, thanks to you!". Gives me at least a warm a fuzzy feeling. :-)

The new Prince of Persia installment plays around this challenge, the designers apparently plan for less fights but more "mini-boss-fights".
 
MMOs are at the bottom of the food chain when it comes to enemy AI. Really, the mobs are little more than resource "nodes" that require a bit of time to "farm" and allow you to progress by giving up some XP and loot.

Smart enemies would be computationally very expensive to implement in an MMO world where the server software is responsible for handling a huge amount of mobs across the whole virtual world simultaneously.

Games in other genres implement more advanced AI routines. However, in most real-time games the AI is just a series of scripts stiched together by a few condition checks and decisions. From an academic point of view, the games most likely to empoy "real" AI are simple turn-based games. Thier rules yeild well to having the computer build and recursively parse a tree of possible outcomes several moves ahead and pick the best next move based on some fixed critera.

Real-time AI is much more messy, since the number of variables involved and lack of well-defined fixed states makes a hard mathematical approach computationally unfeasible. So, we mostly see scripts that may look impressive but are easily foiled by an observant player.
 
"If the first mob in Karazhan would do the logical thing and shout for help, so that all of the mobs in the place come running to repel the intruders, even the world's toughest raid guild wouldn't have beaten the place yet."

Well, that would be a pretty stupid response too - send one toon in the front door, then everyone else sneak in the back door.

The problem with mob AI is that it is so very limited - fight or flee, and not much else. This means the range of possible player interactions is also limited, and that is why players complain. Not enough of a puzzle to keep them interested.

Change up the range of mob IA responses and you'll make the game more interesting.
 
Most MMOs at present use dumb enemies and just up their hitpoints and put them in large bunches to increase the challenge. If the AI improved you need to reduce the hitpoints and reduce the number of enemies but overall I think it would make for a more enjoyable game.

This is my response also.

While it doesn't make sense that a person would just be able to kill a bunch of creatures in plain sight without a response, it also doesn't make sense that a single person would be sent to deal with a whole village of bad guys either.
 
@MrGamer
Agreed, turn based games offer a much more AI friendly environment (chess).

A more challinging AI within non-turn-based games would be possible though i think. Since it will always act within the strict boundaries set by the environment and more precisely, the programmers (until we can create truely adaptive AI)it does not need an exessive amount of resources. It would behave like moving through a multiple branching script, which could be implemented with nowadays technology. The situations it can react to are set and limited of course, so strictly speaking it is not real AI.

Genetic algorithms can make this process even more interesting, as mobs will be able to change the reaction trees by 'mutating' towards the ideal state given the environment and or opposition (for example: 1 mutation per 1000 respawns).

Still, this would be all within set boundaries (the "ideal" state being defined by the designers) and thus within the realms of prediction. Real AI will not be predictable.
 
It would be quite easy to make a more intelligent AI - but that AI would stil be quite stuipd.

This GameDevelopers decided to not do something they cannot do perfectly. Which is a good decision.

I remember in Aoc you sometimes pulled half the village by just pulling one guy. A lot of players complained about this bug .. which indeed it was/is .. but .. you see my point.

Classic Dungeons have always been completely stupid when it comes to 'pulling'. Often they are even stupid in P&P RPGs.

Even if you could make a good AI you first needed to get players with the right attitude, otherwise they would have any fun.
.. and there were MANY exploits btw.
 
Oh, while I remember it .. mob AI shouldn't be limited to just tactical responses either. Consider the possibilities at the strategic level, where you might have a whole zone of mobs acting in concert - some would be patrolling the north, some would be assaulting the player's base camp, all might be ignoring capturing a mine resource for the moment.
 
Actually, Everquest had many superior designs on their mob aggro system. It has however been dumbed down over the years as people complained it was too hard. But if you think back you will remember that mobs in Everquest would go after the healer(s) first and would also be unkind to anyone who tried to sit and med, ect...
 
I think I read that for one of the siege battles in one of the LOTRO movies the programmers gave the CGI soldiers too much AI, and realising the odds they were facing the CGI soldiers turned and ran. The programmers had to dumb the soldiers down so they would continue attacking despite seeing their fellow soldiers slaughtered by the hundreds.
 
An example of where blizz have coded the NPC to play the way a human would, is heroic MGT.. going in there with my shadow priest, if the warrior or rogue is up, I know my time to live is going to be short.
 
I like more AI. I like smarter AI. Challenge is a good thing and there's a lot more fun for me to try to outsmart clever AI than to work out the Nintendo-style boss code of step 1, step 2, yawn, step 3.

If we had better AI in these games, we'd have more choices.

I was actually taken aback by the AI of the mobs in Atzel's Approach (AoC). It's not that they are super clever, it's just not what I expected. After getting past the initial shock, the longer I'm there, the more fun I'm having.

I think the money in the MMORPG business follows what people expect, what they're comfortable with. Such a shame.
 
Icewind dale 2 had this.... Baldur's gate 2 had this... I don't think its a technological hurdle we need to overcome here, its definitely a social and business one.
 
WoW currently has four boss damage mechanics: 1) aggro-based melee damage that requires a dedicated tank. 2) aggro-based magic damage that requires an unusual tank, like a mage or warlock. 3) random or group damage that is unavoidable, and that any class can survive through. 4) damage that is designed to be avoided somehow (for example through interrupts, movement, kiting or simply standing in the right place at the right time).

Those mechanics are designed to make every class useful in a raid setting, but they are at odds with more clever AI. The thing is, I agree that it would be most "logical" for the mob to attack the healer (this is basically also the most obvious PvP strategy in arenas). What I don't see is why it's a given in MMOs that the healer is the weakest link.

I guess this mechanic comes from D&D via Everquest. It works fine for group PvE if you accept a certain loss of immersion (i.e., players look at the spreadsheet-like threat meter more often than they look at the big scary thing they're supposed to fight). It doesn't work that well for PvP, and it's also not that great for solo PvE. Which makes me wonder if it's possible to balance a game around PvP and solo PvE first, and then invent group PvE mechanics around that.
 
"An example of where blizz have coded the NPC to play the way a human would, is heroic MGT.. going in there with my shadow priest, if the warrior or rogue is up, I know my time to live is going to be short."

But on the other side you have a skill called VE as a shadow priest that heals everyone for a portion of the damage you do. In a normal fight its hard to use due to creating too much aggro, but in this MGT heroic fight there is no aggro - so spam it away.

For that reason I actually find a shadow priest is the best class for that fight.

But back to the previous poster with 4 different damage mechanics, I guess it is true. I wonder if one way to balance it out would be to make the tank the most dangerous - if you don't block them.

I.e. no artificial taunts etc but aggro is still based on damage with the tank creating a bit more normally although its blocked away by the mob, while if the mob breaks for someone else the tank is no longer blocked.

I have the feeling this would make the fights too edgy though for most people, maybe MMO's will only truly be upgraded in AI when people tire of the current crop.
 
It's a hard thing. People like to feel powerful in thier games. If the mobs are too intelligent they can always act faster than you can. The day AI can fight on par with a human will be the day you have unbeatable games because of reaction time.

I'd like to see better AI on the bigger bosses and some NPC's but you need a certain amount of stupid mobs that can be owned by anyone to keep the game fun.
 
I just came back to this thread, because I realized how much I think the premise is wrong.

People ~say~ they hate Murlocs. They bitch about them, they complain, they laugh, they cry.

And in the end, the Murlocs are one of the most loved things about WoW.

Players need more challenge, they need harder AI. It's more satisfying. It can frustrate, but in the end, whether the player succeeds, or the player fails, they've been challenged.

That's one of the essences of gameplay that's been lost lately. Not enough challenge.
 
Post a Comment

<< Home
Newer›  ‹Older

  Powered by Blogger   Free Page Rank Tool