Could WoW be a good AI research point?

25173620a178828c4af639e495d588be
0
onyxthedog 101 Feb 03, 2008 at 00:03

Could bot building be a good research point for AI? They are fairly complex worlds, thus the ability to give them more complex virtual brains. Some colleges are already starting to look into this with, I think Unreal Tournament, a FPS, but the skills are limited.

In contrast to this games like Runescape, Guild Wars, and WoW are all full of skills that need to be arranged and worked on in order to get an advantage in a specified area, not necessarily combat. In these games you can try to get an advantage in the economy, in combat, in quests, or just in overall rankings. As you can clearly see, there are much more options to choose from than an FPS. This was just an idea I had about a week ago while watching this video. It describes evolutionary programs in AL (Artificial Life), called Polyworld. One thing he says is that the reason the NNs (Neural Networks) were not evolving enough is that they have relatively simple worlds.

6 Replies

Please log in or register to post a reply.

46e3aeb425a3801138dc3f04e4a9acb5
0
Seliris 101 Feb 03, 2008 at 08:08

Absolutely, MMORPGs have arguably the BEST environment for AI development! Just as you said its because of the plethora of various “play” areas and the vast difference between what one agent (creature) can do and what another cannot.

However, in their current state I think that games like WoW and EQ2 use FSMs for three reasons…
1) Predictability/Ease
2) Processor Power versus Server Cost
3) Content Development

In argument 1, I don’t believe players really want emergent behavior in their encounters. Granted I only played WoW briefly, but alot of the players depend on forming a single strategy to take a raid on and they don’t want suprises. I suppose that might be because if a raid fails, it does’nt just hurt the ego of the leader but the whole of 25-40 people can be impacted. Since they’re paying monthly, as a business move that wouldn’t be very smart to just piss the players off :) On this point a little further, players who aren’t a part of an elite first-encounter group, have the ability to study another guilds strategies so that they may more easily brisk through it. If you’re looking at it from an academic stand point, then this is completely irrelevant.

In argument 2, this is an obvious reason. There are hundreds, if not thousands, of AI agents being generated and controlled by a server at any given time. FSMs just allow the CPU to plow through a few statements and determine a behavior, rather than work with something more complex like fuzzy logic or a neural net. Obviously, as processor power increases and server costs decrease I imagine this will become a moot point.

Lastly argument 3… this is pretty straight forward. I just don’t believe that after looking at my first 2 statements that producers see much need to invest in an AI programmer like Jeff Orkin. They’d rather focus their efforts (again theorizing here) on developing artwork, stable clients on every possible piece of hardware, server and network code.

Hopefully, I’ll continue on the course I’m on. I always thought an MMO that focused on AI would be more interesting then any of the existing ones. Since the game is already persistent, it would just help reinforce that illusion of another world. Imagine seeing a human child NPC one day, and in a few weeks you come back and he’s grown up and after exploring the world has picked up a whole set of skills. On another server, he may not grow up (gets killed) etc… This sort of behavior would work really well for cities/towns.

But how would you handle boss encounters or dungeons? Might be interesting to have a dragon that memorizes how players encountered him last (and does he carry that memory to his next respawn?)

So many questions to ask after that point :)

25173620a178828c4af639e495d588be
0
onyxthedog 101 Feb 03, 2008 at 19:13

Those are all very good arguments, but I think that you misunderstood what I meant. I meant that getting special permission from the designers to make a character that is automated character, that could use neural nets to make it understand the game and how to play it to the maximum that it can. But I realize, in the context that you were using, those are all legitmate concerns.

46e3aeb425a3801138dc3f04e4a9acb5
0
Seliris 101 Feb 03, 2008 at 20:24

@onyxthedog

Those are all very good arguments, but I think that you misunderstood what I meant. I meant that getting special permission from the designers to make a character that is automated character, that could use neural nets to make it understand the game and how to play it to the maximum that it can. But I realize, in the context that you were using, those are all legitmate concerns.

Do you mean to write a bot that would behave like a player? Not an agent run on their server?

Sure… they’ve already got bots that use really simplistic (and stupid) AI. So the ability to thread through the data being sent is there, and to be able to return keystrokes/input is also available.

25173620a178828c4af639e495d588be
0
onyxthedog 101 Feb 03, 2008 at 23:01

@Seliris

Do you mean to write a bot that would behave like a player? Not an agent run on their server? Sure… they’ve already got bots that use really simplistic (and stupid) AI. So the ability to thread through the data being sent is there, and to be able to return keystrokes/input is also available.

I know they always annoy me when I am playing online. I was thinking to see how far we can get them to play the game like a real player/make intellegint decisions. Like see how they might be able to adapt to different situtations. (I.E. Economies often change.) One thing is to decide through experience how the best way to make money would be, not based on a price guide alone, but rather look to see the average price and then look for the amount of demand, and make a decision based on that.

3612bf90aaf71b238da89b1934391b25
0
Sekm 101 Apr 27, 2008 at 09:33

This sounds interesting. I am still studying through my CS degree right now, but eventually I want to end up working primarily with AI. I think that what you are suggesting could be valid, although painstakingly hard (if possible at all). Also, it would be limited to interacting with the server, I dont see this AI looking for groups for h.MgT or running a the flag in EoTS anytime soon lol.

25173620a178828c4af639e495d588be
0
onyxthedog 101 Apr 28, 2008 at 00:41

It may just be because I came up with the idea, but I really like it. There are three major problems that I see:

1) The sheer programming time and skill required to create such a “player-made bot”.

2) I am not sure but I know it would take a lot of processor power.

3) I doubt that the game administration and the player community would go for the whole idea of a bot running around in their game that would actually be smart enough to be competition. (Most games anyways.)