Games AI seems most associated most with single-player computer games. This is because with single-player games there is no social dimension to highlight (it is up to only the AI for dynamic content). However, AI has an important role yet to play in social "massively multiplayer" spaces, I believe. Consider the features readers of MMOWatch thought were important in RPGs (see site poll) as illustrative. In order of importance (at the time of my reading): "NPC interaction", "Character development", "Exploration", "Combat" (only 5%), "Puzzles", "Questing"...
A call for papers is out (AIIDE '08), details below. Readers here might take this time to suggest to those considering writing what they think are the important qualities of entertainment AI in MMORPGs.
The Artificial Intelligence and Interactive Digital Entertainment Conference (AIIDE '08) is soliciting papers and presentations on research and applications involving artificial intelligence in computer games and entertainment. The conference will take place on October 22-24, 2008, at Stanford University in California, USA. Paper drafts and talk proposals are requested before April 22, 2008. More details can be found at the conference website: http://www.aiide.org ...
This year's technical program includes a new Industry Track strictly for presentations in addition to the more traditional Research Track. Industry professionals who do not want to submit an academic paper are invited to propose an Industry Track presentation by submitting short abstract.
AIIDE is hosted by the Association for the Advancement of Artificial Intelligence (AAAI), and is the premier conference on artificial intelligence in computer games and entertainment. It brings together technical leaders of game development and research to examine how artificial intelligence can improve games and entertainment, exchange information about new developments in the field, and share experiences in deploying AI in commercial products. The conference will feature invited talks from prominent industry professionals and academics, a full technical program, and a lively poster/demonstration session.
The speaker list and conference program will be announced in the summer of 2008. More information about AIIDE '08 can be found at: http://www.aiide.org
"This is because with single-player games there is no social dimension to highlight (it is up to only the AI for dynamic content)"
I guess I'd quibble with the first bit of this. The distinction between single player and multiplyaer seems to be shifting fast. Though I think the second part of what you say probably does put the finger on what is different between the genres.
That is dynamic in-game content. As opposed to social dimensions. As with the advent of integrated social networking like technology in to so many games now and networks that sit around these games, online social elements of the practice of game playing, as opposed to game-play per se, seems at least to be an option these days.
Sorry that's a bit off topic.
Posted by: Ren Reynolds | Apr 02, 2008 at 07:42
The resource requirements for heavy duty AI are so large that it's just not an option for most virtual worlds. WoW has its mobs run on rails most of the time rather than use a route-finding algorithm: what would having mobs with decent AI do to their cpu load?
Richard
Posted by: Richard Bartle | Apr 02, 2008 at 13:47
What should be done here depends a lot on what you consider to be 'NPC Interaction'. Mass Effect-style dialogue trees? The Sims style natural use of their environments, replete with needs and desires? Gears of War quality of AI for squadmates and enemies? All are drastically different gameplay paradigms, with drastically different costs.
Posted by: Damion Schubert | Apr 02, 2008 at 14:33
Someone (Mike Sellers?) used the term "artificial personalities" to try and differentiate AI (the ability to actually learn) from making NPCs more than just "ye olde vending machine".
Posted by: Mike Rozak | Apr 02, 2008 at 19:05
We've used the term "artificial psychology" and sometimes "artificial personalities" for computer-driven agents that relate, learn, etc. We've also referred to this as "social AI" (apologies for all the "quotation marks" :) ). That is, when we've talked to people about AI, most of them go right to pathing, combat maneuvers, etc. Good and necessary stuff, but not what excites me about AI. To me, social AI is the necessary underpinning to make the social network meaningful within the game: being able to include the NPCs as part of the social landscape makes it more engaging and less something that's just bolted onto the side of an existing game.
FWIW, available CPU is always an issue, but much less so than it used to be for online games. I don't believe CPU is the reason that most NPCs are vending machines; it's more the case that engaging NPCs are phenomenally difficult to create with existing technology, and that's not been the gameplay focus for most AI anyway.
To reel this back in, I hope to be talking about a lot of these issues at AIIDE this year.
Posted by: Mike Sellers | Apr 03, 2008 at 16:41
Richard said: "The resource requirements for heavy duty AI are so large that it's just not an option for most virtual worlds. WoW has its mobs run on rails most of the time rather than use a route-finding algorithm: what would having mobs with decent AI do to their cpu load?"
I wonder how much of the CPU load could be done by the client? Shop keeper responses could possibly be done on the client side. Any content where you know you're only going to have one player interact with the environment could be driven by client side CPU. Only the final results would need to be sent back to the server.
Posted by: SVgr | Apr 04, 2008 at 15:51
1) basic security - I think it would be difficult to design a schema where the client could process AI reactions and there wouldn't be a danger of abuse from hackers who figured out how to jinx the system for themselves that didn't bog down the server with data and sanity checks (erasing the CPU gains from shifting the AI to client side)
2) basic philosophy of most MMOs: Most of these virtual worlds have a critical paradox at the heart of their narrative: The protagonist is doing something for the first time that has been done before by a bunch of other people (who were also doing it for the first time). Because of this, MMOs tend to have an odd stateless quality because nothing is ever permanent. For this reason, there's no sense in having NPCs remember a protagonist, with the exception of faction hits or the individuals progress along a specific quest line.
And in my experience, the NPC is always in the exact same place. If they aren't, something is wrong and you either /petition or wait for a respawn.
This is one of the things I hate about MMOs. But God, if they changed it....
There's two problems with this idea for MMOs:Posted by: Peter Hanley | Apr 11, 2008 at 17:15