Dr. Castronova's first post was about how better AI can create more effective virtual communities. Everyone should agree that good AI certainly creates more enjoyable games (see Barney in Half-Life). This is probably a correlative of the fact that bad AI ruins some games. But query: does smart AI create better communities?
Of course, real-life bots are the subject of a vast amount of pop literature. Asimov is a good place to start, but golems (not gollums) have an older pedigree. Also pre-dating I, Robot is the play R.U.R., a.k.a. Rossum's Universal Robots which popularized the term "robot." More recently, you can choose your poison, but 2001, Blade Runner, AI (riffing on Pinnochio), even The Terminator flicks, all revisit Capek's questions -- if we create things that seem like us, how will they fit into our society? The recurring theme in all of the above (and look at Asimov's First Law) is a deep Oedipal fear that they won't. This is interesting stuff, but as a practical matter, I'm not too worried about replicants taking over the world or the ethics of deleting subroutines, even ones that imitate Haley Osmond.
So back to the question: do smarter bots make for better virtual communities? Take for example, the interesting decision in The Sims Online that there wouldn't be any bots. (Given the repetitive OMG! LOL! statements, I'd almost like to be informed otherwise.) In some ways, I think this was a good idea. While I am somewhat embarrassed to admit that I kind of like Clippy, I'm more embarrassed by my time-to-time Julia experiences, when the avatar you're dealing with turns out to be a cousin to ELIZA. Or even reverse-Julias, when you mistake an avatar for a bot. (Yes, I've had these too.)
So now that it is clear that we're approaching the point where the Turing Test can be passed (in virtual environments) I'm curious to see what societal role we'll allocate to bots in social game-spaces. Will MMORPGs be filled with Clippy sidekicks and virtual peanut galleries, or will subscribers feel uneasy with receiving simulated praise? Will stronger bonds be formed if players are banding together to fight a mob as smart as Deep Blue? Will the Julia problem be "fixed" with clear indicia or will line-blurring be the norm?
Virtual world bots and mobs may not raise all the questions Capek anticipated with his play about future robot revolutionaries, but they do make for an interesting prequel.
Update 10/03: Dave's article (mentioned by Ted) and Raph's slides are both good for further reading.
This new blog looks great!
Could you expand a bit on why you say "it is clear that we're approaching the point where the Turing Test can be passed (in virtual environments)?" Perhaps we already have AI-driven game opponents as "skillful" as human players, but outside of strategic gameplay itself, I'm not sure any bot yet comes close to measuring up to real people when it comes to socializing, chatting, hanging out together, that sort of thing.
That said, I think there's all kinds of interesting possibilities that game developers should push on, in regards to bots in virtual worlds. I'm curious to hear your thoughts on how bots could help improve virtual worlds, i.e., in what ways could bots make virtual worlds even more fun for players?
Posted by: andrew stern | Sep 18, 2003 at 07:45
Well, Ted's idea (I think -- correct me if I'm wrong here, Ted) was that the bots could mimic a community that actually looked like a real community. I saw this idea somewhere else recently (sorry, but I forget if it was MUD-Dev, gamegirl, or somewhere else...), where an author was advocating for a more richly textured bot society where the village folk didn't just patrol, stand in shops, say staged lines, or mill about aimlessly, but actually did things that recreated the feel of society -- e.g., working, building, interacting with other bots, etc. I think that playing in an immersive artificial community that reacts to you, has its own issues, and evaluates your actions, could make for a better game experience. I've seen this in solo games (e.g., Ultima III, Black & White springing to mind) but I haven't seen it work as well in most MMORPGs, which seem a lot more primitive in that regard.
Posted by: Greg Lastowka | Sep 18, 2003 at 11:16
One example I can think of off the top of my head are the NPPs (non-player pirates) in Yohoho! Puzzle Pirates (http://www.puzzlepirates.com/). Since the game centers around interactions involving simplistic puzzle games, the developers were able to program their AI bots to be rather good puzzle players themselves (though they sometimes do cheat a bit at the advanced levels).
With the current level of affordable basic AI we have, I would say NPC bots are a good idea if your gameplay mechanisms are simple enough that something like the above example can be implemented. However, that being said, even the most advanced ALICE bot is not going to trick a player into believing they're a real person after a few interactions.
And that's why we're finding more of the new crop of social online games (ATITD, 2nd Life, There.com) removing NPC bots entirely like the Sims Online did. Though it makes the gameworld seem a bit empty at times, players generally feel more immersed when they know everything that looks like a person is a person. However, both ATITD and 2nd Life (don't know about There.com since I haven't tried it yet) include limited AI in game objects like buildings which function much like traditional NPCs do in most online games. It makes sense. If your online banker can't talk and can only perform basic banking commands, why not just make him into a banking terminal instead and drop the NPC elements?
Posted by: Jia Ji | Sep 18, 2003 at 11:21
It may be worth observing that the Robots in Capek's play are stripped-down versions of artificial humans... like the Deltas and Gammas of Brave New World (which are stunted and conditioned so that they will prefer their lower-caste lives), the Robots we see in the play are mass-produced living beings. Capek doesn't spend much time describing the technology involved (he has a non-technical businessman explain it to a visitor). In the play, a Robots-rights activists convinces one of the scientists to modify the "formula" for Robots, in an effort to refrain from stunting the Robot soul.So, Capek's play is less about intelligence than it is about the soul. The plot may sound corny, but it was from the poplarity of RUR that this plotline became a SF stand-by.But the play also gives a glimpse into the future community that the Robots form on their own; they are insanely productive, but since the Robots themselves have few needs, and they have destroyed their Human creators, their mass-produced goods pile up unwanted. Due to a convenient plot complication, the only things the Robots cannot produce is more Robots -- thus they are motivated by the extinction of their community.
Posted by: Dennis G. Jerz | Sep 18, 2003 at 13:27
>>I saw this idea somewhere else recently (sorry, but I forget if it was MUD-Dev, gamegirl, or somewhere else...), where an author was advocating for a more richly textured bot society where the village folk didn't just patrol, stand in shops, say staged lines, or mill about aimlessly, but actually did things that recreated the feel of society -- e.g., working, building, interacting with other bots, etc.<<
Greg, I believe John Arras is the man working on that (MUD-Dev). His current project is on sourceforge as "genmud", though you'll get more out of it if you ask him directly.
Ignoring the ethics of the issue for the moment, it'd be easily possible to have a lot of robots rolling around a village or country and doing things normal people would do. Except the socialization factor. You could program attitude trends easily enough (those on the west-half dislike those on the east-half) by programming the results of those attitudes, but it'd be much harder to program the personality causes and their dynamics.
Egh, sorry if I got too technical.
So that's really a different level. It's not a community; it's a simulation of a community. Which is interesting, since MMORPGs are communities (theoretically/in some regard) and "simulations" of the real world.
Posted by: Michael Chui | Sep 18, 2003 at 21:57
Jia: Thanks -- good points. You're right. I was probably a bit too glib about the Turing Test. I only meant to suggest that *for a second or two* there could be misidentification, which is enough to unsettle me a little. Even to make that happen, much depends on the interface. Julia and the MUD bots are probably still the best candidates, but that's just because the MUD interface is comparatively opaque. If VOIP catches on, blurring the lines in MMOs will be even harder. And your terminal/banker distinction makes me realize that the whole responsive game environment is really the artificial intelligence, though bots qua bots seem to raise different issues.
Dennis: Thanks, I actually taught RUR to a high school class many years ago. I do recall my impression that Capek was struggling with social engineering as much as (maybe more than) he was thinking about technology.
Michael: I'm not extremely interested in the ethics of AI either, but it is fun to think about. I'm more interested in the technical question of what can be done with social AI and how likely it is that it might be done. It is interesting that much of the good AI in games today is oppositional (I'm thinking of RTS games mostly), but I wonder if we'll start seeing games with better social AI, even full-fledged communal AI... (Again, I'm trying to draw out the future of things like Barney in Half-Life.)
Posted by: Greg Lastowka | Sep 18, 2003 at 23:36
Something in me wants to comment that the people who would be satisfied with AI social interactions are probably that same people whose best friends in childhood were Teddy Ruxpin and Furby and who look to Real Dolls as potential life companions.
Are we training humans, just a little, to accept virtual pets instead of real, messy ones and feel warm after dealing with a virtual phone customer rep?
Posted by: BridgetAG | Sep 19, 2003 at 01:27
The teddy bear analogy reminds me of the teddy bear in AI -- which was a wonderful foil by Kubrick, because the bear was as smart as Osmond, maybe smarter -- it just didn't look human...
Speaking of non-messy virtual pets, it looks like the bots will be on their way to the previously bot-less Sims Online -- and in addition to pets, they will be performing in the role of *maids*. Capek would surely appreciate this.
http://www.gamasutra.com/resource_guide/20030916/lewis_02.shtml
Posted by: Greg Lastowka | Sep 19, 2003 at 10:53