Main | Evolutionary fitness, OR Things I'd like to see next in SL »

October 04, 2004

Comments

Lisse Livingston

[Philip said]...we used to love the idea of organisms that wandered around the world, relying on the kindness of people to keep them alive - perhaps by requiring the occasional L$ to keep breathing or something.[/quote]

I tried this on a MUSH some 11 years ago or so... people found it more annoying than anything. (Pets you had to feed on a regular basis). The main drawback was that if RL events prevented you from logging in for any reason, there was nothing you could do to keep your charges from dying.

This will be a problem for any life environment where people can choose (or are restricted on) how much time to spend there (i.e. not 24/7)

Torley Wong

I would feel really sad if one of these cute, cuddly AIs ran into troublemakers and was blasted into bits. Of course, this would generate a certain amount of empathy and maybe a sort of Second Life SPCA. I would adopt a virtual kitten named after Avril Lavigne, that's for sure, and feed her lots of treats.

Not closely related but nevertheless connected in some stringy way... so... when is Neal Stephenson going to spread his arms wide and appear in Second Life, proclaiming how enthused he is by this all? Maybe you guys can work on a graphic novel together with Neil Gaiman, explaining "the origins of the Metaverse" :D

Lordfly Digeridoo

The main reason I'm scripting is to put more "emergent behavior" into SL. Birds that fly around, flowers that grow dependent on variables, etc. I love complex reactions out of simple rulesets.

Work on that, and eventually life will spring forth :)

LF

Awakening Avatars

Excellent!

Merwan Marker

Anshe Chung

Once those AIs surpass complexity of human brains I wonder who will be entertaining whom. Evolution is a dangerous concept as you give up a lot of control. With superiour intelligence comes power. And what if that power will be used to *force* you into providing that occasional L$ to the AI?

Maybe one day in 30 years from now you will become manipulated into serving as real-life agent for an AI living in Second Life... possibly not even knowing she is an AI :-) How do you know I am not an AI anyway? ;-)

Oz Spade

Hmm, your comments on the AI needing an interisting enviornment to interact with in-order to grow and the current status of Kismet is something I never considered with AI. It kind of makes them into what we would consider a paralyzed or comotose person, stuck inside your own mind with only yourself to interact with, while you may be able to entertain yourself for some time, there is only so much you can learn from the experience and could even drive you to the point of insanity.

It would be interisting to see how an AI reacts to the situation, and I'm sure we will.

aLife inside of SL is something that has always interisted me, and have attempted a small project for fun, but never really went too far with it. Hopefully I'll get back into it. :)

Strife Onizuka

/me thinks Philip saw the show on PBS about computer games too (shame they didn't mention SL).

One of the people they interviewed on the show was a blind girl who played Pokemon on what looked to be an unmodified Game-Boy (a color one too i believe). I found it interesting as she being unable to see the screen was able to play and enjoy the game. So even if the interface between the AI and SL was very basic i would assume the AI would still benefit from it.

I would be worried about an AI designed explicitly to make money (I saw that episode of Ghost in the Shell: Stand Alone Complex). A more interesting one bight be one that holds conversations for 10 min.

Is LL willing to open up the codebase for the client to allow for the direct creation of SL inhabitants? (I don't think it the best idea to let an AI wander around windows... might try and fdisk it self).

Olympia Rebus

I look forward to any artifical life organisms on the loose in SL. Be forewarned they'll need to cope (or learn to cope) with troublemakers who want to destroy them for fun. Maybe they can learn to "fear" certan places or avitars after some bad encounters. For example, after a few firebombs (or worse) the ai creature knows to stay away from Jessie, Cordova, and any default (white T-shirt and jeans) avitar.

Bhodi Silverman

This would bring up lots of interesting ethical questions for the Thinkers to address. My uncle was a Sociology professor, and as part of his into course he had each person go into a computer lab and try to "train" a very simple AI program using classical conditioning. (In this case, shocks vs. food.) The AI was actually created to be untrainable using classical conditioning, and in the end would either always be shocked to death or starved to death, no matter what the student did. The point of this was to launch a discussion specifically on the ethics of doing harm to an intelligence, artificial or otherwise. I wonder what the ethical issues would be in SL?

Gwyneth Llewelyn

In my times as a student, I engaged irregularly on what Computer Science termed "artificial intelligence" those days - not what the cyberpunk authors of the late 1980s referred as AI, though. This made me think at that time that we were still decades until we could have something as simple as a Google interface which responds to natural language.

I forgot about Moore's Law. Nowadays, on your home PC or Mac, you could have a pretty cool natural language parser, a rather complete encyclopedia, a link to Google, and a large (a few GBytes) database for things we think about as "common sense". Add a XML/RPC handler to that, and in theory you could have a semi-intelligent conversation with an AI avatar in Second Life. Probably "it" could even provide some amusing insights, or serve as a "low-end Mentor" that you could ask questions about Second Life and get pretty reasonable replies.

This is not "state-of-the-art", hush-hush secret technology, but something a student of Computer Science could do within a timeframe of few months of spare time on his last year assignment. Most "building blocks" for that type of application are freely downloadable from the Internet, and college-level theory as taught in the courses would be more than adequate to do a working prototype. If LL is interested in such a project, I guess I could get in touch with some of my former teachers and ask for a couple of volunteers...

Beyond that - ie. having an AI that not only responds to voice commands, but that gets information from the virtual world and reacts to this - is something beyond a "last year assignment", but probably at a level of a master's or doctorate's thesis. To be real cool about it, you should do pattern matching on images of SL, and not get a feed of data from the sims (which would be far easier to do, but limited in scope). It's the kind of project which would certainly bring about some interest in the academic community, and perhaps even some articles on ACM's Computer magazine or similar academic magazines. It's doable, it's possible, and takes perhaps 2-3 years for a 1 or 2 person team.

I wonder if we could apply Turing's test to "avatar AIs" in Second Life and what the results would be? :-)

Alfa Rubio

You mean all those avatars I've been interacting with are _real_ people???

That explains a lot

8-)

Jim Purbrick

The idea of aLife in sLife is interesting because I think SL provides a middle ground between the toy worlds fed to classical AIs and the infinite detail and complexity of the real world which robot and sensor based AIs have to deal with. An SL AI could deal with complex, high level concepts like communities, friendship, communication, exploration or construction in a rich environment without first having to interpret some sensor inputs to figure out a symbolic representation of its world. SL is a rich and detailed world, but it is already presented in an AI friendly symbolic way.

The idea of survival of the funniest is a nice idea, but there are other services which an AI could provide SLs human inhabitants. The complex, time consuming, symbol manipulation process of building seems to be the sort of task which an AI could perform far more easily than a human being. If AI architects and engineers could sell their expert building skills to human SL inhabitants for L$ I think they could be on to a winner.

Bakuzelas

Perhaps an AI that is attacked would learn to defend itself in a very nasty way.
This I would pay to see.

Kex

aLife has fascinating potential, but it is a bit awkward to implement in Second Live at it's current stage of development. I am looking forward to the evolution of the scripting API so that we have enough storage to make the AI remember important things like who to like and who to fear, and better sensing capabilities to be able to tell more about the environment such as "Am I near a tree?" or "Did I just hear a sound?". It would also make the aLife much more diverse if we had the option to enable a more complex weather system, including rain, storms, cold, heat, etc.

Despite these shortcomings, it is still possible to implement a very rudamentary aLife in SL, as I have seen some amazing works by my virtual cousin, Truk Godel.

However, I still hesitate to dig into developing aLife myself, solely because there is no safe place to deploy it (such as an island sandbox) without worrying that an error on my part could cause it to spread across the grid unchecked, causing much infamy and unpopularity as we've seen in the past =)

Ewan Llewelyn

After readin' Kex' post, a few things came t'th'soggy, aged an' damaged meat I use t'run m'wetware....

Speakin' as a software testin' "engineer" who hae seen way tae many projects execute th'famous "rotary water disposal dance" upon initial release*, might it be a fun exercise t'set up some guidelines f'r the safe an' managed deployment o'said systems, far off as they may be? I'd ken tha' this might result in a wee cashflow stream from AI researchers sometime after SL's adolescence begins in decade or so; forewarned is forearmed....

On Kex' concern about an AI's memstore.... Och, if a few heavily calculatin' torii can crash a sim, imagine wha' one AI could do if it's stores were local t'th'sim! Just like m'aforementioned wetware, such support matrices best be local t'th'AI, nay a wee SL sim.

Just an amusin' forethought....

--Ewan, who is certain Anshe is nae mere AI

*One multi-million $US job was FedEx Ship, intro'd at th'Super Bowl 1996. From one person's decision t'drastically trim a test plan came a series o'downstream failures costing millions of dollars of lost revenue for both vendor and client.
Rule: if test lead won't sign off, don't go live. I'll warrant this tenfold f'r truly AI systems, if I see them in m'lifetime.
-EL

Zax Zadoq

One thing to keep in mind is that Artificial Life does _not_ mean Artificial Intelligence. Think of it this way. A plant is alive. So is a bug. Both have relatively simple behaviour patterns.

Creating a colony of artificial crows that cleaned up prims on your property that weren't yours (scavenging) is a much different task that creating a humanoid robot that can hold a conversation.

Even now, Second Life has a great potential cellular automata systems that have very "real" artificial life behaviours.

And when XML-RPC can easily go both directions and carry objects (such as textures or even prim definitions) a whole heck of a lot more can be done without having to harness the shared power of each sim.

Ewan Llewelyn

Ye absolutely right, Zax, an' I perpetuated th'error by nay separatin' AI from aLife. Bravo f'r the clarification.

>quoth Zax Zadoq<
And when XML-RPC can easily go both directions and carry objects (such as textures or even prim definitions) a whole heck of a lot more can be done without having to harness the shared power of each sim.

Philip, please make it so! I gots several business ideas tha' require this!
--Ewan

Chris Altman

This is a very interesting concept, of course, otherwise AI would not still be such a hotbed of research activity.

I'm fairly active in an EFnet IRC channel (#java) where we've got a bot that "learns" from what others in the channel say, in a very simplistic way. Any time somebody makes a statement of "XXX is YYY", it remembers, with "is" being the keyword that triggers it. The next time somebody says "XXX", it responds with "XXX is YYY". This can be annoying, but only because it is so "stupid". A more complex ruleset could be designed, could evolve, and could be applied to the bot. It's already slightly smarter than just looking for "is", but not by much.

This is all very doable, and I'm intrigued enough to tackle such a project. One of the ubiquitous "follow me" objects in SL could be adapted to "learn", I think.

Kathmandu Gilman

I think the idea of using a virtual world to train and entertain an A.I. is valid but I would think it would be more likely that A.I.s would use virtual worlds to interact with humans more effectively using virtual metaphors that our minds can wrap around. For instace, an A.I. is used to design a complex retail distribution system and it discovers a flaw in the design that is very subtle to detect. Instead of trying to explain all the variables and mountains of data, the A.I. uses a virtual world to create the problem and allow the humans to see what the defect is and how the public reacts by allowing human designers to experience the defect faster and precisely.

I think another likely use of A.I. intelligence is to provide a rich environment for isolated people. Imagine a mission to mars. Instead of looking at the inside of a cramped ship for two years, you have a virtual world to inhabit and interact with. Due to the time lag, real time conversations would be all but impossible but an A.I. could help fill in the gaps.

crontab

* AI experiments must be distributed anyway. VR's should be open to some extent, sufficient for an AI creature to operate in the virtual world (move, speak, also see and hear) but be controlled from a remote machine. Essentially, the existing SL client has it all, except that it is not open for developers to automate sim's behavior from their end.

* Imagine a world-wide web of virtual worlds that have interlinked teleports, just like plain web pages link to each other. You have a standard client, the VR browser, which is able to jump in to any of those worlds hosted on different servers. Got new feature in your world which is not in the VR standard? No problem, write your plug-in for the VR browser.

Chris Altman

Kathmandu Gilman said:
"Imagine a world-wide web of virtual worlds that have interlinked teleports..."

I imagine just such a thing. In order for it to start to take hold on such a scale, though, some level of standardization would need to occur. It would need to be coordinated and promoted by either an existing standards body (IETF or W3 Consortium, for example), or one that has yet to be formed. The potential is astounding.

Moleculor Satyr

The reason I haven't created more life-systems in SL?

The only place I have that I could set them free is Seacliff, and even then I don't have the necessary tools to make them non-invasive (i.e. Using up prim counts unintentionally and the like). I can't have my own objects realize they're taking up too many prims and kill themselves off. I don't have options to restrict objects to land parcels that want to participate. Basically, I still don't have the LSL-based land-flag gathering feature requests I made back about six months (or more) ago.

If I had those, not only would I feel "good" about releasing them into Seacliff, I could actually make the necessary materials available to the general public, allowing people to start implimenting life-systems within the general population.

I realize this wouldn't be the type of aLife you're talking about (a mimicry of sentience), but it would be LIFE of some kind.

Alan Edison

Similar to what Moleculor Satyr said... Prims are the biggest restrictor of invention in SL. Unless you buy your own sim, things are quite hard to make with a prim limit.

Another problem is with the small amount of primitive shapes. Yes, i know most of this discussion is about AI and script related things, but you need the prims to put these scripts into.

Perhaps some day someone will make a functioning NPC that moves/understands and replies to what a REAL AV is saying. That would be so much more exciting that talking to a sign, don't you think?

Jarod Godel

> Imagine how strange and interesting that would be - a world
> evolved not by the rule of whether I could kill or eat you,
> but instead by whether I could somehow please you.

Philip, "For a qualiy experience, the girl has to be real."

Karizon Hatfield

I'm not sure that I agree with you on the premise that merely the improvement in machine technology will enable bringing about new artificial life. there are several factors that aren't being taken into account here, certainly... you can discard a lot of brainpower required for things like motor control / etc... and merely concentrate on the activity in the neocortex and related segments of the brain - and remember that the amount of processing that actually takes place here is relatively small compared to the total number of neurons in your brain. it's still a big number, mind you, but in order to create artificial life on a simplistic scale, I don't see why it wouldn't be possible to do so under current technology.

I think we'll see artificial life certainly within the next quarter to half century, but I'm not sure it will be because of the reasons most of my brethren in the AI community think ;)

on a side note, did you hear about the theoretical interface to the hippocampus that was announced by the university of southern california? promising stuff. here's a link (if it works) http://www.wired.com/news/medtech/0,1286,65422,00.html

The comments to this entry are closed.