AI Allowed to Create Their Own Culture

nomoreself writes "New Scientist reports that five European research institutes are building a virtual world with about 1000 virtual (AI) citizens, in order to observe the society these artificial agents create for themselves over the course of three years. From the article: "Each agent will be capable of various simple tasks, like moving around and building simple structures, but will also have the ability to communicate and cooperate with its cohabitants. Through simple interaction, the researchers hope to watch these characters create their very own society from scratch... [further], by pointing to objects and using randomly generated "words", characters should be able to conjure up their very own language and communicate with others inside their world." One of the researchers involved thinks the dwellers of this artificial world may even develop ritualistic practices."

Comments

Is buying 5 copies of

[twilight30]

... the Sims 2 really all that expensive? :)

The test

[mabu]

The test of the realism of the sim's AI would likely involve how long it takes for one of the sims to seize power and exploit the rest.

Agents?

[SDMX]

"Each agent will be capable of various simple tasks, like moving around and building simple structures...

...absorbing other agents, usurping gigantic networks, eliminating Keanu Reeves. You know, all the things a good AI should do!

Learning about ourselves.

[Leroy_Brown242]

This has a lot of interesting possiblities.

One being that given a blank slate, what they tend to do on multiple runs. (Do they always end up the same place, or does chaos theory win out?)

Another, is that the AI could be programmed to have a pre-disposition, and see how they progress. (Homosexuality, self destruction, etc, etc)

And yet another could change their environments and see how they react. (Plague, overcrowding, etc, etc.)

Rituals

[ndansmith]

One of the researchers involved thinks the dwellers of this artificial world may even develop ritualistic practices.

What, like reading Slashdot?

Re:Culture is for Bacteria.

[rbarreira]

Combining an infinite series of light-switches will never produce conciousness.

I would like to see your proof of this. Because, combining a finite series of neurons apparently produces conciousness in humans. Have you or anyone else proved that neurons are more powerful than "light switches"?

By the way, we humans fool ourselves all the time...

Re:Culture is for Bacteria.

[kmahan]

I've got to disagree.

To make the typical politician all you need are a couple of Dimmer switches.

Re:Yep, sorry

[hab136]

Anyway, I don't understand why they'll need for a simulation like this. Will the virtual beings build a whole universe? Simulations aren't usually so slow.

Because if it ran in a week, how would they eat for the next 3 years? No, they've probably slowed down the simulation so they can "study" it as it progresses.. read: eat Doritos and play video games for the next 3 years, occasionally looking at the simulation and write something up about it.

These guys are my heroes. :)

Re:Err

[node 3]

Idiots... 30 years for a project like this? In 30 years we'll have much better methods of doing this, so any project started 3 years from now wouldn't be valid for so long.

You should put some thought into your words before saying, "Err. Idiots". They're not talking about running the same simulation on the same hardware for 30 years. What they want is for the project (not necessarily the simulation) to last 30 years (and beyond), and it's pretty idiotic to believe they'd learn all there is to learn in just 3 years.

30 years certainly seems like a long time, but on the subject of AI and alife, there's a *lot* yet to study. This 3 year simulation will answer just a handful of questions, and raise many more, leading to another simulation, and with the newer, faster hardware, and more capable software, they'll answer some of the new questions, raising yet more, and so on.

Those of you who have read Ray Kurzweil's essays probably know that there is a very good chance that we will pretty much understand how the human brain works by that time (like we understand the genome now).

Mapping the human genome (you've already noted this difference) is mere bookkeeping. It's raw data. Faster computers and newer methods sped up the project so that it was finished, it sometimes seems, before it even began. But that's just data, it's still going to take a *very* long time to really understand the data.

Take a simulation system far, far, far simpler than AI and alife--chess. Even after thousands(?) of years of study, and decades of computers aided study, we still have yet to fully explore that system--there's still work to be done, and will be, perhaps, forever. What makes you think AI and alife, which is far more complex than chess, will be so much easier that there won't be enough work to last even a mere 30 years?

As for Kurzweil's essay. He is making the case that we'll understand the brain in the same way a beginner at chess understands chess. We'll know pretty much what each part does, and how they work together in simple terms, but we won't have all the answers--there will *still* be work to be done.

Erm, Idiots indeed!

Lem-ing

[jefu]

Stanislaw Lem wrote a great story (actually a fictional book review) called "Non Serviam" (found in his collection of fictional reviews : "A Perfect Vacuum") in which precisely this is done and the scientist running the experiment eavesdrops on his artificial creations discussing the nature of God.

An excellent read (as are all the pieces in "A Perfect Vacuum").

I hate the asymmetry in news reporting

[exp(pi*sqrt(163))]

When someone creates a new project it gets press releases and news coverage. But when the project fizzles out because it was completely lame and didn't go anywhere in the least bit interesting nobody bother reporting it. It's like news stories are written in a dialect of C++ with constructors but no destructors.

Well I think it's time for the garbage collector of the news world. Someone who goes through year or three old science and technology magazines looking for projects where the leaders say things like "this technique will replace everything else" or "I expect my system to develop self-awareness over the next 18 months" and brings a bit of closure to them. If the project has failed then the project leaders need to be asked "what do you have to say about your extravagant claims?", "how do you feel about the grant money you frittered away?" and "how do you respond to the poeple who claimed you were a crackpot at the beginning?".

I'll have to put this story in my queue for re-examination in 2006.

Re:Culture is for Bacteria.

[el_jake]

Neurons do not act like transistor gates. (AND/OR/NAND etc.)
Neurons has multiple states. Active / Semi active / Sleep. Operating many connection lines in between other neurons.

The task whould be to simulate single neurons (with code) and make theese simulated neurons interact on a weighted level. You whould need at least a billion of those sim-neurons just to be equal to a insects brain. Every sim neuron should be pre-coded with a specialized task, but allso be able to take another sim-neurons task is it should be necesary.

Makes me think of john Conway and his life algo's http://www.bitstorm.org/gameoflife/