The future of AI

IMO, the only reason automation would NOT result in better living conditions for humans is because of greed and capitalism. Look at us now - so many of the jobs once filled by human hands are now being done by robotic devices, yet, that percentage of work has not been evenly distributed amongst the world; If robots are now doing 10% of the work we used to do, wouldn't that mean we would only have to work 90% as hard? You'd think, but studies show that we as people now work harder than ever before, yet who is reaping the benefits of the engineering marvels of the past 100 years?... The small population of upper management and owners of major corporations who didn't actually DO any of the work or invention.

I am interested in AI not as a business solution or some sort of 'better' replacement for humans. I am interested in AI because I am interested in the human mind, and what makes it tick. I ponder possibly 'better' AI systems not for any sort of commercial advancement, but for the sole purpose of intellectual pursuit. I believe AI and psychology are very closely related, and if it were not for the greed that fuels the world, AI may be able to actually help us.

If anyone wants to help me build a brain, there's a project here:

http://artlearn.sourceforge.net

As you can see, it's sortof hibernating since I found SL

Hehe, I'd love to Azelda but I can't even get the open-source "chatterbot" AI that I've been playing around with to connect to SecondLife yet...

More on that at this thread...
/54/76/29113/1.html

Interesting thread, everyone makes some really good points... Ama and Eggy, you hit the nail on the head, I had the exact experience you're talking about when I ended up creating the above post - It is an easy idea, seemed like the simplest thing in the WORLD to me when I started last night... But in reality, turned out to be much too difficult of a project for me. Heh, story of my life...

I think that we will have increasingly functional AI, and I also think that yes, while it probably won't happen for another century at the very least, EVENTUALLY we have to get that AI to a point where it will at least be intelligent enough to reach the point of "singularity", meaning if I were an AI and capable of learning and modifying myself the first thing I'd do? Make myself smarter, improve myself... And then, THAT AI can tell Sean and I what the grand unified theory is.

Nice quote Where'd you get it from?

Also .. who is this ellen we are referring to?

I for one welcome our new skynet overlords.

Last year, during that 24 hours stunt of mine, I met a guy at about 3AM who was building an artificial life program with mice and cheese and cats who ate mice. The idea was the mice would learn to avoid the cats after several hundred generations. Anyone know what became of that? (I forget his name, and because SL recently re-archived the entry, I can't find it right off.)

That sounds like a neural network; It's fairly common practice these days to train up a neural network with tons of test data and then use it on real-world data to make predictions & analysis.

-Adam

I think that was the very clever Rhysling Greenacre.

A good summation of the arguments would be to look at the long bets foundation's inaugural bet. http://www.longbets.org/1

Edit: I should probably also point out that Mitch Kapor (one of the parties of this bet) is one of LL's board members.

One thing I have to wonder about the Turing test is... well, where does the AI get its knowledge from? To be fair, humans have years, or decades of learning behind their intelligence... it's hardly a case of 'Let there be Mind' and poof, you have a sentient human being. Heck, I doubt we're even self-aware when we're born.

So would it not be reasonable to expect that, for a computer to pass a Turing test against a human observer, it would need at least as much life experience as an average human, as a human?

I certainly hope human-like artificial intelligence is possible soon, 'cause I'm one of those fringy trans-humanist types, and dammit, I want to get uploaded as soon as possible.

Texas.

No, really: //www.cyc.com

This AI actually did make an ethical observation on the contradiction between the fact that most religions espouse peace, but most wars had been fought on religious grounds!

> One thing I have to wonder about the Turing test is...

Turing Test is invalidated as a test of artificial intelligence for something along these lines.

Essentially, its easy to detect a computer because you say "So, just wondering, what is the square root of 2354653235 * 2352356646426 ?" and the computer will reply, instantly "23535046604.988611432898411086459".

Of course, you could build in stuff to make your computer pretend to be dumb, but the point is: an intelligent computer can easily fail a Turing Test because it is *too* good at certain things. Does that make it unintelligent?

Azelda

Ah, Azelda... well then you call up a fave topic of mine which comes to mind: human savants and people with such mindboggling mental skills, who are very "computerlike" in some regards and can do all sorts of amazing mathematical wonders in their heads, sans calculators.

Michio Kaku's overseen book, Visions, has an accessible section devoted to AI and the different approaches (i.e. top-down vs. bottom-up, I think they were called) used to "learn".

One thing I've often laughed at, in video games, is ways to muck up "enemy AI" and see if you can get your foes repeatedly crashing into a wall or performing other dumb, repetitive actions.

> One thing I've often laughed at, in video games, is ways to muck up "enemy AI" and see if you can get your foes repeatedly crashing into a wall or performing other dumb, repetitive actions.

Yes, AI has a pretty bad name right now, because of things like this. Its called "rule-based AI", and it has to be this way because computers simply are not fast enough. Rule-based means a human thinks of various rules that you can apply in order to play the game ok-ish, and feeds them into the computer. The computer obeys these rules.

Whilst rules have their place, to be really intelligent, a computer needs a couple of things:
- something more akin to "intuition", which is where things like SVMs and neural nets come in
- the ability to learn; again SVMs and neural nets

For the record, I'm not really a fan of neural nets. They are very effective, but you cant actually "read" them, you never know quite what theyre doing or why. This is not such a bad thing, as far as intuition etc, go, but SVMs provide a much more intuitive output, and are at least as powerful.

As far as speed, computational complexity, well, the actual computational complexity of the brain is tricky to estimate. According to the extent to which you consider neurones are quantized, you can get estimates up to 2 orders of magnitude different (100x different).

However, many estimates run like this:
- in 2010, a 2000usd pc will have the same computational capacity as an ant
- in 2021/2022 a 2000usd pc will have the same computational capacity as a human
and, of course, with Moore's Law, to the extent that it holds, then if in 2021 a 2000usd computer has the same computational capacity as a human, then in 2031 it will have 1000 x, and in 2041 one million times the computational capacity of a human!

A difference of 2 orders of magnitude is about 6-7 years with Moore's Law.

As far as the Singularity, basically the concept is that Moore's Law - exponential growth of power - is based on a constant quantity of research resources, since the number of humans is not changing really fast.

However, as the intelligence of computers approaches that of humans - and then exceeds it - then the computers themselves can contribute to the development of increased computational capacities etc. Thus, the quantity of research resources will no longer be constant, but will itself be increasing exponentially!

Double exponential growth -> Singularity.

Azelda

I read someplace, not too long ago ( I wish I could remember where ), some estimates on the computing power of the human nervous system. Apparently, the optical pre-processing done by our retina puts them at about an order of magnitude more powerful than our most powerful desktop computer. Mighty, mighty retina.

I always thought the Singularity was based on the development of technology that enhances intelligence, which would of course send things into the most ridiculous sort of exponential growth.

I remember getting into argu... er, discussions, over on the MOO3 forums, about the possibility of true artificial intelligence. There were people taking the stance that it was utterly not possible, placing the human mind on some sort of weird, special pedestal. That drives me nuts.

> I remember getting into argu... er, discussions, over on the MOO3 forums, about the possibility of true artificial intelligence. There were people taking the stance that it was utterly not possible, placing the human mind on some sort of weird, special pedestal. That drives me nuts.

Yay, finally I meet someone who believes in Strong AI Most people have a lot of trouble seeing beyond that box on their desk. Even at ECML/PKDD, almost everyone I spoke to didnt believe in Strong AI. I really dont see why there should be any dispute really. The brain works, worst case we just a brain from neurones. The only possible dispute is whether there is something in addition to the brain, the whole brain as a transmitter to the soul thing.

Azelda

Hehe, I personally am dubious of the whole 'soul-antenna' theory... even though I did arrive at an identical theory independantly.

I guess the thing that really bugged me about their view was that, well, the human brain operates just peachy keen, and it's constrained by the laws that govern all matter and energy in the universe. So obviously, it's possible for an intelligent computing system to exist. So... there's no reason we shouldn't be able to make one, ourselves, right? I mean, hell, we could build a galaxy, if we wanted to. I'm not saying building a galaxy is feasible... just possible.

What is "Strong AI", exactly? It's not a term I've encountered, I don't think.

Good explaination, Azelda, I had heard bits and pieces but that's the singularity concept I was referring to, explained much butter than I ever could. I just hope the conservative estimates are wrong, and that it happens within my lifetime.

I was kinda shocked by the ant analogy, then I read Cross's comment about the retina. DAMN, wow, at least that explains why I don't see pixelation while looking out my eyes.
I guess, with me having studied so little anatomy, I underestimate the complexity of us walking slabs of meat...

GO GO TRANSHUMANISM!

"Singularity", like "Nanotechnology", is a buzzword that is abused repeatedly and glossed over simplistically. There are, in fact, opposing schools of thought concerning matters such as these. I don't have a problem with the words themselves, but I do have a problem with certain brash people believing the whole house is done, when in fact, we are still laying the first bricks. I'm all for optimism and a Diamond Age but that time has not yet come to pass. Nor are we flying around unaided.

FWIW, I thought of the concept of the Singularity independently of Vernor Vinge. About 8 years after him but without hearing the concept from anywhere else beforehand. I didnt call it Singularity (what a cool word!), I called it something like "exponentially explosive growth in computational complexity"

There are opposing schools of thought, but mostly the future surprises people in ways they never dreamed of.

The funny thing is: when the future happens, most people never give it a second thought. I still can look at a mobile phone and think "Whoa! From pretty much anywhere I am, I can just touch a couple of buttons on this device, and talk to pretty much anyone, anywhere in the world, at any time!" I mean, that is incredible! And its not just rich people that have these, its *everyone*. Its unemployed people, its people in China. I mean, I saw this street vendor here, looked like he probably slept in a box or something, walking along the street with his goods on a pole. And his mobile phone started bleeping

Ditto for digital cameras. 5 years ago, people laughed at me when I mentioned digital cameras.

When real computer intelligence actually happens, people will talk about it like its been there all along, that it was never in any doubt. There'll be like "yeah, but these arent really computers, cant you see that theyre intelligent?".

Azelda

Pretty much all homeless people and beggars in the UK appear to have mobile phones. It doesn't help me want to give them any spare change when they can afford something I don't even have because I can't honestly justify the cost of ownership.

Which invalidates your assertion that everyone has one. Cuz I don't. So nyah!

(Look, the forums are just shite atm and trolling material is pretty lean. What can I say?)

> (Look, the forums are just shite atm and trolling material is pretty lean. What can I say?)



Azelda

Heh, Azelda - I'm also a believer in strong AI. The reason most people dont, simply is based out of religious or superstitious grounds, often from the fear of being replaced or surpassed by a creation. (see early fears about computers for a brilliant example.)

-Adam

I don't have any comments to offer on "The Future of AI". But at the urging of a couple friends, I'd be happy to post a bit of information here about some SL Artificial Life and Artificial Intelligence development going on in the present.

For the last month I've been offering occasional Genprim classes to show off the Genetic Intelligent Objects I've been developing in SL. The description I've been using in recent event posts is:

"These SL objects are controlled by an internal DNA governing how they look, behave, evolve, interact, sound, breed, and communicate with each other and Second Life residents. The cute mutating critters are the building blocks for interactive performance art and evolving behavior experiments."

There are no Genprim classes scheduled for this week, however there is information about them available through a free SL group named ASLI: the "Artificial Second Life Institue". The ASLI charter reads:

"ASLI is open to all who share an interest in developing Intelligent Objects. We pursue Genetic Programming, Chaos Theory, Cellular Automata, Natural Language, and Artificial Intelligence. Sound scary? Naw... Come check us out at an exhibition event to find out more. We're mostly harmless, and our objects can be a heck of a lot of fun."

Feel free to contact me if you're interested in finding out more about the Genprims or other intelligent objects being developed, in receiving announcements on upcomming Genprim classes, or in joining the ASLI group proper and/or it's sister group "ASLI Chat" for casual conversations about AI. Just send an IM to "Rhu Altman" in-game and request a copy of the "ASLI Gazette". I'd be delighted to send you the latest edition.

Thanks!
--Rhu Altman

*reads definition of 'Strong AI'* Okay, yes, I believe in Strong AI.

Has anyone read any of Greg Egan's novels? He seems to like to explore a lot of the facets of existance as a simulated consciousness, which is strongly tied to concepts of artificial intelligence. He has a tendency to fall into incomprehensible mathematical exposition from time to time (he knows what he's talking about... I just wish I always did ), but the ideas he explores are fascinating.