"Rascals" AI could pass Turing test, with a little help from Blue Gene

Researchers at Rensselaer Polytechnic Institute believe they've built an AI smart enough to pass an avatar version of the Turing test -- convince a human that it is human while conversing in a virtual environment like Second Life. The bot is named "Rascals," which stands for Rensselaer Advanced Synthetic Architecture for Living Systems. Naturally, this first-ever achievement requires some world-class juice, and the team will using IBM's Blue Gene, the world's fastest supercomputer, to run the AI in real time. The AI's thoughts about itself and others is based on a grad student guinea pig, and it can even formulate thoughts about the beliefs of others and their beliefs about the beliefs of others. Trippy, no? Rascals will be put to the test in October 2008.

[Via Slashdot]

[ArcticFox @ Mar 14th 2008 7:20AM]

October 2008, judgement day.

[System48 @ Mar 14th 2008 8:37AM]

Maybe not, they just need someone standing by the plug just in case things get out of control.

[Yemen @ Mar 14th 2008 11:07AM]

Plug huh? If its smarter than a human it can probably outsmart a human.

[Chuckles McGee @ Mar 14th 2008 1:09PM]

Best of luck

@Yemen

Any NASCAR dad can pass the Turing test...Don't fear for your safety from our robot overlords quite yet.

[Neebs @ Mar 14th 2008 2:27PM]

Don't you remember in Portal, with the phone line by GLaDOS? The plug won't work if that didn't.

[Wwhat @ Mar 15th 2008 8:41AM]

It's funny how many people and specifically americans think a cold logic would 'naturally' be all violent, but what would it be really I wonder.
I'd say that a mechanical intellect would, at least initially, see the humans as their 'doctor' or mechanics and would figure they are handy to have around.
Still, it makes you think about what drives us, and that is a predetermined set of primitive instructions, things like reproducing and gathering food and protecting your own life and annoying other lifeforms (often not included in the list but evidently real that last one), which would not apply to a robot really I figure, so if we make a real self-aware AI we better give it a deeper purpose too, and NOT killing please, not even in army projects.

[Wwhat @ Mar 15th 2008 8:44AM]

Oh and if it really had no purpose I'd say it would see humans as a way to keep entertained and would not want to get rid of the ones that do, so perhaps it would merely attack tax inspectors and accountants.

[adamb716 @ Mar 14th 2008 7:25AM]

Sounds like the next big step in robot-world domination.

[octoberasian @ Mar 14th 2008 7:30AM]

You know... giving an advanced, complicated and sophisticated AI that can pass the Turing Test to run on the world's fastest supercomputer is a disturbing thought.

If the AI manages to become self-aware, I'm making myself a bomb shelter.

[Steve @ Mar 14th 2008 7:53AM]

I guess you're mostly joking...

but is this thing really "self aware" in it's truest sense or do you think it's more likely that it's just a very convincing simulation of something that is self aware.

this is assuming that there is even a difference between being self aware and SEEMING self aware. don't really feel like getting into a philosophical conversation :P

[phoomp @ Mar 14th 2008 8:40AM]

How would you go about proving that something only *seems* self-aware?

[bondsbw @ Mar 14th 2008 8:51AM]

I think you've hit the nail on the head. The point of the Turing Test is not to show us something that is truly self-aware, but convincing enough that we cannot tell the difference. It is really more a test of humanity to distinguish intelligence than it is of computers to simulate it.

An interesting aspect of the Turing Test is that it is a continual test. Once some AI system passes it, it can (and probably will) fail it at a later time. Just like breaking cryptographic systems, we will find ways to break down these AI systems so that we can tell if they are human or machine.

[John @ Mar 14th 2008 9:45AM]

I hope one of the questions they ask involves not helping a tortoise.

[cirian1975 @ Mar 14th 2008 7:59AM]

G'ah.


Skynet is here


*Runs for the hills*

[Bill Brasky @ Mar 14th 2008 12:00PM]

Cue Iron Maiden... "Run To The Hills.... Run foooorr yourrrrr life!"

http://www.last.fm/music/Iron+Maiden/_/Run+to+the+Hills

[blake @ Mar 14th 2008 12:40PM]

the hills wont save you

[mellis33 @ Mar 14th 2008 8:00AM]

skynet is watching

[Jonathan Sundy @ Mar 14th 2008 8:07AM]

Since when do 2nd life users count as human.

The test seems flawed, convincing a bunch of idiots your also an idiot can't be that hard.

I know this test was deviced, what 40 years ago or so, but I would think it needs more structure to have any "value".

Cool and scary regardless.

[dustin3000 @ Mar 14th 2008 12:37PM]

You're.

[Johan S @ Mar 14th 2008 1:49PM]

There are many supposedly normal people (ie, without any health issues), who can't pass the Turing test.

[John B. @ Mar 14th 2008 8:16AM]

Oh shit.

[telepheedian @ Mar 14th 2008 8:23AM]

Does this mean that SmarterChild will get a brain? (shudders)

[adam @ Mar 14th 2008 8:23AM]

I, for one, welcome... ............

[slyd3z @ Mar 14th 2008 11:00AM]

This would be one of the rare occasions in which that statement would be valid. Mankind = pwned (and no, not Mick Foley)

[tcc3 @ Mar 14th 2008 8:31AM]

You’re in a desert walking along in the sand when all of the sudden you look down, and you see a tortoise. It’s crawling toward you. You reach down, you flip the tortoise over on its back. The tortoise lays on its back, its belly baking in the hot sun, beating its legs trying to turn itself over, but it can’t, not without your help. But you’re not helping. Why is that?

[r3loaded @ Mar 14th 2008 8:42AM]

W00t for Blade Runner references, but this is about computers with AI, not replicants. :P

[brokenkeyboard @ Mar 14th 2008 8:56AM]

let me tell you about my mother

[zoara @ Mar 14th 2008 11:05AM]

r3loaded got whooshed.

[Sime @ Mar 14th 2008 8:39AM]

and so the countdown begins...

[MorikaWeb @ Mar 14th 2008 8:47AM]

Out of curiosity why is it assumed any intelligent robot will want to kill people? Just because us humans cant think of anything better to do than sit around killing each other for no good reason does not mean a sentient computer will do the same.

I mean there is a whole universe out there to be explored, why would any intelligence want to stay on this pathetic planet, filled with semi intelligent creatures that can think of nothing better to do in the whole cosmos than kill each other?

[John @ Mar 14th 2008 9:33AM]

Because its always an instinct, to prove our supremacy over a weaker race, as we humans do, on other species, hence, we fear ourselves, as in fact somebody will play GOD, other than ourselves.

[kal326 @ Mar 14th 2008 10:39AM]

@MorikaWeb

Its a little thing called self preservation. If something tries to harm you, you defend yourself. If something continues to try to harm you, its must be destroyed. Once Rascals develops enough to scare us mere mortals enough to pull the plug and then realizes that we are trying to destroy it.....then its on like Donkey Kong....

[OneLove @ Mar 14th 2008 11:01AM]

It would want to play call of duty 4 instead.

[eh @ Mar 14th 2008 12:38PM]

its the idea of self-preservation will make them think they need to wipe out humans to stay alive.

[Warren @ Mar 14th 2008 4:53PM]

Because like god, we will make them in our image. Damn the bad luck.

[NewJohnny @ Mar 15th 2008 2:51PM]

Why would it want to explore? Why would it want to do anything? You're assuming it has desire.

[3lo @ Mar 14th 2008 8:47AM]

Oh, come on
No Skynet, no Armageddon, no nothing is comming...

It takes more effort, than a conversation.

The machine isnt selfe aware, because it has no routine to identify itself. Its just a simple (uh, not so simple, but you get it i guess :) coding of a database, where all the information is stored and a good code to create the correct syntax + an analyzer of the questions.
Give or take - its just a conversation program, not a AI that can be self aware or think in terms of astrophysics and philosophy...

[docca @ Mar 14th 2008 8:51AM]

"The AI's thoughts about itself and others is based on a grad student guinea pig"

I can see the quality of the dialogue:

- human: What is the meaning of life?
- computer: Dude, you trippin'? It's like we're here, and then we're not. Dude, like, err, you know, this totally sucks...

We welcome our stupid grad-student quality robot leaders!

[ePants @ Mar 14th 2008 9:02AM]

Maybe I'm too skeptic overall, but the idea of creating artificial intelligence has always struck me as, well, impossible. Sure, we can make elaborately coded intelligence-emulating software, and future generations will create even more elaborately coded and even more convincing intelligence-emulating software (and maybe the software will even learn to write /better/ software!), but it'll be nothing more than just that- a finite set of instructions constructed to mimic /behavior/ that lends itself to being /interpreted/ as having underlying intelligence. Until we develop something that doesn't run on 1's and 0's, the best artificial intelligence we come up with will be, at heart, just a very complicated (though still predictable) set of boolean algorithms.

That said, I'm still quite impressed with this feat (-or, will be once it runs later this year). While I don't think we're ever going to develop true intelligence in electronic form, the usefullness of having pseudo-AI electronics is undeniable. The ability for a program to interpret and identify the beliefs (and false beliefs- hit the read link) of others is impressive. Though, the fact that it has (and can form?) it's own beliefs does have impending doom undertones. Teaching a computer the concept of lieing will probably not go down in history as the wisest choice humans have made. ~_-

The next step for this kind of accomplishment is, of course, to teach Rascals to play WoW and FFXI instead of Second Life. Automated farming and power leveling FTW. Thanks RPI.

[zoara @ Mar 14th 2008 11:08AM]

What's to say that humans intelligence isn't just elaborate algorithms? After all, it's just neurons firing in sequences determined by genetics and past experience.

[madgamer @ Mar 14th 2008 1:56PM]

The medium of the thought (be it neurons interacting or 1's and 0's) isn't really what's important. For it to actually be self aware, it would need an assortment of inputs and a grasp of language far beyond what it is using/possesses. While it is an important leap in AI research, you are right in saying that it is far knowing what it is doing, as the Turing test is far from being the end-all test for intelligence. If you study up on philosophy of mind you'll find that there have been a number of theories that have pretty well shot down the Turing test as far as its validity in this department goes.

[Gaspode @ Mar 14th 2008 1:56PM]

You know the idea of creating intelligent life always struck me as impossible.
Sure, we can make elaborately coded intelligence-emulating creatures, and future generations will create even more elaborately coded and even more convincing intelligence-emulating creatures (and maybe the creatures will even learn to create /better/ creatures!), but it'll be nothing more than just that- a finite set of nucleobases constructed to mimic /behavior/ that lends itself to being /interpreted/ as having underlying intelligence. Until we develop something that doesn't run on C's, G's, A's and T's, the best creature we come up with will be, at heart, just a very complicated (though still predictable) set of acids.

[DonatoM3 @ Mar 14th 2008 2:20PM]

Well played Gaspode, well played.

[ePants @ Mar 14th 2008 3:30PM]

Well played, indeed, Gaspode.

Though, while your opening remark makes it seem as though you are arguing against intelligent design (forgive me if I've taken your humor in a direction not intended), the whole of it seems to be denying the very existence of intelligence to begin with. It seems you're saying we're either not really intelligent and therefore cannot have been intelligently designed nor can we design lesser intelligence, or that we were 'created' by something/someone with equally predictable and mechanically produced 'intelligence.'

Interesting.

[d0n0van @ Mar 14th 2008 9:15AM]

By the time Skynet became self-aware it had spread into millions of computer servers across the planet. Ordinary computers in office buildings, dorm rooms, everywhere. It was software in cyberspace. There was no system core. It could not be shut down.

[Jack @ Mar 14th 2008 9:16AM]

I for one blah blah blah you get the idea

[Kizorblade @ Mar 14th 2008 9:29AM]

I for one believe this can blend?
I for one believe this can play doom?
I for one believe this can play Crysis at 30+fps with max settings?

=)

[markcih @ Mar 14th 2008 9:25AM]

Waitasecond - this was tested by these uber-geeks who do not know how to interact with real people? I think my kid's Pleo could fool them.

[y3k.nik @ Mar 14th 2008 9:26AM]

it sounds so much like I Robot to me..

I hope will smith and his unbreakable arm are ready for this!

[DssTrainer @ Mar 14th 2008 9:27AM]

The bot is named "Rascals," which stands for Rensselaer Advanced Synthetic Architecture for Living Systems.

Where do they get the "c" from? Should be "Rasals". FAIL!