The Long Bet: Will a computer pass the Turing Test by 2029?

[bascule]

I thought this was interesting: crazy Singularity futurist Ray Kurzweil has accepted a bet from some angel investor I've never heard of:

 

http://www.longbets.org/1

 

The bet is simple: will a computer pass the Turing Test by 2029?

 

What do you think? I think it's entirely possible, but when you start talking about near-term hard dates like 20 years in the future it seems like little more than a guess.

 

If they changed the betting window 50 years (i.e. 2059), then my answer would be a definitive yes.

 

Note that this is more than just a friendly wager. If the angel investor wins, Kurzweil will donate $20,000 to the Electronic Frontier Foundation. If Kurzweil wins, then $20,000 goes to the Kurzweil Foundation.

 

I certainly wouldn't be willing to put money on a bet like this.

[GutZ]

I think there may is a slight chance that they will have figured out how to create A.I. to pass the turning test, it would be the hardware aspect will probably not be sophisticated enough to carry it out.

 

2050 seems more reasonanble.

[J'Dona]

Well, the Blue Brain project recently completed its first phase where they used a Blue Gene supercomputer cluster to simulate an entire neocortical column in real time, with results matching those of actual observations.

 

If quantum computers with capabilities along these lines are developed soon enough (as this sort of parallel processing is presumably what they're theretically best at), then as the model already exists, by scaling up the project (with multiple NCCs) we might see progress rather quickly, given perhaps some sort of business angel like Paul Allen to bankroll the research.

 

But by 2029? I think that if they have a coherent model of any sort by then, then tied with existing AI software it might be intelligent enough to persuade most humans, but still be recognisable, so I think maybe one more decade beyond that would be needed to be sure. (20 years isn't very long from now!)

 

I'll say though, that while I'm actually trying to get into quantum computing for explicitly this purpose (creating strong AI), it sure isn't for the trans-human, futurist slant of exponentially increasing intelligence to the point of obsolescing the human race that Mr. Kurzweil seems to promote.

[John Cuthber]

Has anyone checked that people can reliably pass the Turing test?

[insane_alien]

i know there are many that cannot pass the turing test. some of them have even appearedon this forum. most go to 4 chan though.

 

Note: do not go to 4chan, its full of the not so nice to the downright depraved.

[Reaper]

4chan is not that bad. At least not once you get to actually know the people there, even in /b/

[Cap'n Refsmmat]

How do you "get to know" them when they're anonymous?

[bascule]

If quantum computers with capabilities along these lines are developed soon enough (as this sort of parallel processing is presumably what they're theretically best at), then as the model already exists, by scaling up the project (with multiple NCCs) we might see progress rather quickly, given perhaps some sort of business angel like Paul Allen to bankroll the research.

 

The stated goal of the BlueBrain project specifically precludes trying to create strong AI. They're simply trying to do neocortical simulations.

 

In the near term I think tools like NuPIC provide a much shorter path towards strong AI.

[ajb]

A question. Is the Turing test necessary and/or sufficient to determine AI?

 

Will the Turing test be modified if a machine passes it?

[Reaper]

A question. Is the Turing test necessary and/or sufficient to determine AI?

 

Will the Turing test be modified if a machine passes it?

 

We can already make computer programs that can pass the Turing Test. The real problem lies in the fact that some people just don't want to admit that it really is a badly designed test. Especially since humans are known to have failed it.

 

 

This article covers all the basic problems and failures of the test: http://qntm.org/?turing

 

The quote from the article basically sums it up:

There's one other problem with the Turing Test' date=' and that is its scope. [i']All it can test for is humanity[/i].

 

Of course, we haven't been able to quite answer the question of what, exactly, we mean by humanity.

[John Cuthber]

Please provide a link that supports this assertion.

"We can already make computer programs that can pass the Turing Test. "

(except of course for the trivial cases like a machine that can emulate a person perfectly; assuming the person to be asleep, mute, unable to type...)

[bascule]

A question. Is the Turing test necessary and/or sufficient to determine AI?

 

Will the Turing test be modified if a machine passes it?

 

Sorry to dodge your questions and go off on a tangent, but:

 

The Turing Test provides a nice solution to the p-zombie problem, which suggests that there can exist things which appear to be conscious but aren't conscious.

 

If conversation with an intelligent computer program is indistinguishable from conversation with a human, then what difference does it make if the computer program doesn't contain some magical mystical metaphysical undefinable quality of consciousness that philosophers like John Searle would argue something like the Chinese Room lacks?

[J'Dona]

The stated goal of the BlueBrain project specifically precludes trying to create strong AI. They're simply trying to do neocortical simulations.

However, it is one of their stated goals to provide a foundation for whole-brain simulations, which is clearly applicable to strong AI (although far beyond the scope of their research). Their research at least shows that real-time simulation of a macroscopic portion of a brain (a 0.5mm cubed neocortical column) is possible, and by doing so it may assist in developing software (like NuPIC) to simulate the same process without as much computational power.