Darpa’s Physical Intelligence (PI) Initiative, and Its Ridiculousness
I was doing some research and ran across this story again -- Darpa's loony "Physical Intelligence" initiative. The way that Wired passes along this story with a straight face shows us that the magazine, or at least this blog, the popular Danger Room, is not very reliable or qualified when it comes to science or philosophy. One may recall that I criticized the grant solicitation initiative back when it first hit the news.
There are two absolutely ridiculous elements to the initiative. First is that the text of the solicitation implies that it needs to be concretely demonstrated that intelligence is physical, or that there is any doubt over whether intelligence is or is not physical. Second is this quote:
A central tenet is that intelligence spontaneously evolves as a consequence of thermodynamics in open systems.
No, it doesn't. Only a computer scientist that has never studied a single piece of the brain would ever even say this. Intelligence is a very precise thing that evolved due to selection pressures over hundreds of millions of years. It doesn't evolve spontaneously anywhere. The above statement seems to be derived from the pseudo-mystical notion that the universe is imbued with intelligence, and rocks everywhere are just waiting to burst forth with intelligence if we nudge them the right way.
Boltzmann brains are the exception, not the rule.
November 23rd, 2009 - 19:37
I think they are probably talking about Dewar-like material:
http://en.citizendium.org/wiki/Life/Signed_Articles/John_Whitfield
The “spontaneous” emergence of intelligence does seem to have taken several billion years, though.
November 23rd, 2009 - 22:27
Whether or not the limited subset of math that defines physics will also be sufficient to define intelligence is unclear at this point. However, physics clearly defines the system in which organisms evolved.
I was talking to some members of a team from Cornell that was able to use a double pendulum to regress out from its motion some of Newton’s laws. It was suggested that computer power equivalent to 1000 core i7 processors may very well be sufficient to equal the human brain. Support vector machines, neural networks etc still have the potential to be very powerful.
Ultimately I am confident that once we understand the underlying algorithms of intelligence. It will be clear that an perceived innate intelligence exhibited by humans is merely an illusion.
November 24th, 2009 - 03:27
> Intelligence is a very precise thing that evolved due to selection pressures over hundreds of millions of years. It doesn’t evolve spontaneously anywhere.
Well, basically you’re right, except for the fact that evolution obviously is a spontaneous process, and it’s, in fact, quite “thermodynamic” (like pretty much everything else in the universe). Which basically means that in the literary sense, the “central tenet” of those DARPA guys is true (although trivial).
But from looking at their website some months earlier, I got the impression that the actual spontaneous thermodynamic evolution of intelligence has been quite different from what those guys believe it to be.
November 24th, 2009 - 12:27
It may evolve spontaneously in an extremely restricted range of environments, but what these guys seem to be saying is that evolves generally and spontaneously over relatively short timescales in (complex?) thermodynamic environments. It’s really hard to determine what they’re trying to say, which does partially lead to me determine that it’s a bunch of nonsense.
November 24th, 2009 - 13:45
Yep, it seems that they expect to come up with some rather low-level expression which in some way would “explain” the obvious fact that there is (at least, some) intelligence, basing on some choice of thermodynamic parameters forming a more or less understandable relation. If I understood what they mean correctly, they’re aiming at the wrong level of scientific description: they could just as well start with quarks and gluons instead of thermodynamics.
However, there’s always the possibility of having misunderstood their actual intentions in one way or another :)
November 24th, 2009 - 14:51
Michael, I wonder if you could define to me precisely what this “intelligence” is?
November 24th, 2009 - 16:26
Intelligence is the ability to achieve complex goals across a wide range of environments.
November 24th, 2009 - 18:07
“Intelligence is the ability to achieve complex goals across a wide range of environments.” This is the IQ / general intelligence viewpoint. I does not tell you anything about what Intelligence is only what it is capable of achieving. It is no more a definition then saying bees make honey. Intelligence is currently defined as anything that a human can do better than a machine. Intelligence is a moving target.
A route-planning algorithm is viewed as dumb yet is able to outperform most humans aside from a few savants. Humans are capable of running multiple algorithms only because of evolution. Even among humans, a large variance in capability to perform tasks is observed. Nor are humans unique in the ability to incorporate new knowledge neural networks or any updating algorithm can also do this.
Intelligence is merely an illusion. Intelligence is not a semi-mystical thing that you can only know when you see it.