Only when we have robots which approach the cost of the materials that are needed to make them will they be economicallyvalid without mass production. Our fully automatic design and automatic manufacturing approach is not "self-replicating", but it uses software to design a machine in virtual reality to fulfill a particular purpose, and then automatically builds that machine. It isn't scary, its cost-effective.

If you say I need ten humans earning a hundred thousand dollars a year for five years to do the design of a particular robot, say a horse or a tank, and only ten of those are going to be sold, well then each of those is going to have to sell for ten million dollars in order to justify the expense of the human design. That's really where the economic problem is in robotics. By doing automatic design, by having software which actually invents, we're beginning to broach the issue that, computer chess players, have broached a long time ago. If playing chess is really a human thing and then a computer starts to do it, what does that tell us about humanity?

We think of creative work as a purely human thing, and wrap invention in mystery and legal monopoly of copyrights and patents. I've written a lot about information as property, actually, because software is the most important recent development in human culture, and its been misunderstood as a box containing just a book and a compact disc. It is a cultural solvent.

EDGE: The astrophysicist Martin Rees argues that 21st century-technology will threaten humanity far more seriously than any natural hazards. One of his concerns is that we could be wiped out by rogue nano-machines that replicate catastrophically. Any comment?

I agree with Rees and with Bill Joy that there are risks, especially when people are playing with artificial chemistries, genetic engineered species, and micro-weaponry. But I think the risk from pollution, disease, and war are much greater risks. Poverty and desperation of the dispossessed are triggers for catastrophes which could be ameliorated with games favoring the middle class.

But at the scale of electromechanics which we work in, out of control self replication is impractical. No robot will be able to eat an old fax machine to make babies. No robot will be in control of massive factories and natural resources for manufacturing themselves. Look as what's really here, not what's in science fiction books.

I do think that there is a very slight risk of networked server and routers becoming self-aware through aggregation of amazing amounts of CPU time. A phase-transition of a computer network to self-awareness is the terminator hypothesis, not self-reproducing robots with Swedish accents. We don't really understand how brains, composed of billions of neurons become coherent personalities, so we have no idea if a million servers on the internet could self-organize. I once proposed a SETI-like project to the Pentagon to watch for signs of artificial life arising in the communications and power infrastructures.

In the near term, I think the risk of Microsoft Exchange Visual Basic Viruses crippling critical infrastructure is far more serious than Nanomachines or Terminator.

EDGE: Ideas concerning computation are a moving target. Where are you on this issue?

POLLACK: The notion of computation itself is something I've worked on. John Kolen and I wrote a paper in 1991 called "The Observer's Paradox" which essentially overturned the traditional notion of generative capacity that was brought to the fore by Chomsky and the whole cognitive science movement. The question is whether or not you can infer the power of a machine's internal computation by observing the way it operates. Earlier we thought we could analyse human language, and could say something about the human mind being context-free or context-sensitive. But when you actually look at a physical system, a dynamical system, the choices made by the observer affects the level of computational complexity that the observer actually finds. This means that, in a certain way, the Heisenberg principle actually applies to the traditional notion of computation as it arises naturally. In my view, computation is everywhere, and we could look at life itself as a process which involves information and energy and matter, working together and forming state structures in very complex ways. And we understand them best through our computational notions of process and representation.

EDGE: Last year I talked about similar ideas with Rod Brooks in his Edge feature ("The Deep Question"). Here are a couple of sentences:

"Computer scientist and AI researcher Rodney Brooks is puzzled that 'we've got all these biological metaphors that we're playing around with — artificial immunology systems, building robots that appear lifelike — but none of them come close to real biological systems in robustness and in performance. They look a little like it, but they're not really like biological systems.' Brooks worries that in looking at biological systems we are missing something that is already there — that has always been there. To Brooks, this might be called 'the essence of life,' but he is talking about a biochemical phenomenon, not a metaphysical one." Any comments?