(This is the second post in a series written "live" from Hot Chips 19 at Stanford University.)
Vernor Vinge is best known as a science-fiction writer, but he's also a computer scientist; he retired from his professorship at San Diego State University five years ago. (I mentioned his participation in a panel at Siggraph earlier this month here.)
Vinge's talk was titled "Digital Gaia," a reference to the Gaia Hypothesis. (I see Vinge used the same title for a January, 2000 essay in Wired, here.) Vinge described several scenarios for the future of the integrated-circuit industry, building on some of the same themes he raised during that Siggraph panel.
Vinge's "slides" for the talk consist of a Web page which you can see on his SDSU site.
As embedded systems become smaller, smarter and better connected, with knowledge of their location and environment, the network of these systems becomes its own computing and communication platform--and more, its own new kind of reality--a digital Gaia. Vinge argues that new services should be designed to run on this new platform; services that try to make do with the relatively clumsy systems we use today will be at a serious disadvantage.
Vinge's future scenarios begin with "The FOOBAR Limit," in which the progress we've become used to under Moore's Law comes to an end. ("FOOBAR" in this sense is like "fill in the blank.") Vinge finds this unlikely, especially as long as there continues to be economic demand for further improvements.
The second scenario is that hardware complexity exceeds the ability of software to use it. Vinge finds this possibility more credible, and fears it may ultimately prevent practical artificial intelligence. He points out that multicore processors are already a problem from this perspective. (Some of these chips are coming up later in the Hot Chips program, so stay tuned to this blog!)
Vinge's third alternative is that there could be "wide-area hardware failures of embedded microcontrollers." I suppose it may seem strange to speak of "single failure points" in multi-vendor distributed networks, but as computers become more complicated, they become harder to analyze, and failures are certainly possible. Vinge says he isn't speaking of externally induced failures such as an EMP (electromagnetic pulse) attack, but failures of the system itself, perhaps from some presently unknown failure mode common to all silicon integrated circuits.
A fourth scenario is that hardware improvements are eventually devoted more or less entirely into computer security--stronger DRM, virus prevention, and possibly government-mandated features as well. Vinge showed a hypothetical chip block diagram in which a large section of the chip was "government property"--logic defined by the government and required to be present on all processors. As he points out, if networked computer chips define a new kind of reality, the government could begin to legislate reality--something that has been impossible before now. (He refers to Lady of Mazes, a novel by Karl Schroeder, but I haven't read it.)
Vinge's final topic was the full Digital Gaia scenario. He asked, "How ubiquitous can ubiquitous computing come to be?"
His suggestion is that perhaps integrated circuits could become a new domain of life, peers to the domains of plants, animals, fungi, bacteria, etc.
He was quick to admit that ICs can't reproduce on their own--but pointed out that most forms of life on Earth can't live or reproduce without others. Human life, for example, is utterly dependent on bacterial life. I see his point; perhaps it would be useful to expand our definition of "life" to help us deal with the future of IC technology--or at least, to help us understand and discuss it.
It's lunchtime here at Hot Chips. I'll be back later with more updates.