Scott Hotes: December 2007 Archives
More interesting, perhaps, is this series of rebuttals to this tired assertion by The Reality Club at edge.org. Choice quotes include:
Alas, Davies also brings up the anthropic principle, that tiresome exercise in metaphysical masturbation that always flounders somewhere in the repellent ditch between narcissism and solipsism.-- PZ Myers
Science is, indeed, founded on the working hypothesis, one amply borne out by four centuries of scientific practice, that the world, or at least some aspects of it, is ordered in a stable and intelligible way. But that tentative and partly testable working hypothesis is a far cry from religions' reliance on sacred texts and personal revelations. To characterize these radically dissimilar endeavors as both based on "faith" is to point out a superficial commonality while obscuring the fundamental difference. And at a time when humanity is wracked by conflict between incompatible versions of faith, in the genuine sense of the term, to muddy the distinction between religion and science is worse than philosophically misguided: it is irresponsible.-- Alan Sokal
A few years ago, when I read the recently released On Intelligence by Jeff Hawkins, I was, to say the least fascinated, if not downright inspired. Jeff Hawkins, inventor of the Palm Pilot and Handspring Treo, had codified and clearly explained recent results in our understanding of how the brain works, and in particular how memory could be modeled in the neocortex.
In the book, Hawkins not only provides an overview of recent work, but combines and extends these ideas into a concrete hypothesis for effectively modeling neocortex brain function. These ideas have been formalized into a whitepaper. Hawkins published the book in 2004, and a year later formed a company called Numenta (along with long-time business partner Donna Dubinsky and technologist Dileep George) to help translate these ideas into reality.
When I first heard about Numenta, I was skeptical about its viability. It's certainly a noble goal to make practical the theoretical ideas outlined in the book, and nothing would lend more credibility to these ideas than commercial success. However, the ideas/concepts seem to be fairly low level from the perspective of building commercial solutions. Say, for example, that I build a commercially viable image recognition system based on the toolkit provided by Numenta. Is it likely that a great deal of IP would be attributable to the toolkit? Or to the concepts underlying Hawkins' program (something called HTM)? If it's the latter, then really Hawkins is commercializing the concept (I have doubts about the former.) HTM can be defined as a type of Bayesian network. It's not clear that patenting these concepts would not run afoul of previous academic research into brain function (recall that really these concepts were known previously, the contribution here was aggregating them into a practical whole.)
In any case, in a recent press release, DARPA has announced that it has awarded a $4.9M grant to Lockheed Martin to develop just such a program. This brings great credence to what Numenta is doing, and outlines a fairly clear path to success. Now more than ever, this book may be worth a read!
