Receive the Bulldog Edition


Economic Blogosphere

Economic Journalists


economicprincipals.com banner

November 22, 2010
David Warsh, Proprietor


| contents |

The Situation Bears Watching

There was a distinct flutter of excitement at the Massachusetts Institute of Technology earlier this month when Jeff Hawkins talked at MIT’s new Center for Biological and Computational Learning. Hawkins was a co-founder of Palm Computing and Mindspring (and inventor of the Palm Pilot and Treo devices). These days he is better known for his 2005 book, On Intelligence, and the subsequent start-up, Numenta.

On Intelligence, a remarkably accessible book (thanks in part to the collaboration of science writer Sandra Blakeslee), propounded what Hawkins called a “memory-prediction” model of how the brain works – a considerable novelty at the time. James Watson compared the book to Erwin Schrödinger’s 1943 What Is Life, which put the problem of molecular transmission of hereditary information at the center of biology.  Another Nobel laureate,  Eric Kandel, of Columbia University, praised its synthesis of new and existing ideas about how the brain produces intelligence.

Three years before, Hawkins had bankrolled the Redwood Neuroscience Institute, in Menlo Park, to further his investigations of “this very complex system [that] doesn’t look so complex any more.” His memory-prediction model produced a computer technology he designated Hierarchical Temporal Memory (HTM), intended to imitate the brain, whose initial algorithms were devised by graduate student Dileep George.

Hawkins and George, along with former Palm Computing CEO Donna Dubinsky, then started the venture-funded Numenta to offer a new method of computing that Hawkins describes as “trained” rather than programmed – potentially useful for recognizing patterns in large messy data sets of all kinds, everything from faces in crowds to meaning in texts.

Charm was added to the story by the fact that Hawkins had to fight at every turn for his conviction that biological principles gleaned from brain research (“real intelligence”) should govern computing, rather than the rules-based approach (“artificial intelligence”) that dominated work on universities.

A 1979 B.S. in electrical engineering from Cornell University, with experience as computer architect, Hawkins was turned down when he applied to MIT graduate programs, and turned off when he sought to form a dissertation committee at the University of California at Berkeley.

Revenge is sweet. The Redwood think tank has been assimilated by Berkeley’s Neuroscience Institute, MIT has formed an Intelligence Initiative designed to integrate neuroscience, cognitive science and computer science, and many other universities have followed suit. The penumbra of excitement around Numenta was sufficiently provocative that Hawkins received a respectful hearing from senior faculty at MIT, who have some experience, both with outsiders who change fields in which they have not been previously trained (Broad Institute director Eric Lander, for example), and with many other wealthy and ambitious thinkers who try and fail (chemical engineering magnate Ralph Landau, for example, or computer engineer Jay Forrester).

Interestingly enough, on the floor above the Redwood Center in Berkeley last week, another genius, this one an economist, gave a job-talk paper. E. Glen Weyl, 25, of Harvard University, has been a star from his days at Princeton, when he graduated from college as class valedictorian one year and completed work for his doctorate the next. Hermann Weyl, the mathematician who was Einstein’s friend was a great uncle.

During three years as a member of Harvard’s Society of Fellows, he edited a two-volume collection of Simon Kuznets’ previously unpublished writings, Jewish Economies: Development and Migration in America and Beyond, to which he contributed a long introductory essay; and wrote a series of technical papers,  as well. An accomplished theorist,  Weyl is the hottest prospect on the job market this year.

His job market paper, Materialistic Genius and Market Power: Uncovering the Best Innovations, which is jointly written with Jean Tirole, of the University of Toulouse, is an ingenious, intricate and extremely difficult formalization of the trade-off between two great systems designed to stimulate the growth of knowledge: establishing intellectual property rights on the one hand (patents, trademarks, copyrights and so on); and, on the other, awarding prizes (for an accurate clock, a vaccine for malaria) which minimize the inefficiencies of monopoly power.

At the center of the paper is what Weyl describes as “Friedman’s Conjecture:” that it is appropriate to pay, often heavily, for entrepreneurial genius. He (and Tirole) quote from Friedman’s Capitalism and Freedom.

The great advances of civilization… have never come from centralized government… Whitney, McCormick, Edison, and Ford… no one of these opened new frontiers… in technical possibilities… in response to governmental directives.  Their achievements were the product of individual genius, of strongly held minority views, of a social climate permitting variety and diversity.

Leaving aside the obvious problems with Friedman’s Conjecture – Columbus received his sailing orders from a monarch; government directives gave rise to nuclear power, jet engines, computers and the Internet – Weyl proposes a measure of the value of “materialistic genius” associated with entrepreneurship that might permit empirical calibration. (Too complicated to go into here, it has to do with the magnitude of potential profits of the most successful innovations.) Moreover, Weyl and Tirole speculate that their analysis might be reinterpreted to apply to non-pecuniary incentives, such as collegial esteem.

Computers that think? Thinkers who compute the value of various systems of innovation? Could this be the beginning of a beautiful friendship between innovation policy and innovation itself?  Not any time soon. It will take years to evaluate the merits of Hawkins’ ideas, decades for those of Weyl.

Meanwhile, in an indication of the complexity of the problem that Weyl is trying to compute, Hawkins has put many of Numenta’s proprietary algorithms up on the Web in what he says is an attempt to jump-start an industry, and co-founder Dileep George has left Numenta to go off on his own.  The situation, as they say, bears watching.

| contents |

Post a Comment

Your email is never published nor shared.
Required fields are marked *

*
*


Skim past columns here.


Support Economic Principals by subscribing to its bulldog edition—receive the weekly via email a day before it is posted on the Web, and, as well, a quarterly Report to Subscribers.

To reach the proprietor, ask a question about the website or report a problem email warsh@economicprincipals.com.