Tuesday, 7 July 2009

Challenge problems

Apparently a team has reached the threshold in the Netflix contest to improve quality of viewer suggestions by 10% (for a $1 million prize!):


A blog on the process here:


(and, somewhat tangentially, a nice "SVD/LSI [a method often used for recommendation systems] for dummies" article here: http://www.igvita.com/2007/01/15/svd-recommendation-system-in-ruby/ (including an implementation in Ruby as the URL suggests) )

This "bounty" system of development seems to be proving very effective and relatively inexpensive. See also the DARPA Grand Challenge effort that managed to produce an effective driverless vehicle system for a tiny, tiny fraction of what a full blown traditional DARPA program would have cost, assuming a traditional program would have managed to do it at all. My claim: This provides evidence that in a post-industrial society, and maybe in all, a gift economy is superior to a market economy for purposes of providing innovation, people valuing prestige and satisfaction of solving challenges even more highly than material gain. (While the prizes here are, prima facie, substantial, the actual reward to participants is likely far, far less than they'd have received in typical market production scenarios even if they were just being paid for their time. If one considers, for example, what a DARPA program would have paid for the person hours that the winning team alone would have cost in the Grand Challenge, I imagine it would have far exceeded the prize money actually paid out.)

No comments: