Distributed Selection

From P2P Foundation
Jump to navigation Jump to search


F. Gregory Lastowka and Dan Hunter:

"Today distributed selection is an emerging reality. In various ways, distributed selec- tion is replacing the past functions of the entertainment industries by sifting through and prioritizing large numbers of works.

Increasingly, “social software” allows for the profiling of personal preferences, cross-indexing of those preferences among individuals, and thereby predicting with relative reli- ability the preferences of consumers.

Perhaps the best known social software–reliant tool is Google, which ranks the relevance of any given website by determining the number of other sites that are linked to it. As computer scientist Edward Felten has explained, “Google is not a mysterious Oracle of Truth but a numerical scheme for aggregating the preferences expressed by web authors.”34 Google fil- ters out the vast panoply of irrelevant material by collecting relevance assessments made by other users.

Capturing individual preferences and writ- ing preference algorithms that rank information’s relevance are generally known as collab- orative filtering. Analog collaborative filtering has existed for a long time. For instance, the notion of good “word of mouth” to drive up sales of movie tickets, Billboard’s listing of top singles and albums, or the New York Times’s listings of “bestsellers” are processes by which, to some extent, the public casts votes that buoy the sales of information products. But well-written collaborative filtering software can offer much more personalized and nuanced varieties of recommendation.


Distributed selection is increasingly a more reliable predictor of preferences than are the traditional industry selection agents—commissioning editors, movie executives, and so on. Distributed selection is real-time, individ- ually tailored, and resistant to the personal generalities, inconsistencies, and information deficits that plague traditional industry agents. The average selection agent makes a gut reaction decision about the interest level in a particular market or submarket. The algorithmic distributed selection agent makes individualized predictions based on the end user’s interests.


Central selection agents will lose their relative power in much the same way that the proliferation of cable television channels has led to the decline in prominence of the three major American broadcast networks. In situations in which we can actually compare centralized ex ante and decentralized ex post selection directly — for example, the ex post distributed Google search engine as contrasted with the ex ante centralized, human-selected Yahoo! directory — the distributed agent has garnered greater market share because it apparently works better. And for scope of material covered, the work of the volunteer, amateur, and socially distributed Open Directory Project is more comprehensive than the Internet directory produced by Yahoo!

Distributed networks are transforming the selection function. The conclusion is simple: Traditional centralized ex ante selection is costly and decreases total available content. Now that distributed selection is possible, ex post selection among works by decentralized agents seems to be a better alternative." (http://www.cato.org/pubs/pas/pa567.pdf)