A New York Times article today (via Marginal Revolution) took note of Netflix’s contest to develop a movie recommendation system better than its current Cinematch system. The prize? One million dollars.
It seems that prizes like this are one of the most effective ways to drive innovation. Governments and scientists dislike them because they create uncertainty about the ultimate winner and the timing of payments. Prizes, though, have a long and distinguished history of producing useful ideas and breakthroughs. As the Times article points out:
BACK in the 1700s, prizes were a fairly common way to reward innovation. Most famously, the British Parliament offered the Â£20,000 longitude prize to anyone who figured out how to pinpoint location on the open sea.
Grants, which have largely replaced prizes as research drivers, have the benefit of providing funds for researchers to purchase the equipment needed for their research. Prizes, though, provide some other benefits:
These are the two essential advantages of prizes. They pay for nothing but performance, and they ensure that anyone with a good idea â€” not just the usual experts â€” can take a crack at a tough problem. Much to the horror of the leading astronomers of the day, a clockmaker ultimately claimed the longitude prize.
Although Netflix expected little quick progress (the prize was announced in October) on the algorithm contest, machine learning experts quickly jumped in and have been at it ever since. The winner must produce a 10% improvement over the current algorithm and contestants are currently up to 6.75% (Netflix maintains a leader board).
This contest notion fits right in with the current open source and user generated content trends. The idea that anyone can enter is a key aspect of the whole thing. A contest among a company’s small R&D team won’t produce the same kind of results. It’s not clear how large the universe of contestants must be and how large the prize must be to make a contest like this truly effective but common sense suggests that effectiveness increases with the size of both. Many companies offer bonuses for good ideas, but I think that’s different and less effective. The type of Netflix prize described here sets out a clear problem and has clear criteria for winning. While a good prize takes the form of “If A (solve the problem), then B (earn the prize),” corporate innovation bonus schemes often take the form of “If A, and if A2, then some level of B.” A prize works best for specific problems, although it could be adapted for specific outcomes (e.g. a new product reaching $20mm in sales 2 years after launch). The less specificity, however, the less incentive.