Archive for October 25th, 2009

The New York Times recently published a lengthy article about Pandora, the personalized online music site. In stark contrast to most Web 2.0 systems, Pandora has eschewed the model of collaborative filtering and social recommendations (think Digg or iLike); instead, they manually codify hundreds of attributes for individual songs to create a database that, in my experience, is able to provide very sound musical recommendations. Have they codified musical taste?

Jonathan McEuen told me he heard about Pandora a couple of years ago and started using it immediately, “with the goal of breaking whatever algorithm they had.” A devoted music fan and a musician himself, McEuen says he did not believe an online service could understand what sort of music he would like and introduce him to new artists based on some deconstruction of his listening tastes. “You can’t just reduce it to a bunch of numbers,” he recalls thinking. “This is a romantic, emotional thing,” and Pandora’s approach to it “can’t work.”

He has changed his mind. A 28-year-old clinical neuroscience researcher at the University of Pennsylvania, he’s a listener who lacks the time to keep up with music news the way he did while amassing hundreds of CDs as a student. Sometimes he runs Pandora as background music; sometimes he’s more engaged, using it as a way to learn about contemporary classical and opera — and as a result has become a fan of the music of a young composer named Eric Whitacre. “I don’t know how else I would have found out about it,” he says. “Except through the exhaustive process of making new friends on the Internet. Which is something I’m kind of loath to do.”

Have they succeeded in quantifying musical tastes through their labor-intensive algorithmic process? Is there a lesson for all the other realms in which people are attempting to quantify human experience?

They may have been successful, but I think there are limited lessons. The reason is because the cost of failure is so low. If they recommend a song that I don’t like (and that doesn’t happen too infrequently), all I need to do to fix the situation is click the ‘Thumbs Down’ button and it ends. In fields like economics, the thumbs down function is far more costly.

Update: Via my friend Alex’s Twitter feed… an algorithm that predicts music success with 80% success:

So far HSS programming boasts an 80 percent success rate, classifying tracks such as Outkast’s “Hey Ya!” and t.A.T.u.’s “All the Things She Said” as potential hits, according to a 2006 study from Harvard Business School. That compares to a 10 percent success rate for songs promoted by record companies as singles, according to the study.

That’s certainly a high rate of success, but:

• Though impressive, that rate of success, it would seem, is far from acceptable for scenarios where failure is far more costly to society.
• Does it evolve? Would it have predicted unique new sounds like M.I.A. or just boring stuff like Drake? Is the evolution of music what’s found in the 20% error zone? This is especially important when you look at things like GDP and see that they have largely been the same since the 1940s, despite its increasingly detrimental effects.
• I believe it was Mick Jagger who once said that the key to the success of a song was exposure. Science seems to back him up: the exposure effect posits that experience with something increases likelihood of liking it. So, if musical tastes are constructed, not given, how does the algorithm account?