Mon, 06 Nov 2006
Olin alerted me to a website called pandora.com (update – no longer available in Canada). Maybe you’ve heard of it, or maybe you’ve heard of the Music Genome Project, which is the organisation that runs the website. The idea behind the Music Genome Project is that they take popular songs and break them down according to their musical characteristics – rhythm, melody, orchestration, vocal style, and so on – producing a “genetic map” of each song. So far they’ve mapped tens of thousands of songs. The deal with Pandora is, you go to the website, type in the name of a favourite artist or song, and their computers generate a mix of other songs from their database that they think you’ll like. Then they stream the mix to you for free.
Because the Genome Project looks only at the characteristics of the songs themselves, rather than what style of music the performer usually plays, you get exposed to artists that you’d probably never seek out on your own. For instance, my starting point – the musician I entered when I first logged in – was Aimee Mann. Usually I get an Aimee Mann song in the mix every hour or so. Some of the musicians that come up I could have predicted, like Elliott Smith. Other stuff comes as a surprise – it turns out I like the Cardigans and Marianne Faithfull. Most of the bands are completely unfamiliar to me.
But it’s not exactly foolproof. You tweak your mix as you listen, giving thumbs-down to songs you dislike (which removes them from the mix) and thumbs-up to songs you like (which tells the computer to play more songs like that). But you have to be careful. I gave a thumbs-up to a Liz Phair song and spent the next half hour thumbs-downing shiny female pop artists like Natalie Imbruglia and Jessica Simpson. Cos, you see, there’s one musical characteristic that the experts at the Music Genome Project have neglected to incorporate into their genetic map: quality. Superficially, Aimee Mann has quite a bit in common with Alanis Morrisette, who has popped up several times in my mix – they both sing catchy mid-tempo pop songs about failed relationships. The difference is that Alanis’ lyrics are usually pretty woeful. A million superficial similarities won’t compensate for the single highly important distinction between good and bad. Which is why I’d be willing to listen to an Aimee Mann song even in a genre I didn’t care for, like rap or heavy metal, because I know she’d do something interesting with it; whereas no matter how many catchy mid-tempo pop songs Alanis Morrisette writes about her disastrous teenaged love affairs, I’ll probably never become a fan.
Still, I’m kind of addicted to Pandora because it’s responsive to input, even if it’s not always responsive in the way I’d like or expect. I’m listening to it right now. For the last couple hours I’ve been stuck in a ghetto of folkie girl singers, so I’ll have to start punching the thumbs-down button pretty soon. But earlier this morning I was being flown over some pretty diverse musical territory – from Marianne Faithful to Donovan to Broken Social Scene to Richard Thompson.
While I was writing this, Norah Jones drowsed by, which reminded me of a web video I watched a few days ago. It’s a talk – really, a meander – given by Malcolm Gladwell at a conference hosted by the New Yorker magazine. He starts out by relating how a computer program predicted the success of Norah Jones a few years back. Then he proceeds to describe an outfit, similar to the Music Genome Project, which breaks down screenplays according to attributes like setting, characters, conflict, etc. By analysing the screenplays, these consultants are able to forecast the box office returns for the completed movies with remarkable accuracy. The premise is that with this technology, Hollywood studio executives will no longer have to rely on their gut instincts when greenlighting hugely expensive movies – they can feed the script through a computer and find out instantly whether there’s any chance of the film making a profit.
So Gladwell brings these guys the script for the forgettable Nicole Kidman / Sean Penn flick The Interpreter, which earned only $50 million or so, and asks them to use their technology to turn it into a blockbuster. The consultants spirit the script away to their secret laboratory and, after applying a few algorithms, they return to assert that by adding a helicopter chase, replacing Sean Penn with a black guy, and doing away with some complicated exposition set in Africa, they could double the film’s box office potential. “Wow,” says Gladwell. “Now, what if we were to throw away the script entirely and create a new screenplay based on the premise – Nicole Kidman as a U.N. interpreter in peril?” The consultants mull it over and calculate that if it were turned into a Bodyguard-style romance, the movie could bring in $200 million. Gladwell is dazzled. But he notices that the consultants look rather glum. “What’s the matter?” he says.
“Nothing,” says one of the consultants. “It’s just…I kind of liked the script the way it originally was. If I were a studio executive, I’d keep the original script, budget it at ten million bucks, release it to a few film festivals and hope for good notices.”
Gladwell’s conclusion is that technology can be useful, but there’s something to be said for following your gut. And by the same token, a DJ with idiosyncratic but human tastes will always beat out a computer who can’t tell the difference between Liz Phair and Kelly Clarkson.