hen Norah Jones released her first
album, she was a long shot at best. ''Come Away With Me'' was filled
with mellow, sultry tunes -- precisely the opposite of the
histrionic diva pop crowding the charts. Virtually no one expected
Jones to score a major hit.
No one, that is, except for a piece of artificial intelligence
called Hit Song Science, a program that tries to determine, with
mathematical precision, whether a song is going to be a Top 40 hit.
When the scientists fed Jones's album into that computer, alarm
bells went off: the program predicted that eight tracks would hit
the charts. ''We were like, whoa, that's funky,'' says Mike
McCready, the C.E.O. of Polyphonic HMI, the Barcelona-based company
that developed the software application. A few months later, Jones's
album went multiplatinum -- and Hit Song Science had proved it could
pick a hit as well as Clive Davis.
But how? At the heart of the program is a ''clustering''
algorithm that locates acoustic similarities between songs, like
common bits of rhythm, harmonies or keys. The software takes a new
tune and compares it with the mathematical signatures of the last 30
years of Top 40 hits. The closer the song is to ''a hit cluster,''
the more likely -- in theory -- that the kids won't be able to
resist it. Yet the weird thing is, songs that are mathematically
similar don't necessarily sound the same. The scientists found that
U2 is similar to Beethoven, and that Van Halen shares qualities with
the piano rock of Vanessa Carlton. Even more bizarrely, 50 Cent's
throbbing rap tune ''If I Can't'' correlates with ''(There's) No
Gettin' Over Me,'' a twangy country ditty by Ronnie Milsap.
This year, several record companies began using Hit Song Science
to help pick which songs on an album to promote. Others are now
using it in the studio, taking a rough mix of a new song, checking
to see how hit-worthy it is, then tweaking it until it has ''good
mathematics,'' as McCready puts it. He can foresee a day when most
major hits will have been vetted by algorithms.
Which is, depending on how you look at it, either a wonderful
breakthrough for science or an incredibly bleak statement about the
music industry. Critics for years have complained that record labels
produce only bland albums that mimic what's already popular. But Hit
Song Science takes that trend to its logical absurdity: it does not
merely aim at the middle of the road -- it calculates it,
with scientific precision.
