AI vs A&R? New Computer Algorithm Picks the Hits

Many of us imagine we can identify talent better than the labels' "Artist & Repertoire" (A&R) staff. Recently a company tried using artificial intelligence (AI) and announced it could reliably predict the success of recordings.

In a recent New York Times article, music columnist Neil Strauss admitted that he first though the press release from Barcelona-based Polyphonic HMI was a joke, "poking fun at the desperation and cluelessness of the music business." But soon he realized that it was the real thing.

HMI visualization of an album superimposed into the recent "hit universe"

"Hit Song Science," as the application is known, is based on a technology from AI firm Grupo AIA. Polyphonic says their product analyzes melody, beat, harmony, pitch, octave, tone quality and other patterns in a piece of recorded music, and compares the results to a "universe" of other songs. Their claim - hit songs tend to lie in the same part of the universe, that is, have a lot of those characteristics in common. The product also can dig inside a song and find the best excerpt to use in advertisements and films, the so-called "gold content," according to Polyphonic. Some outsiders are skeptical. Jordan Berliant is quoted as saying "What creates a hit is that people have an emotional reaction to a song, in particular the lyrics. It's difficult to believe that a machine could gauge that." Virtual reality pioneer Jaron Lanier thinks the science behind this application is "sloppy," but he also admits that he "doubt[s] pop music could get any worse, so using even a meaningless tool like this might result in some improvement."

Visit the Polyphonic HMI company site

Visit the Grupo AIA company site

Read a related commentary by Rupert Goodwins (ZDNet - UK), and further coverage in Australia's The Age

Read Neil Strauss' column (see second part) in the March 12, 2003 New York Times [fee required]

Read an older article (New Scientist, 15 June 2001) recounting some related developments by Savage Beast Technologies