Fantastic article from Bloomberg BusinessWeek on the never-ending quest to get computers to predict human behavior.
Turns out, people can invent computers who can best Grand Champion Garry Kasparov in a chess match, but they haven’t found a way to determine how people will think.
Although Barcelona-based Hit Song Science correctly predicted the success of Norah Jones’ runaway hit album, Come Away With Me strictly using a computer analysis and comparison of the rhythms and pitch of her music to those of other hit compositions, the same computer analysis concluded that Michael Jackson’s “Billie Jean” would have been a dismal failure.
Human tastes are simply too fickle for computers to predict—at least right now.
However, the power of persuasion is such that people can and do influence the tastes of other people, and computers can leverage these preferences to effectively sell goods, services and ideas. (If they didn’t then marketers wouldn’t be beating a path to the doors of Facebook, Twitter, et al, right?)
Case in point: according to a study conducted by a team from Columbia University, who asked people to rank 14,000 songs by groups they’d never heard of found that those who were aware of the ratings others had provided were more likely to like the songs other people preferred, even though different groups of people liked different songs.
Companies such as New York-based BuzzFeed are leveraging this knowledge to analyze web content and predict which bits of content are most likely to go viral. BuzzFeed recently teamed with an outfit called Undercurrent to drive a web-based marketing campaign for DonQ Rum called “LadyData” that ultimately increased sales for the brand by 55 percent.
So I guess you could say, “Computers don’t create killer campaigns, PEOPLE with computers create killer campaigns.” (Apologies to the NRA.)