Sunday, October 15, 2006


Ryan tipped me to the Netflix contest: they're offering a million dollars to the team that improves the reliability of their customer recommendations by 10%.

It's apparently a realistic goal, software recommendation programs have been improving dramatically, they're quite good now, and they're getting better all the time.

In less than 5 years (if it hasn't happened already), algorithms will be far more knowledgeable about human tastes than any human critic.

Eventually, the big wave in robot recs will be having AI anticipate life trends, so when the teenager who listens to punk is ready to make the switch to country or jazz in his twenties, the robot will already know ahead of time which crossover artists to use to ease the transition, and will automatically substitute Hank Williams for Henry Rollins. And the AI will be ready again when he switches to easy listening in his forties. The robot might also recommend appropriate changes in dress and select items from the IKEA catalogue to go along with his chosen aesthetic.

Will AI also soon be ready to predict what social policies we're most likely to adopt, or is politics somehow distinguishable from fashion?