Nature Future Conditional

The story behind the story: Coin-operated dancer

By day, James Reinebold is an AI programmer in the video-game industry. Fortunately, when he’s not programming, he writes, and this week’s Futures story is the result. Coin-operated dancer considers the potential glitches with AI — and does so with some happening dance moves. You can keep up to date with James via his website and his Twitter feed. Here he explains the background to his latest tale — as ever it pays to read the story first.

 Writing Coin-operated dancer

I listen to a music streaming service while I program.  It works like an eager-to-please robotic DJ:  suggesting new songs based on what I’ve already told it that I like.  Although it can be frustrating when it makes silly suggestions, overall the algorithm does a pretty solid job of picking out what I want to listen to.  And all I had to do was give it a few examples.

The concept of how a few lines of code can model and even offer new insight into problem spaces as diverse as computational fluid dynamics, car navigation and music fascinates me.  My story is a reaction to that:  we can hook up fancy sensors and engineer the most elaborate learning algorithms possible, but unless we provide these things with some actual experience they’re going to be pretty useless.

I’m pretty optimistic for the future of artificial intelligence.  It’s amazing what computers can do.  How well our algorithms work in practice, though, ultimately depends on us providing solid, representative data to them as inputs (which is yet another reason for diversity in the sciences!).  If we hope to train robots to dance and sing, we need to prepare their neural networks and radial basis functions for a big, beautiful and noisy universe.


There are currently no comments.