Diving even deeper

In his comment on yesterday’s post, Doug suggested a philosophically interesting direction: Capture variations in singing style and then re-apply those variations to different songs. As Doug points out, this is in the same vein as the Image Analogies work by my brilliant former student Aaron Hertzmann.

I use the phrase “philosophically interesting” because herein lies a sort of divide in current thinking about ways to use computers. There is no right or wrong side to this divide. It’s more of an aesthetic difference.

Doug’s suggestion falls on the “Machine Learning” side: Feed a lot of real-world examples of something into a computer algorithm, and then apply that now-trained algorithm to new situations. In this case, we would be training an algorithm to recognize a singer’s style by first examining how they have performed some set of songs, and then use our tuned algorithm to simulate how that same singer would sing different songs.

My own aesthetics (and where I was going with this series of posts) leans more toward the “Model Driven” side: I would want to create a design tool in the form of an interface that represents a model of the singer’s choices. Using this design tool, a designer could reshape those choices, in effect creating their own custom singer.

The difference is crucial: The Machine Learning approach does not provide a way to understand the choices the singer makes. It can effectively apply the singer’s “style” to new songs, but in doing so it operates as a black box. The Model Driven approach reveals what is going on under the hood.

In other words, I’m not so interested in algorithms that just do things for us. I’m more interested in tools whose workings we can clearly understand, and that therefore can be guided by our own intuition.

Don’t get me wrong — the Machine Learning approach is immensely powerful, and in fact is the key to the power of such data-driven search tools as Google search.

It’s just that I don’t just want a player piano, even if it’s the best player piano in the world. I also want to play the piano — or even make my own piano.