What You Don't Measure You Can't Improve: AI from the View of an Applied Scientist

Q: One of the conversations that have come about with the AI discussion is about a new role that would develop out of the AI boom — data scientist. So that's you, correct? What is your official title and main duties?

A: I guess you would call me an applied scientist rather than data scientist although I prefer the term research scientist. To me the term "data science", although it's a decade old, is a bit faddish. Such roles are typically backend AI, focusing on engineering better features, building better models, cleaning data. They have domain knowledge to communicate their findings to higher ups but it's really all about the mechanics of prediction. I tend to think of my role as there being more to AI than just better data and cleverer math. There's a process that you need to wrap around it, a human-machine interaction design, an evolution and iteration. This is especially true in the legal field because you have much less data available to you than in consumer web domains where there are literally billions of users. In legal, the role of interaction in the process becomes a lot more important. That's more than just data. It's understanding data science but it's also some human factors when designing good solutions.