I've been thinking a lot recently about the deeper meaning of data science and prediction in practical contexts. One idea I keep returning to is the notion of data science as a form of advocacy. A valuable model doesn't just generate predictions, it generates signals that are intended to inform people and spur them to take specific actions. That sounds a lot like advocacy to me!

I suspect many people who are aware of data science in their industries would be surprised to hear the field described in this way. People often promote predictive analytics and related disciplines as an objective means of understanding problems. Whether with good or ill intentions, the tendency to talk about data science as an objective arbiter of decision-making conveniently hides many of the ethical and practical pitfalls that predictive methods can introduce into a system.

Talking about data science models as a type of advocacy creates an opening for people to engage directly with the likely consequences and potential side effects of introducing predictions into a system of action. What are we advocating on behalf of? What change are we trying to produce in the world? Why did we make that decision? What are the potential trade-offs? What kind of adverse side effects can we tolerate? What adverse side effects are we not willing to tolerate?

There's a lot of interesting work going on in the field right now about ethical machine learning and ways to implement safeguards into these projects. It seems to me that a great starting point is recognizing that any attempt at prediction is a form of advocacy.