mathbabe

I’ve noticed a recent trend in coverage of data science. Namely, there’s backlash against the hype and the over-promising, intentional or not, of data science and data scientists. People are beginning to develop smell tests for big data and raise incredulous eyebrows at certain claims.

This is a good thing. We data scientists should welcome the backlash, first because it’s inevitable, and second because it allows us to have a much-needed conversation about how to behave and what is reasonable to claim or even hope for with respect to big data. There is a poseur problem in big data, after all.

But, fellow data nerds, let’s take this as a cue to start an internal discussion about data science skepticism. Let’s make sure that it’s coming from our community, or at least the surrounding technical community, rather than from yet another set of poseurs who don’t actually know what data is and would only serve to lampoon and discredit our emerging field rather than improve it. We should be the ones leading the charge and admitting when we’re full of shit. We need to own the backlash.

Let me give you an example. A serious data scientist friend of mine recently got asked to be interviewed as part of a conversation on data science skepticism. After thinking hard about what her contribution could be, she wrote back to accept the offer, but was then told she was “off the hook” because they’d found someone else who was “perfect for the assignment.” It turned out to be a journalist who had previously interviewed her. That was his credential for this conversation.

But how can you actually have informed skepticism if you are not yourself an expert?

Another example. David Brooks recently wrote a column wherein he declared himself a data science skeptic and then followed that up by referring to no fewer than eight random statistical studies that made no coherent sense and had no overall point. My conclusion: this is the wrong man to lead the charge against poseurs in data science.

If we are going to rebel against big data soundbites, let’s not do it in soundbites. Instead, let’s talk to people on the inside, who see specific problems in the field and are willing to talk openly about them.

I liked the recent Strata talk by Kate Crawford entitled “Untangling Algorithmic Illusions from Reality in Big Data” (h/t Alan Fekete) which discusses bias in data using very concrete examples, and asks us to examine the objectivity of our “facts”.

For example, she talked about a smart phone app that finds potholes in Boston and report them to the City, and how on the one hand it was cool but on the other it would mean that, if naively applied, richer neighborhoods like Lincoln would get better services than Roxbury. She explained an important point: data analysis is not objective, which most people know. But often the data itself is not either – it was collected in a certain way with particular selection biases.

We need more conversations like this or else we will be leaving a hole which will be filled with loud, uninformed skeptics who would be right to raise the alarm.

One last thing. I’m aware that tons of people, especially serious academic statisticians and computer scientists, criticize data scientists for a totally different reason, namely that we are overly self-promoting (although academics have their own status plays).

But I don’t apologize for that. The truth is, a data scientist is a hybrid between a business person and a researcher. And this is a good thing, not a bad thing: it means the world gets direct access to the modeler, and can challenge any hyperbolic claims by asking for details, rather than having to go through a marketing person who acts (usually quite poorly) as a nerd interpreter. I for one would rather represent my work directly to the world (and be called a self-promoter) then to be kept in the back room.