Database Reference
In-Depth Information
Smallwood:
Titles are always a challenging thing. We didn't even use the
word “data science” a decade ago. So titles across the industry are a challeng-
ing thing. In terms of what people do, it's more self-selecting. People tend to
be either more interested in the product and experimentation kind of work,
or more interested in algorithms. Even in the algorithms space, some people
have very specialized interest in particular types of algorithms. So I think, in
general, we try to group people in ways and in areas that they're most excited
about so that you're getting both the business benefit and people are happy.
We're pretty flexible on titles at Netflix. People can call themselves what they
feel like calling themselves.
Gutierrez:
What does the future of data science look like?
Smallwood:
I think it's going to continue to explode and it will become
quite ubiquitous. Kind of like how trend charts are everywhere. I think it
will be the same thing for models, algorithms, and the deeper data sciences.
The techniques are already amazing and they'll continue to get better. Even
where we stand today, they've really evolved amazingly with the depth and
sophistication of the math and the different kinds of techniques. And it's due
to having so much data available that's let people really investigate different
nuanced techniques. So I think those techniques will continue to explode. And
what will become more challenging for data science is being aware of so many
different techniques, being good at knowing how to formulate a problem with
the proper kinds of techniques, and not misusing the techniques.
I think that there will be a little bit of a danger involved as well. I think it will
be very easy to develop bad models because of open source availability, how
easy it is to program these techniques, as well as the vast array of complicated
techniques. It is already easy to develop bad models, and unfortunately I think
that will become even easier.
I think this is true in particular for companies that are trying to start a data
science team where they didn't have one before. I think it's a bit of a danger
zone for them. It will be very easy to hire someone who's built one regression
model and uses fancy terms in their interviews like “support vector machine,”
and the executives will go, “Wow! Come in and build our data science team.”
For these companies, there's no way to gauge whether the person they hire
knows what they're doing, because a great model versus a bad model is still a
model. They both spit out the same type of result and you have some mea-
sure of whether that's a good result or not, but it's impossible to really know
whether it's a good model or not. In my mind, it's all about the quality of the
people building those models, and I think it will be hard for inexperienced
companies to discern that.
Gutierrez:
How can people discern the good people from the rest?
