Database Reference
In-Depth Information
I think we need another five to ten years to get a better sense of best prac-
tices and expectations on when things really work and are useful, and when
they are not. I do not think that anything will change fundamentally about how
data science is done. With the hype going down, I think some of the more
hard-core concepts that have been going around, like optimization, will gain a
little bit in appreciation, and some of the hype-y things will go down a little bit
when people realize that ultimately they cannot really make any impact.
For instance, I am very hesitant to embrace every time somebody asks me
for actionable insights. It scares me. It really scares me because how can you
expect me to tell you the actionable insights if you do not tell me first what
your actions are? Unless you communicate very clearly and think about what
exactly it is you can do and are willing to do based on data, sending me off on
a wild goose chase with data to come back with actionable insights is a fool's
errand. The more data you give me, the worse it gets because I will find more
stuff that is probably meaningless. I think we need to change that process and
that comes back to communication. I think that will improve, but I do not
think anything very fundamentally will change.
I do not believe data science will be automated. I do not believe that your
secretary will do data science for you, as much as that is kind of the way it is
positioned—that anybody can do data science. That is not true, because more
often than not, it is a problem with the data that you have to find first, and if
you do not know data, then you cannot do that. Just because you can run an
algorithm on data does not mean that you get something meaningful. You do
need to look at the data. And that skill has to be in there. There has to be a
human there, somewhere. You cannot hide it all away and abstract it in data
layers and cool tooling. It is not going to work.
Gutierrez:
Are there any areas you think data science should focus on?
Perlich:
I think there is a tremendous opportunity for data science to have
a huge impact in the medical field, in particular general well-being. To make
this impact occur, we have to go back and really figure out how we deal with
privacy and data sharing in order to get a good stake in the ground. I have
personally walked away from medical applications repeatedly because I just
got so hung up in HIPPA regulations. It was depressing because I felt like I was
getting my hands tied behind my back. It was very depressing to know that
I could have otherwise been making a large impact.
In the medical field, I think it is crucial that we understand a lot better what
the tradeoffs are with these regulations. We need to have the medical experts
communicate exactly what can be done and who the people doing the work
with the data would be. As with most things, part of the problem is a lack of
information sharing. Every doctor basically has his cases to read and knows
