Page 37 - The Real Work Of Data Science Turning Data Into Information, Better Decisions, And Stronger Organizations by Ron S. Kenett, Thomas C. Redman (z-lib.org)_Neat

Get Out There 23

So work hard to understand their perspectives, but don’t depend on them. In particular, note
suspected sources, their nature, why you expect they cause variation, and their dynamics.
Then look for them in your data analyses, especially seeking to understand how to control or
represent them.
A great example of why understanding variation in data inputs is so important is illustrated
in a sketch on the BBC Scotland Burnsitoun comedy show (BBC 2017). Two Scots are in a
voice‐activated elevator that does not understand their accents. Obviously, not everyone speaks
the Queen’s English and the system was not designed to handle variability in the spoken word.
Predictive models build using AI must also address variability in inputs. Understanding varia-
tion is key to predictive maintenance in advanced manufacturing (Barkai 2018).

Selective Attention
Now a couple of words of caution and possible opportunity. First, beware “selective attention,”
where one pays too much attention to certain details, while missing the larger context. In a
popular video, two teams of three players pass basketballs to one another. One team has white
shorts, the other has black shirts. People are asked to carefully watch the video (which is
1.21 minutes long) and count the number of passes by the black team (Simon 2010). Typical
counts range from 9 to 15, which itself calls into question people’s abilities to count events just
by watching.
The punch line is different, however. Halfway through the video a black gorilla steps in,
jerks its hand around, and leaves the frame. More than half of people seeing the video for the
first time do not see the gorilla. In fact, when we rewind the video and point to the gorilla,
many believe we tricked them. So keep your eyes open for the black gorillas wandering around
that no one sees. The insulated probe in the oil field is just such an example!

Memory Bias
Second, beware memory bias as you talk to people. Elizabeth Loftus, a psychologist at the
University of California, Irvine, studied the forces that taint people’s memories after an expe-
rience is over, and she has consulted on hundreds of criminal cases. “Just because somebody
tells you something, and they say it with a lot of detail and a lot of confidence and a lot of
emotion, doesn’t mean it really happened,” she says (Loftus 2013; Baggaley 2017).
Much soft data resides in the memories of employees and may be distorted. And people
often have their own agendas and biases, further distorting their memories. So talk to people,
but be cautious.

Implications
So much of the nuance in the real world is not properly captured in the data, the metadata,
or management reports. Data scientists need to understand this nuance. Engaging in water
cooler discussions and talking to experts helps, but it has limitations. Touring facilities,
visiting customers, joining a service technician on one of his or her shifts, and riding with a
trucker provide insights that cannot be gained any other way. At a minimum, the soft data
gathered in this way provides a context for the numbers you analyze. And often so much more.
Thus, the real work of a data scientist involves talking to people, visiting the places where
the data is created, and delving into the details.