(In Firefox or
Chrome, right click link and select open link in Private or Incognito window to
bypass subscriber log-in)
On August 30th, a
contributor to Foreign Policy Magazine wrote about an algorithm which could be
used to help policymakers predict the future geopolitical landscape. My first
thought was that such a predictive program would put me out off a future career.
The skill of an international relations specialist comes from combining knowledge
of a group's interests with knowledge of culture and history. The goal is to predict that group's reaction
to a given stimulus. The task of synthesizing a response based on the
integrated, and almost sub-conscious, calculation of many variables is
completed by utilizing the amazing parallel processing power of the human
brain. The output of this subconscious, parallel processing is perceived by
individual humans as intuition. As the power of man-made computers increases, parallel
distributed architectures can be developed that are capable of making
"leaps" previously reserved for the human brain.
The Global
Database of Events Language and Tone (GDELT) is a program that uses the
exponential increase of computing power to turn a jumble of hundreds of
variables into a concrete trend line.
The GDELT works by
analyzing all sorts of global media, coding the media with different values for
each variable, and using these values to create a trend line of general unrest
(link to the math in article for those who know stats). The predictive power of
GDELT is derived by comparing the trend line of an ongoing event with the trend
line of a similar previous event. GDELT is designed to provide a concrete
measure of the degree to which history repeats itself.
Luckily for me,
the GDELT will not be replacing political analysts any time soon. The system
has two major flaws. The first flaw is endemic to any model with a bunch of
variables. If a variable is missed when constructing the model, it is hard for
the researcher to find his mistake because one can't see what one doesn't know
to look for. The second big flaw with GDELT is the input data. Since GDELT
relies on media coverage of an event for data, the data is tainted by media
bias before it even enters the system. Even if the model were technically
perfected, the reliance on media coverage would always contaminate the data.
The designers of GDELT say that even if the system cannot be used to predict
events, it can be used to predict the way the media responds to events.
Here is a link to
a cool animated map of global unrest the GDELT team created using their
program. They are definitely on to something.
http://gdeltproject.org/globaldashboard/
No comments:
Post a Comment