How numbers on Facebook change behavior
So
it seems this article falls into the overall disturbing theme of analytics and
data being used to effect, change or even shape people’s lives and even
viewpoints. I found Ben Grosser’s ‘Facebook Demetricator’ very interesting, as
well as people’s responses to it. The fact that many people expressed that once
the numbers were gone, they realized they had been using them to decide what
they liked or whether or not the like something was particularly intriguing. It
reminded me of mob mentality, but in an extremely prolonged state. My brother
used to always say, “A person is smart, but people are stupid,” meaning that
while individuals can think through problems and be reasoned with, a mob of
people is irrational and cannot easily change its course or its mind. Perhaps this
can explain people’s reaction to the numbers on Facebook – they are so used to
following the group’s decisions that they have temporarily lost sight of their
own preferences. This is a scary thought
when you relate this piece to Quinn Norton’s “The Hypocrisy of the Internet
Journalist” and her warnings about opinion shaping and how data analytics potentially
allow those collecting the data to remake people’s worldviews.
The promise of big data
I
did find the majority of the research and work being done described in this
article to be very promising. The work that IBM’s T.J. Watson Research Lab is
doing with the Ureport System I think is particularly inspiring, because it
simply isn’t possible for a person to comb through that much data and the
impact on people’s lives is very real and very important. However, I can’t help
but immediately worry about the computer program making mistakes. My first impulse
is to want an actual human being reviewing the data or reviewing the computer’s
work, even though I know this is improbable. There are countless examples of
analytic software making mistakes – several are mentioned in later articles.
But what is the solution? Just not using this software is not an option and it’s
simply too much data for human beings to handle without computer aid. I think
this is a very important question to focus on and research.
There
were two points in this article in particular that stuck out to me as problematic.
First, Google research scientist Diane Lambert’s comment that, “If you’ve ever
put a query into Google, then you’ve been in an experiment.” I guess consenting
to participant in experiments in one of those items in the impossibly long
terms of service that no one reads.
Second
was Dean Cherry A. Murray’s comment about the new breed of data scientists,
that graduates would “have the opportunity to inform decision-making in
science, business, or government settings.” Taking into account later articles’
point about inherent human bias in these analytic software programs, I found it
disturbing that she foresees creators of such programs informing government
decision-making.
The pitfalls of big data
The
most infuriating part of this piece, besides the point of the article itself,
is that the Princeton Review does not seem to have any intention of addressing
the problem brought to light by the study. I tried to find an update on this
story but was unsuccessful. Has the Princeton Review made any effort to change
its discriminatory policies? It doesn’t matter if it was unintentional, it
should still be fixed.
Kate
Cox, the author of the article, makes the point that the Princeton Review’s algorithms
are designed by humans that bring their own assumptions and biases into the
code. So perhaps it’s not all that surprising that the organization that told Asian
American college applicants to play down their Asianness by not attaching a
photograph, not answering the optional question about ethnic background, and
advising against application essays that discuss how their culture affects
their lives to increase their chances of admission in their 2004 Cracking College Admissions (http://www.bloomberg.com/news/articles/2014-11-21/princeton-review-tells-asians-to-act-less-asian-and-black-students-to-attach-photos)
has been found to discriminate against Asian Americans.
The
source for this article, ProPublica, even belies an inherent bias with the
title it chooses for its version of this story: “The Tiger Mom Tax: Asians Are
Nearly Twice as Likely to Get a Higher Price from Princeton Review.” ‘Tiger Mom’
being a negative term often applied to Asian mothers who, according to
stereotypes, push their children too hard and are overbearing and controlling.
Interestingly, this article lists several other recent
instances of discriminatory or suspect behavior by companies’ programmed
analytical software:
- In 2012, Staples was varying prices by zip code with the inadvertent effect that people in lower income zip codes had the higher prices
- In 2014, Northeastern University found that top websites like Home Depot, Orbitz, and Travelocity were steering their users toward more expensive products
- In 2015, another study found that users identified by Google as female received fewer ads for a high-paying job (also mentioned in article below)
How social bias creeps into web tech
The
biggest thing that stood out to me in this article was the quote from Vivienne
Ming: “Computers aren’t magically less biased than people.” The problem of bias
and discrimination in data analytic software is not a simple one and not easily
solved. But I think a great first step would be to move away from the idea that
computers and software are neutral and impartial, therefore unbiased, solely
because they are not human beings. They are still created and programmed by
human beings who cannot escape bias and program their personal biases into
them. The sooner we all realize that, the sooner we can move closer to reducing
and potentially one day solving this problem.
No comments:
Post a Comment