Profiling by Algorithm
By Albert B. Kelly
Today, high school and
college-age students face enormous obstacles and challenges when it comes to
higher education. Obviously, the cost of college is a huge worry for students
and their families; not only pulling the tuition money together on the front end,
but student loan debt on the back end.
One cost effective
approach for college is the use of online courses which come with less overhead
and allow students more flexibility with schedules and assignments. I’m all for
doing whatever it takes to facilitate learning and helping people to a college
degree. As useful as it is, the online model is not without its issues.
A piece in the NY Times by
Natashia Singer featured a young woman taking online courses from our own
Rutgers University. It zeroed in on the process she had to go through for
mid-term exams. Like the course itself, the exam was done online and in order
to ensure that students were not cheating, they had to download an
anti-cheating program called “Proctortrack”.
The exam fee to use
Proctortrack was $37. Students using the program place their face and hands on
the computer screen so the program can verify the student’s identity. Once the
test started, the webcam recorded the student’s activity while Protctortrack monitored
the computer being used to ensure that there’s no cheating.
You can’t have students
bringing up Wikipedia to find the answers; nor do you want them glancing at
books next to the computer, listening to prerecorded answers on headphones, or
reading from poster boards taped to the wall over their desks.
At first glance, this
makes a lot of sense. The whole point of the exam is to make sure that students
have a certain understanding of the material being taught. Beyond that, if a
college is to issue a degree, it wants to know that graduates have earned the
credential and will be competent in their chosen field.
But there’s more to it.
Proctortrack uses algorithms to detect unusual student behavior that could
possibly mean the person is cheating. The program, having amassed this
information, then categorizes the student as having either high or low
integrity. And Proctortrack is only one such anti-cheating program.
According to the story,
Utah Valley University has a program called “Spotlight”, dubbed an early
warning system, it uses academic and demographic details about students to
predict how likely they are to pass the course and this information is then sent
to professors so they can keep an eye for cheaters.
This is where things get a
little dicey. With Proctortrack, if a student bends down to scratch an itch on his
or her foot will that be considered unusual? How about a student gazing off the
screen as they think? God forbid if a bathroom break is needed; imagine what
that might do for one’s integrity quotient.
As for Utah Valley’s
“Spotlight” program, you have to wonder if students with average grades coming
from lower-income backgrounds are the ones that will be deemed most likely to
cheat. And if so, you also have to wonder what that means for minority and
foreign-born students.
Even with something like
Proctortrack analyzing body language and “flagging” a student so the professor
can evaluate footage; much comes down to who is doing the evaluating and if
they have some bias; be it for or against males, females, blacks, whites,
Latinos- the same gesture might be judged as innocent from one person or
cheating from another depending on who’s doing the deciding.
I wonder if test grades
suffer simply because the student is so uptight about not being perceived as
cheating, that they do poorly on the test for that reason alone. With Proctortrack,
apparently everything from changes in lighting to stretching or leaning can
flag you as a potential cheater.
It’s no small thing. Over
3, 500 institutions use an automated plagiarism detection system called
“Turnitin” which scans a student’s paper to see if they’ve lifted passages from
some other text. As online courses become the new normal, these anti-cheating
algorithms and programs will be commonplace.
As I said at the outset,
we need to get as much value out of online learning as possible. But maybe there
are some things, like a human proctor in a classroom at test time, which should
not be changed out for a virtual proctor at $37 a pop. With all the challenges
we face in higher education, we can’t afford to add algorithmic profiling to
the list.