Thread by @doctorow

On the one hand, high-stakes testing is pedagogically bankrupt, but on the other hand, it sure produces numbers that universities can focus on increasing, and then trumpet when those numbers are higher than they used to be. I guess that's important?

1/

The problems with high-stakes testing were magnified by lockdown, with universities demanding that students infect their computers - often shared with family members - with spyware that claimed to perform "invigilation" (anti-cheating surveillance).

pluralistic.net/2020/04/15/inv…

2/

Having decided that invigilation software was easier than finding a way to evaluate students without useless high-stakes tests, universites began a campaign of cruel bullying to crush student opposition.

3/

At Wilfrid Laurier in Ontario, computer science students were required to purchase webcams that were not available to sale, on pain of flunking.

pluralistic.net/2020/05/13/mal…

4/

And when students around the world complained about the problems with invigilation tools, the CEOs of these massively profitable ed-tech profiteers did what any responsible exec would do: they doxxed their underaged critics.

pluralistic.net/2020/07/01/bos…

5/

Writing in @techreview, @SheaSwauger describes the disproportionate impact that invigilation software has on marginalized students, especially Black and trans students and students who are parenting young kids from home.

technologyreview.com/2020/08/07/100…

6/

For starters, the facial recognition software has the well-understood algorithmic racial bias thanks to deficits in training data.

7/

That means that when Black students sit their exams, the tools demand that they increase the lighting to aid in facial verification, and often reject them outright, so they can't sit the exam at all.

8/

Similar problems occur for trans students who are transitioning, whose faces are no longer recognizable by the facial recognition system.

9/

Meanwhile, the systems' unblinking eyes are incapable of distinguishing between students who are cheating by having a confederate in the room and students with young kids being interrupted as they sit their exams.

10/

Likewise, they can't tell the difference between someone who gets up to cheat and someone who gets up because they have a medical condition that requires them to take frequent toilet breaks.

11/

Creepily, the tools require students to pan the camera around their living spaces for "room checks," and then let their profs download and view these images of intimate living quarters, including any family members who have nowhere else to go.

12/

These are important considerations, and there's an equally important principle lurking behind the surface, which is that the problem isn't merely that the algorithm is racist or that the tool discriminates against marginalized students.

13/

The fact that facial recognition struggles with Black faces is the result of a training data deficit, which can be easily fixed, by feeding the algorithm LOTS of Black faces to chew through.

14/

That's what Chinese quasi-state surveillance companies did, using Zimbabwe's driver's license database to perfect its ability to recognize Black faces.

qz.com/africa/1287675…

15/

That means that China - and repressive regimes elsewhere that buy Chinese surveillance tools - can spy on Black people really accurately.

This is not an improvement.

16/

And as to high-stakes testing: it's a fool's errand, completely uncoupled from real-world knowledge work AND any pretense of rigorous pedagogical assessment.

17/

Even if we found a way to do high-stakes testing without victimizing Black, trans, poor, and parenting students, it would still be garbage.

Image:
Cryteria (modified)
commons.wikimedia.org/wiki/File:HAL9…

Rawpixel Ltd (modified)
flickr.com/photos/byrawpi…

CC BY:
creativecommons.org/licenses/by/2.…

eof/