Pupils and teachers question the new grading algorithm used by the IB Diploma Programme to predict student scores after it canceled in-person tests this spring

The International Baccalaureate program canceled its high-stakes exam because of Covid-19. The formula it used to “predict” scores puzzles students and teachers.

Wired 2020-07-11 Tom Simonite

Context & Ripple Effects

When Covid-19 forced the International Baccalaureate to cancel its in-person Diploma Programme exams this spring, it substituted an algorithm that predicts student scores — and the pupils and teachers whose results hang on that formula say they cannot see how it works. The episode previews the collision that followed in the UK, where an exam-regulator algorithm downgraded tens of thousands of students from their teachers' estimates, triggering enough outcry that the national statistics regulator opened a review of Ofqual's grading model.

First-order effects

IB students awaiting university offers receive predicted scores they did not earn through any exam they sat, with teachers unable to explain or contest how the formula weighed coursework and school history.
Schools administering the IB must defend opaque results to parents and admissions officers without access to the model's internals.

Second-order effects

The UK downgrade backlash becomes the cautionary case for every exam body weighing algorithmic substitution: Ofqual's review signals that statistical fairness in grade-setting is now subject to formal regulatory scrutiny.
Teachers' workload concerns migrate into adjacent automated assessment — as later coverage shows, educators already spend hours correcting tests mis-scored by auto-grading bots matching answers to keys.

Third-order effects

If the pattern holds, high-stakes credentials converge on hybrid models where teacher judgment anchors algorithmic scoring — a tension visible again in 2024, when schools adopt AI grading tools for speed while critics flag harsh or glitchy marks as AI grading tools spread through classrooms.
Opaque predictive scoring of individual life outcomes invites standing oversight bodies for educational algorithms, much as statistics regulators now review exam models.

The trend: High-stakes assessment is shifting from human-marked exams to algorithmically predicted scores, with each contested rollout hardening demands for transparency and regulatory review.

Discussion

@myrthereuver Myrthe Reuver on x
Insane! Students for the International Baccalaureate (international, high-level high school) obtained their final grade from a prediction by a black box ML model. WTF? Why? Who thought this was a good substitute for an actual exam!? https://www.wired.com/...
@leavittalone John Leavitt on x
And here we are with the increased automation of white collar jobs will lead to shittier, but cheaper, outcomes https://twitter.com/...
@zittrain Jonathan Zittrain on x
The strange thing about it is that the line is blurry between just assigning a final grade and calling that final grade a “prediction” of what a “real” final grade would have been. I guess the big differences are the bizarre attempt to use statistical modeling, and the secrecy. h…
@cfiesler @cfiesler on x
TL;DR We can't give a test so we'll just have an algorithm predict what you would have scored. Who thought this was ok?! Why even bother with school at all anymore? Let's just tell kids when they enter kindergarten what college they're getting into. https://www.wired.com/...
@lpconstance Constance L on x
The sad story of poorly designed machine learning scripts ruining bright students lives #ibscandal #MachineLearning #bigdata #aifail #ai https://twitter.com/...
@caitlin__kelly Caitlin Kelly on x
When the International Baccalaureate program canceled its exam because of Covid-19, it decided to use an algorithm to “predict” grades instead. It...has not gone well. By @tsimonite: https://www.wired.com/... https://twitter.com/...
@amphithea Angie Svensen on x
@harari_yuval @audreyt Please take a look at the #ibscandal and how an algorithm can ruin your life. It feels like a story from your book #21_lessons https://www.wired.com/...
@worldwise001 Sarah on x
As a former TA who sat with students for assignments and proctored and graded exams, let me tell you that students surprise you in the final exam all the time and I doubt that any ML algorithm could adequately predict a students' performance. Give them take home exams. https://tw…
@warren_allen Warren Allen on x
This story is pretty wild, and I'd think it's exactly why the automated decision making restrictions of GDPR (Art 22) exist. Do these providers have their (EU based) students sign the best drafted consent ever? What does the DPIA look like? Seems ripe for complaints. https://twit…
@natematias J. Nathan Matias on x
Woah, students had college offers rescinded based on predicted IB scores when tests were canceled. “basing a student's grades on past trends at their school, potentially unrelated to the student's own school career, could be unfair” cc @AndrewDeanHo https://twitter.com/...
@random_walker Arvind Narayanan on x
I'm learning with growing horror about a high school diploma program that cancelled final exams and instead determined the fates of 170K students based on an outsourced black-box model that uses historical patterns as well as teacher-predicted grades. https://www.wired.com/... ht…
@noneuclideangrl @noneuclideangrl on x
when I say final exams are bad, I don't mean “replace them with AI” https://twitter.com/...

Chronicles