Tens of thousands of UK high school students were downgraded from teacher estimates by an algorithm that replaced exams, affecting university applications
Even with the late changes, GCSE and A-Level results still highlight the flaws in a biased education system
WIRED UKAmit Katwala
Context & Ripple Effects
When COVID cancelled the 2020 exam season, exam regulator Ofqual handed grading to an algorithm that standardized results against each school's historical performance — and it systematically dragged down pupils at schools whose teacher estimates ran above their past averages. The backlash was immediate enough that within days the UK government scrapped the model entirely and reverted to teacher-predicted grades, after claims the downgrades fell hardest on state-school students applying to university.
The episode didn't end with the reversal: the UK's national statistics regulator opened a formal review of how Ofqual built and validated the model (announced days later), making this one of the first times an automated public-sector decision system faced regulatory audit over bias rather than accuracy.
First-order effects
Tens of thousands of students saw A-Level and GCSE grades cut below their teacher estimates, directly changing which university offers they met — until the government's reversal restored teacher grades and re-opened places on contested courses.
Ofqual's credibility as arbiter took the immediate hit, with its statistical methodology now subject to review by the national statistics regulator.
Second-order effects
Universities had to absorb a whipsaw intake — first admitting students on inflated algorithm-adjusted capacity assumptions, then honoring teacher-estimate grades that exceeded them — pressuring admissions caps and accommodation planning.
The fiasco set a reference case for every institution weighing automated assessment: teachers later adopting AI grading tools still report glitchy or overly harsh scoring that requires manual review, showing the human-review burden never left the loop.
Third-order effects
If the pattern holds, automated high-stakes decisions in education get audited as statistical products — with regulators demanding documented bias testing before deployment, not after student protests force a reversal.
The trust gap compounds downstream: with most UK undergraduates now using AI in their studies and teachers reporting declining critical thinking among pupils, institutions face pressure to rebuild assessment around verifiable human work rather than either algorithmic or teacher-estimate proxies.
The trend: High-stakes automated decision systems are moving from quiet back-office tools to publicly audited infrastructure, with education as the first sector where bias in the model outweighed efficiency in the eyes of both regulators and the public.
“it locks in all the advantages and disadvantages - and means that the talented outlier, such as the bright child in the low-achieving school, or the school that is rapidly improving, could be delivered an injustice.” https://www.bbc.com/... a future determined by algorithms
I was curious about the exams controversy and thought there must be more to the story than the government penalising people because of the school they went to. Nope. Turns out the system they used really was as unfair and stupid as it seems. https://www.bbc.com/...
A story of algorithmic bias is exploding in the UK. A-levels are the equivalent of SATs. They were canceled this year due to COVID so teacher estimates of grades/mock exams were used AND modified by algorithm based on the past performance of their school. https://www.bbc.com/...
This @WiredUK article about results day claims to prove that it's a “diversity disaster”, but doesn't. It drops in the obligatory mention of “ethnic minority students” but ignores that working class boys (of all races) are at the bottom of the pile (1/3) https://www.wired.co.uk/.…
I wrote about the absolute mess that is A-level results day, and why it could still have a disproportionate impact on ethnic minority and poorer students, for @WiredUK #resultsday https://www.wired.co.uk/...
“The problem here is that exams were never a great metric for learning or success anyway” The mess of exams under Covid is revealing the bias in the system more clearly than ever https://www.wired.co.uk/... by @amitkatwala
We aren't going to hear the end of this for a good few years, and rightly so unless the govt U-turns. Grades haunt young professionals for the first few years of their careers. 😞 “Results day is a diversity disaster. Here's all the proof you need” https://www.wired.co.uk/...
Explanation of how the algorithm worked in deciding A Level grades. Sit down before you read. If you're 18, register to vote https://www.theguardian.com/ ...
This year, results day is very different. Because of the coronavirus pandemic, exams were cancelled and replaced with teacher assessments and algorithms. And it has created chaos https://www.wired.co.uk/...
So much luck to everyone getting results today. Awful that calculating grades this way will have left lots of students with unfair grades due to the bias they experience at school. I was predicted an E by my teacher, got an A in the exam 2 months later. Things need to change x ht…
Does anyone know the Ofqual algorithm inputs? Have they published their model? Essentially algorithms are often discriminatory depending on inputs- if made judgements on schools as a whole then flawed and probably illegal. I sense a long court battle here. Govt should U-turn.
No system was ever going to be completely fair but hearing so many real tragedies this morning. Eg told of an exceptional student in an historically low performing school downgraded two grades at A Level. She has a place at @Cambridge_Uni. They're not accepting her.
I assume ofqual undertook modelling of its algorithm? They must have known the impacts of downgradeing would be felt very differently by students across the system according to socio-economic characteristics, geographical location and school types. Did no-one challenge this?
The algorithm used by Ofqual can't be applied to small cohorts. Private schools have far more small cohorts than state schools. So this system they've created favours private schools. Ofqual knows this, the Tories know this. They don't care about kids in state schools. https://tw…