At a time when students’ lives in the UK have already been upended with the school year being cut short by a deadly pandemic, an added layer of chaos and controversy has erupted after officials decided to entrust establishing pupils’ A-Level grades to a computer algorithm.
In theory, the algorithm used to determine the grades would be the “fairest possible for students progressing on to further study or employment as planned”, according to exams regulator Ofqual.
In reality, it turned out to be the opposite, something educations officials should have seen coming.
The algorithm was supposedly engineered in a way that was intended to produce fair results across the board while controlling for potential grade inflation by teachers who want to see their pupils succeed. What ended up happening is that 39 percent of students saw their grades drop when compared to their teachers’ recommendations. That percentage, in itself, is eye-opening and elicits doubt in the capacity of the algorithm to generate accurate results. But when we consider how disproportionately pupils from disadvantaged backgrounds were so negatively affected by the results, it becomes all the more apparent that something is not quite right.
Ultimately, the algorithm sought to establish how this year’s class of pupils would have performed on their exams based heavily upon how last year’s class performed in each individual school district. As a direct result of how this was set up, high achieving students from disadvantaged backgrounds were left unfairly punished while underachieving students from affluent areas were undeservedly rewarded. Essentially, what this means is that students’ grades were largely predicated on their postcode, and socio-economic status.
Apart from being a completely unjust methodology to establishing A-Level grades for students, it further highlights the broader issue of bias in artificial intelligence. Completely preventing bias from creeping into AI is not easy since unconscious human biases often unintentionally worm their way into algorithms without engineers even realising where things could have gone sideways until far after the fact. But in this case, the potential for bias in the algorithm was apparent well before the results were in.
Even Education Secretary Gavin Williamson acknowledged that high-achieving students from disadvantaged areas were at risk of being unfairly downgraded as a result of how the system was set up. This makes it even more perplexing why this process was given the go-ahead at all.
Granted, these are extraordinary times and extraordinary measures need to be taken in many aspects of our lives as a result of the current health crisis. Figuring out a way to fairly establish A-Level grades in the absence of formal assessments couldn’t have been an easy task. That is one thing we can probably agree on, but relying on such a heavily-biased AI solution to produce results that can have considerable and lasting implications for English pupils’ futures is massively off-base here.
Because of all this, students had absolutely no control over the outcome. Instead, their futures were effectively placed in the hands of a flawed computer algorithm that largely based their results on how others before them performed (and there is evidence it did not do a good job of even achieving that).
The process sacrificed the individual for the majority and worked to completely undermine the potential of the students unfairly downgraded. We can expect a flood of appeals, whose process should be swift, robust, fair, and accommodating. If the appeals process doesn’t do justice to the injustice that befell UK students this year, Downing Street should emulate how officials in Scotland handled the situation and execute a complete u-turn. And do it before the GCSE results go the same way.
Contributed by Attila Tomaschek, Digital Privacy Expert at ProPrivacy