British selectivity exposes the shortcomings of the algorithms

Popular anger has forced the British Government to give up an algorithm that calculated the probable scores for selectivity. It's a reminder that even well-intentioned algorithms make many harmful mistakes.

The exams were canceled due to the pandemic. The furor erupted after nearly 40% of students scored lower than expected by their teachers, and students from disadvantaged schools were disproportionately affected.

It is an unusually harsh and public example of a much bigger problem. For example, the models used for lending can make it difficult for less wealthy people to borrow at reasonable rates.

Algorithms are also used in many states in the US to help decide the likelihood of a repeat offender. They are supposed to neutralize judges' biases, but their input, for example if the convict's friends have been arrested, may introduce other biases, as described by mathematician Cathy O'Neil.

She talks about proxy variables , that is, inputs that substitute for knowledge of a person's actual behavior. In the British Selectivity algorithm, one of the proxies was the student's school record. It was a surefire way to discriminate against star students from less affluent backgrounds.
As so often, the objective was rational: to adjust the expected ratings to bring them closer, in general, to the distribution of results from previous years. Authorities had months to consider the possible unforeseen consequences of their algorithm. Like careless teenagers, they wasted their time.

Stories of kids' homelessness do well on TV, and protests from parents, teachers, and a generation of future voters made the government's initial refusal to budge short-lived, predictably. Now the grades estimated by the teachers will be valid. I wish it were so easy for prisoners, for example, to make themselves heard.

The authors are columnists for Reuters Breakingviews . Opinions are yours. The translation, by Carlos Gómez Abajo, is the responsibility of CincoDías