Meg Foulkes discusses public impact algorithms and why they matter.

When I look at the picture of the guy, I just see a big Black guy. I don’t see a resemblance. I don’t think he looks like me at all.

This is what Robert Williams said to police when he was presented with the evidence upon which he was arrested for stealing watches in June 2020. Williams had been identified by an algorithm, when Detroit Police ran grainy security footage from the theft through a facial recognition system. Before questioning Williams, or checking for any alibi, he was arrested. It was not until the matter came to trial that Detroit Police admitted that he had been falsely, and solely, charged on the output of an algorithm.

It’s correct to say that in many cases, when AI and algorithms go wrong, the impact is pretty innocuous – like when a music streaming service recommends music you don’t like. But often, AI and algorithms go wrong in ways that cause serious harm, as in the case of Robert Williams. Although he had done absolutely nothing wrong, he was deprived of a fundamental right on the basis of a computer output: his liberty.

It’s not just on an individual scale that these harms are felt. Algorithms are written by humans, so they can reflect human biases. What algorithms can do is amplify, through automatedly entrenching the bias, this prejudice over a massive scale.

The bias isn’t exclusively racialised; last year, an algorithm used to determine exam grades disproportionately downgraded disadvantaged students. Throughout the pandemic, universities have been turning to remote proctoring software that falsely identifies students with disabilities as cheats. For example, those who practice self-stimulatory behaviour or ‘stimming’ may get algorithmically flagged again and again for suspicious behaviour, or have to disclose sensitive medical information to avoid this.

We identify these types of algorithms as ‘public impact algorithms’ to clearly name the intended target of our concern. There is a big difference between the harm caused by inaccurate music suggestions and algorithms that have the potential to deprive us of our fundamental rights. To call out these harms, we have to precisely define the problem. Only then can we hold the deployers of public impact algorithms to account, and ultimately to achieve our mission of ensuring public impact algorithms do no harm.