IT IS hard to go a single day without hearing about the two huge crises that humanity is grappling with right now, the covid-19 pandemic and the climate emergency. In both cases, science and technology have been crucial in identifying the problems and their possible solutions.
Those two issues might seem like quite enough to be going on with, but we shouldn鈥檛 take our eyes off another troublesome area in which the role of science is vital: the rise of algorithms.
Advertisement
We might hear less about them, but algorithms are just as hard to avoid as talk of the pandemic. Constantly operating in the background of our digital lives, they do a huge variety of jobs, suggesting what we should read, watch and buy online.
They are also increasingly used to help us make tricky decisions offline. The trouble with this is that the workings of algorithms, especially those based on artificial intelligence, are often impossible to fully understand. We outsource all kinds of decisions to computers, yet can鈥檛 easily see how these were made.
鈥淲e outsource all kinds of decisions to computers, yet can鈥檛 easily see how these were made鈥�
One instance came last year, when Ofqual, the regulator of exams in England, had tough decisions to make about assigning grades to pupils who had their exams cancelled due to the pandemic. It decided to ask teachers for their assessments of pupils鈥� performances and then moderate these using an algorithm.
The hope was to avoid wild grade inflation. The result, however, was that many students ended up with drastically worse results than they had expected, with 鈥� initially, at least 鈥� little explanation.
Maybe it is time to admit that we need a healthier relationship with algorithms, one where we understand the basics of how they work. A good first step would be to get to know a few of the algorithms that really matter in our daily lives, which is why we decided to do just that.
The trouble that algorithms cause is nowhere near the scale wrought by the pandemic and global warming. But who knows where we will end up if we carry on delegating decisions to machines we can鈥檛 completely understand?
