Finally, what about tales of AI robot uprisings? OK … maybe that’s a little far-fetched. But the real question is: can we trust AI? How can you know that an AI model will not give you a false summary of Donoghue v Stevenson [1932] UKHL 100, perhaps because it is evil or perhaps just for a laugh?

Well, the specific concern is not related to the risk that these machines gain enough consciousness to develop their own (evil) objectives, and, as argued by Stuart Russell, is instead related to the competence of the machines.

In this area, Russell has introduced the idea of beneficial machines, ‘machines that present no threat to us, no matter how intelligent they are’. Essentially, we need to look at programming these models to work towards human preferences – and we, humans, need to be the ones informing AI of what those preferences are.