As the hype around AI approaches fever pitch, most of our concerns centre around what to do if it gets too clever, too quickly. In that scenario, fears range from AI taking all our jobs, to AI deciding that we need to be culled en-masse for the good of the planet. But what if AI remaining too stupid for too long is the real danger?

Most of us have encountered a chatbot getting things wrong. They are often wrong about small things – for a long while, they couldn't tell you the number of times the letter R occurs in the word "strawberry" – but they can also be wrong in more spectacular fashion, like making up entire legal cases from thin air and fooling a hapless lawyer into using these cases in his client's lawsuit.