Phase change

With mathematicians talking about something of a phase change in the ability of leading AI models to tackle hard problems (the most spectacular example, of course, being the apparent solution of the Navier–Stokes problem), the road towards AGI appears to be shortening by the day. While recursive self-improvement may still lie in the future, AI is already writing more and more of its own code, and there is clearly room for concern.

Oddly enough, at today’s project group meeting, I realised that some of my colleagues who are much more sceptical about technological solutions to climate change, tending instead to emphasise the need for behavioural change, were also much less worried about AI. The irony is that it feels as if we have come full circle here: I, historically the techno-optimist in the room, have come to fear the future and distrust our ability to regulate the AI industry, precisely because I believe in the transformative capacity of technology.