Oxford: forecasting broadcast

Today I went on the 80k podcast. For some reason Rob was set on talking about AI forecasting, rather than speculative theories of human social interaction or weird things that happened in my childhood. So I was nervous about it.

I figured I could probably be less nervous if I just filled in all of the holes in my knowledge of anything that might come up and be embarrassing to not know about. For instance, all work I have previously done, and why it is a good idea to do this kind of thing anyway.

So I set about that. I figured this was not too much preparation, since these topics seem valuable to know about regardless.

I kind of enjoyed the type of stress this produced. It was a bit like a deadline, but without a particular end point that I had to get to. So I could potter about, looking into things and filling in gaps, and feel like every step was concretely valuable and time sensitive, but not as if the whole day’s work was at stake from doing a tiny bit more or less, and making it over some line.

I didn’t finish filling in many of the holes, but I think it went okay nonetheless. Hopefully the listeners will take my repeated claims that I don’t know things that sound incredibly simple to investigate as support for my message that there are lots of tractable things to do in AI forecasting, and I don’t personally have the bandwidth to do all of them. And that maybe it would be more appropriate for this to be a field at least as big as say climate forecasting, rather than me and a few of my friends.