I do AI research in the retail industry, and I used to be a physicist. In this substack, I’m trying to get my head around the coming AI tsunami by working through a new research agenda tackling catastrophic risk.
A quick summary of where I’ve got to: I am looking into how AI’s behaviour changes as its environment changes, including when it can continuously learn. This is downstream of my belief that to align AI in the long-term, we need to create a societal ecosystem that provides it continual feedback, keeping it inside an acceptable subspace of possible values and behaviour. I am currently reading existing research in this area, with a view to reproducing and extending it.
For more information, start by reading the Introduction and about my Generating Process.
I welcome questions and criticism, as these will help me improve my ideas, so feel free to add comments under my posts. If you have a comment but don’t want to put a name to it, I have an anonymous feedback form.
