27 Jan 2022 / 12:00 - 12:45

Keynote: The future of AI

Artificial Intelligence has a long history - perhaps a longer history than computer science - and has had many changes in focus, and, if not in ambition, at least in perceived likelihood of success.

It would be hard not to have noticed that recently, AI has been in a period of intense enthusiasm and activity, with the substantial solution of long-standing problems in perception and natural language processing, and with the successful application of AI techniques to solving problems in fields from molecular biology to plasma physics to astrophysics to materials design. We’re even seeing widespread claims of connection to AI from somewhat distant fields and from companies using barely related techniques. And perhaps crucially, the public perception seems to have shifted from widespread scepticism about the future of AI to widespread acceptance that at-least-human-level AI is likely, perhaps inevitable, and that we should plan for it. There is considerable disagreement however, about what advances are necessary and sufficient for general AI, and how these advances might be integrated into human civilisations. In this talk, I’ll cover a little of the history of thinking, provide some thoughts about what capabilities AI might develop and depend on over the next couple of decades, and what models might already be available for thinking about integrating highly capable AI into our societies.