AI and human agency, Professor Christopher Summerfield

AI and human agency

Narratives in AI safety mostly focus on the growing capability of AI models. For example, there are concerns that highly intelligent AI models could be used for criminal or disruptive purposes, may transform labour markets, or even seek to attain their own misaligned goals. In my lecture, I will argue that the most pressing question in AI safety is not artificial intelligence but artificial influence - the many ways that AI can be used to influence people. The widespread embedding of AI in digital platforms, applications and websites opens the door for a highly automated ‘influence economy’ in which conversational AI systems compete to directly influence our commercial and political choices. I will discuss with reference to empirical work showing that current AI systems can be highly persuasive, socially perceptive, and effective in parasocial relationship-building. I argue that we urgently need to consider how to build AI that enhances rather than degrades human agency.

Know what’s Happening Next — before everyone else does.

Get App

Host or Publisher The Institute for Ethics in AI