How can we design AI systems that reliably act according to the true intent of their users, even as the capability of the systems increases?

This seminar group will hold weekly discussions that explore the question of AI safety from a technical angle, discussing papers from OpenAI, DeepMind, and academia. The format will be a short presentation of a paper followed by questions and discussion. We’re hoping these meetings will also lead to new research collaborations.

A basic understanding of machine learning is helpful, but detailed knowledge of the latest techniques is definitely not required! Some suggestions of papers to cover are here: https://docs.google.com/document/d/1gdJdgIG-qXPLrD9ZjHZB8LsJNpQX4MtQHQPnb07jbq8/edit?usp=sharing

Put yourself down here https://docs.google.com/spreadsheets/d/17JxduAhr1BaAG5W5ZbQEfJVizuADhC7U25QYx7uIdHw/edit#gid=0 or get in touch with Beth (eb642) if you’re interested in presenting a paper. There will be free pizza!!