Human Compatible
- FINISHED
- Finished February 2026
- First published 2019
artificial intelligence ethics
Argues that AI systems should be built to be uncertain about human preferences, as a route to keeping powerful AI beneficial and under human control.