Interruptibility graphic

New paper: Safely interruptible agents

Deepmind and FHI collaborate at UAI 2016