Four: Brian Christian on artificial intelligence
Effective Altruism: Ten Global Problems – 80,000 Hours (October 2021)
00:00
Is There a Reward Function?
The hope is that this would cause them to become corrigible. This means being willing to be corrected because iy've made a mistake. So it's like correctible, i guess. Yet why the existing traditional systems may be over confidentd and resistant to correction or being turned off. And so people like stuart russell, for example, have been thinking about how can we incorporate some notion of uncertainty or doubt?
Transcript
Play full episode
