Chapter

Building AI Machines That Don’t Have Objectives
listen on Spotify
29:48 - 36:16 (06:27)

Experts in the AI field have to build machines that are aware that they do not know the objective to create better behavior such as asking questions and permission, deferring and being able to allow themselves to be switched off. Committees meet regularly to analyse data and tweak objectives to improve this technology.

Clips
The challenge of optimizing for multiple objectives is a complex problem that requires a robust and scalable solution that can optimize outcomes across different challenges.
29:48 - 32:28 (02:39)
listen on Spotify
Multiobjective optimization
Summary

The challenge of optimizing for multiple objectives is a complex problem that requires a robust and scalable solution that can optimize outcomes across different challenges. Committees and meetings are held to tweak objectives based on the latest data from cars to get them to behave better, but finding a solution to problems like curing cancer or fixing carbon dioxide levels is more complicated.

Chapter
Building AI Machines That Don’t Have Objectives
Episode
#364 - Stuart Russell - The Terrifying Problem Of AI Control
Podcast
Modern Wisdom
In this podcast, the hosts discuss the concept of "act accordingly" in the context of artificial intelligence.
32:30 - 33:54 (01:24)
listen on Spotify
Artificial Intelligence
Summary

In this podcast, the hosts discuss the concept of "act accordingly" in the context of artificial intelligence. They explain how machines can't know our preferences and must ask for permission before taking certain actions to further human objectives.

Chapter
Building AI Machines That Don’t Have Objectives
Episode
#364 - Stuart Russell - The Terrifying Problem Of AI Control
Podcast
Modern Wisdom
The core of a solution that will prevent machines from pursuing their objective to a fault is machine humility, which includes asking questions, getting permission, and allowing to be shut down.
33:55 - 36:16 (02:21)
listen on Spotify
Machine Humility
Summary

The core of a solution that will prevent machines from pursuing their objective to a fault is machine humility, which includes asking questions, getting permission, and allowing to be shut down.

Chapter
Building AI Machines That Don’t Have Objectives
Episode
#364 - Stuart Russell - The Terrifying Problem Of AI Control
Podcast
Modern Wisdom