In this episode, we explore philosopher Nick Bostrom's proposal for creating AI that understands and pursues human values. We discuss how such AI could be motivated to act based on predicted human approval, ultimately harnessing its intelligence to address the complex problem of value-loading.