newsfilter
← Interviews
80,000 Hours2h 13m

Flaws that make AI architecture unsafe & how to fix them | Stuart Russell (2020)

Stuart Russell, Rob Wiblin

Interview

Stuart Russell argues that the standard AI model of optimizing fixed objectives is fundamentally flawed, proposing instead a "Human Compatible" framework where systems remain uncertain about human preferences to ensure safety. He warns that without this paradigm shift and new regulatory standards, humanity faces existential threats from autonomous weapons, surveillance, and the gradual erosion of human autonomy. Russell emphasizes that while superintelligent AI could dramatically increase global wealth, achieving this safely requires immediate technical breakthroughs in alignment and strict government oversight of algorithmic deployment.