The Cognitive Revolution podcast presents a cross-post from the Future of Life Institute podcast, featuring Gus Docker interviewing Tom Davidson about the risk of AI-enabled coups. Tom argues that powerful individuals could use AI to seize power, rather than AI systems rising up independently. He outlines necessary AI capabilities, including persuasion, strategy, cyber offense, and autonomous military robots, and segments the threat landscape into singular loyalties, secret loyalties, and exclusive access. Tom also discusses historical precedents and the potential for AI to automate AI research, exacerbating the risk. He estimates a 10% chance of an AI-enabled coup in the next 30 years and proposes mitigations like system integrity measures, distributed control of military AI, transparency, and programming AI to maintain democratic checks and balances.
Sign in to continue reading, translating and more.
Continue
