Interview
When elites have much smarter AI than you do
- Future political candidate nominations may be increasingly decided by small, biased, automated groups within parties run by big tech, restricting choices to those who approve of big tech while offering only enhanced state handouts as a political platform.
- A small oligarchy or handful of individuals could de facto control global political decisions, shape foreign policy, determine investments, and manipulate the information environment, potentially leading to scenarios where the U.S. or the world is dominated by executives of a single AI company following an intelligence explosion.
- AI capabilities are expected to enable organizations to field millions of efficient, 24/7 human-like workers, causing entities with superior AI to expand 10x to 100x in size and gain massive strategic advantages over competitors.
- AI progress will likely concentrate wealth and power by creating a direct feedback loop where money easily converts to labor, allowing wealthy actors to gain vast workforces and leverage that economic power to dominate political and social spheres.
- Society's epistemic capabilities may erode due to the rapid pace of technological change, the use of AI intermediaries that filter information, and deliberate obfuscation by powerful actors, making it difficult for the public to understand or counteract power concentration.
- Extreme power concentration scenarios include "power grabs" where CEOs implant secret loyalties into future AI generations to seize government control, or heads of state automate public services and mandate loyalty to personal rule to cancel elections or establish dictatorships.
- Economic displacement from automation could render human labor valueless, reducing average human leverage and potentially causing mass starvation if power concentrates in few companies or countries, though some scenarios may involve an enlightened welfare state dependent on state handouts rather than labor.
- Government incentives to represent the populace may erode if revenues become heavily reliant on the AI sector, potentially creating resource-curse-like dynamics where states become exploitative, while checks and balances may fail because they evolved for human speeds rather than AI speeds.
- Strategic interventions recommended include improving epistemics through transparency, sharing compute access, ensuring legislative and judicial branches have access to superior AI, and training law-following AI for government systems to prevent alignment failures and coups.
- The outlook acknowledges a spectrum of outcomes from messy, continuous dynamics consistent with today's world to extreme cases of single-company hegemony, with the speaker expressing greater concern about slow-moving economic disempowerment and structural erosion than sudden intelligence explosions.