Foundational Beliefs
AI safety expert argues we have just 4 years before humanity's fate is sealed by superintelligence.
Against Moloch's 'Foundational Beliefs' article, posted on LessWrong, presents a sobering framework for AI safety strategy grounded in 2026's geopolitical realities rather than idealistic scenarios. The author argues that with a 50% chance of superintelligence by 2030 and key decisions likely within the next 4 years, traditional approaches relying on government regulation or international cooperation are insufficient given current political dynamics. The piece specifically references the recent Department of War (DoW) conflict with Anthropic as evidence that government-led solutions face serious viability challenges.
The article outlines six foundational beliefs: short timelines (25% AGI chance by 2027), resolved open questions (AGI development will occur under Trump/Xi leadership by Anthropic, OpenAI, or DeepMind), high variance futures (Taiwan invasion scenarios), need for portfolio strategies (no single solution works across all possible worlds), game theory centrality, and inevitable tough tradeoffs. The author emphasizes that strategies must account for populism, distrust of big tech, job concerns, and potential US-China conflict while developing multiple contingency plans rather than seeking one perfect solution.
- 50% chance of superintelligence by 2030 with key decisions likely within 4 years
- AGI development expected under Trump/Xi leadership by Anthropic, OpenAI, or DeepMind
- Portfolio approach needed as no single strategy works across all geopolitical scenarios
Why It Matters
Shifts AI safety discourse from theoretical ideals to actionable strategies within current political constraints and urgent timelines.