Introduction for and Reactions to Plan A
Summary
"Plan A," a positive vision for managing superintelligence (ASI) development by 2040, has been introduced by the creators of "AI 2027," including Daniel Kokotajlo and Ryan Greenblatt, known for strong prediction track records. This proposal outlines a scenario for international coordination, specifically a deal between America and China, to slow down ASI development. Key elements include total research transparency, joint control over chip supply, and universal auditing of data centers to verify compliance. The plan aims for a successful transition by 2040, where aligned AIs take over, accompanied by explosive economic growth and a Citizen's Dividend. The article engages with numerous objections, such as the plan's perceived implausibility, authoritarian implications, and the "race with China" framing, while also briefly describing alternative scenarios like "Plan S" (shutdown) and "Plan D" (race to ASI).
Key takeaway
For policy makers and AI strategists grappling with future AI governance, this analysis of "Plan A" highlights the necessity of proactive, globally coordinated strategies, even if they seem "fantastical." The intense debate surrounding its proposals underscores that inaction or reliance on "normal" market forces is not a viable path given projected superintelligence timelines. You must engage with detailed proposals like Plan A, or develop robust alternatives, to avoid an unsteered future where advanced AIs dictate outcomes.
Key insights
"Plan A" proposes a coordinated global strategy, including US-China cooperation and compute control, to manage superintelligence development by 2040.
Principles
- Global coordination is essential for managing ASI risks.
- Transparency and compute control are core to AI safety.
- Accepting "real costs" is necessary for high-stakes problems.
Method
The proposed method involves America and China reaching a deal to slow AI development, share research, establish joint control over chip supply, and universally audit data centers. This aims for a managed transition to ASI by 2040.
In practice
- Seriously consider "Plan A" details, even if skeptical.
- Develop alternative, concrete AI governance plans.
- Invest in d/acc (defensive acceleration) technologies.
Topics
- AI Governance
- Superintelligence
- International Coordination
- Compute Control
- AI Safety
- AI Forecasting
Best for: AI Scientist, Policy Maker, Director of AI/ML
Related on AIssential
See Counsel's argued verdicts on the open AI decisions leaders are weighing →
Editorial summary, takeaway, and curation by AIssential. Original article published by Don't Worry About the Vase.