Researchers Propose 'Plan A' to Delay Superintelligence Until 2040, Demand Full Research Disclosure
The AI Futures Project research organization proposes a proactive plan to manage risks from advanced AI, calling for development to be delayed until 2040 and all research to be publicly disclosed to reduce dangerous competition.
The non-profit research organization AI Futures Project has published a significant policy proposal titled "AI 2040: Plan A," which presents a radically different approach to managing the risks of future advanced artificial intelligence. Its lead authors are Daniel Kokotajlo, a former OpenAI researcher, and Thomas Larsen, co-author of the previously noted AI 2027 report. The core of "Plan A" is not a prediction, but a recommendation the research team believes "should happen" to guide humanity towards the best possible outcome. The plan calls for delaying the development of Superintelligence (artificial intelligence that is smarter than humans in all aspects) until at least 2040. A key proposed measure is to mandate all AI laboratories to publicly disclose all their research and models to eliminate secret competition and foster transparency in the field. The proposal further states that full disclosure would enable dozens of companies and research organizations worldwide to catch up with the leading groups' technology, leading to decentralization and preventing any single entity from monopolizing power. Another key concept is the deliberate creation of a state called "mutually assured compute destruction," a principle similar to "mutually assured destruction" from the Cold War era, but replacing nuclear weapons with compute power. Under this, if any party attempts to secretly accelerate AI development, the remaining parties, with their publicly disclosed compute power, can collectively deter or retaliate, thereby discouraging anyone from violating the agreement. This report is the third significant work from the AI Futures Project, following AI 2027 and the AI Futures Model. The project's website is also developed in collaboration with Lightcone Infrastructure. This proposal has been confirmed and reported by leading news outlets such as Axios and the Washington Post, indicating it is being seriously considered in global technology and policy circles.
This proposal directly challenges current AI development approaches, which are heavily focused on fierce competition among a few tech giants. If adopted, it could transform the global AI landscape to be more open and secure, potentially benefiting various countries, including Thailand, by ensuring equitable access to technology and knowledge.