AI 2040: Plan A - the positive vision

#The Core Proposal

Instead of a secretive corporate race to artificial general intelligence (AGI), Plan A advocates for an international treaty (primarily between the US and China) to coordinate a verified slowdown of frontier AI development.

• The Target: Delay the arrival of superhuman AI until 2040.
• Total Transparency: Make all frontier AI research public so global compliance can be easily monitored and enforced.
• Multipolarity: Allow multiple global companies to catch up to the frontier, preventing a single tech monopoly or dictator from controlling superintelligence.

---

#The Proposed Timeline

• 2029: The US and China sign a historic treaty to halt the reckless race to superintelligence.
• 2030–2035: AI capabilities are restricted to the level of top human experts (preventing recursive self-improvement).
• 2035–2040: A strict global pause is maintained to finalize safety and alignment protocols.
• 2040: The pause is lifted, and humanity safely transitions to superintelligence.

---

#Immediate Policy Recommendations

To pave the way for this treaty, the authors recommend:

• Deployment Caps: Keep internally deployed AIs close to publicly audited capabilities.
• Stricter Export Controls: Better enforce hardware restrictions to prevent chip smuggling.
• Inference Verification: Invest in tech that verifies compliance without halting economic use of existing models.
• R&D Compute Limits: Cap the percentage of supercomputing power allowed for training new models.

Sources: @DKokotajlo (Daniel Kokotajlo)