AI 2040: Plan A
Let's gather and discuss AI 2040: Plan A, the new scenario from the AI Futures Project (the team behind AI 2027). Unlike AI 2027, it's a recommendation rather than a prediction: an international deal in which the US and China agree in 2029 to avoid a race to superintelligence, all AI research becomes publicly transparent, capabilities scale slowly through the human range, and the world pauses at top-human-expert level before deliberately building superintelligence in 2040 enforced by verification schemes and "mutually assured compute destruction" (China's datacenters get built in Canada, of all places). We'll discuss whether the plan is coherent, whether it's achievable, whether it's even desirable, and what we'd do instead.
The full scenario is long. If you can't read the whole thing (And there’s a lot to read), read the main narrative through 2030 plus the summary and Scott's introduction below, and see if you can at least get the main gist of AI 2040. I’ve partitioned up the main reading into 45min-1hour blocks that you can sign up and claim here. If the sheet is full then share a section with someone.
Readings
AI 2040: Plan A (main scenario; skim the supplements that interest you)
Introducing Plan A (Scott Alexander)
If you prefer podcast format (I haven’t watched through it, but I expect it will be a good summary)
Supplemental Readings
AI 2027 (the predicted scenario, for contrast)
Introduction for and Reactions to Plan A (Zvi's roundup of objections and responses)
Comparing Possible Plans (Plans A/B/C/D/S head-to-head)