AI 2040
Supplements
We thought about many things in order to write this scenario. We believe in transparency, so we're publishing our thinking here in the form of various supplements, even though some of them are unpolished, scrappy, and/or guesswork. We tried to indicate at the top of each how confident we are in them.
FAQ
Answers to some of the most common objections and questions about Plan A that we have received so far.
How Plan A solves our 5 biggest problems
How the plan addresses the five biggest risks from AI.
Covert AI Projects
How much do hidden AI projects threaten the deal, and how we can prevent, detect, and deter them?
Verification Plan
How can the US and China enforce an AI slowdown without having to trust each other?
Transparency Plan
How can relevant information be shared with decision-makers and the public while avoiding downsides?
Capability Scaling Strategy
The Consortium scales to top-expert AI by 2035, then slows nearly to a pause before handoff.
Security in Plan A
What security is needed for Plan A in model weights, verification integrity, and algorithmic secrets.
Security in Plan A
What security is needed for Plan A in model weights, verification integrity, and algorithmic secrets.
Plan A Assumptions
Core forecasts and recommendations the scenario rests on, and how confident we are in each.
Comparing Possible Plans
Scoring alternative plans on metrics we care about and our odds of them leading to a good future.
Deal Decline
How the risk of the deal weakening over time shapes how fast to scale capabilities.
Takeoff Supplement
The modified AI Futures Model behind Plan A's trajectories, with an interactive explorer.
Economics of Plan A
Why we expect full automation and explosive growth from AI and summaries of our economic model.
Compute in Plan A
Justification of the compute numbers in our scenario.
Economic Growth Explorer
An interactive explorer of our economic model.
Space Governance Plan
How the resources of space could be allocated in the longer term future.
AI for Epistemics
How we can use AI to improve human judgment and collective decision-making.
Alignment Roadmap
Our best-guess roadmap of the alignment objectives needed to avoid AI takeover.