Help us test a way to negotiate global AI safety
Updated: 5 minutes ago
New here?
OR
Artificial intelligence represents perhaps the greatest opportunity and the greatest threat to humanity to emerge in recent years. Watch this video to learn how you can help this opportunity unfold safely.
NOW OPEN: Single Negotiating Framework (SNF) review
Click here to access the progressing versions. This stage will be open until further notice. Tell us what would make this SNF more useful to the party you know best, and help us improve it.
Simpol and Smartsettle are running an open demonstration of how a global AI safety package could be negotiated. Five parties are represented: the United States, China, the European Union, the United Kingdom and a coalition of states that don’t host frontier AI training. Each party’s preferences are kept confidential. Smartsettle searches for agreements that leave every party better off than the status quo, raising the least-satisfied party as far as it can. Simpol lets the parties implement what they agree simultaneously, so no country has to move first.
We’re building the demonstration in the open and in stages, and we need people who understand these parties to help us get it right. You don’t need to be a negotiation expert or commit to weeks of meetings. Most contributions take a few hours, and we keep required reading to about 20 pages at any stage.
What you can do now
The Single Negotiating Framework (SNF) is the agreement itself, with a blank wherever something is still to be decided and the range each blank may take. All five parties negotiate against this one document, so it has to make sense from every side. We’re asking people with different points of view to review it and suggest improvements, especially from the standpoint of the party they know best.
UNDER CONSTRUCTION (some links don't work) - COME BACK SOON
Start with the reader’s guide. It’s short, explains what the framework is and how to comment, and lists the few sections worth reading first.
Reader’s guide (start here)
The full SNF (link is above)
The proposal behind the demonstration (for background)
Submit your comments to info@smartsettle.com (form link coming later) by [date]
What comes next
Once the framework is improved, we’ll publish the revised version and invite endorsements in this form: “This would be a useful instrument if I were representing Party X.” An endorsement covers the instrument only, not Simpol or any policy. After that, we’ll invite contributors to build a model of one party’s preferences, using a worked example and instructions that let your own AI assistant interview you. We’ll combine the submissions for each party into a composite, using the median so that no single view decides the result. Later, with support, groups representing each party will negotiate using those composites.
This page will be updated as each stage opens, and the materials for it will be added here.
Credit and privacy
Contributors who wish to be named will be credited. If you prefer anonymity, we’ll publish only your credentials. We’ll publish what the method achieved, such as whether a consensus was reached and whether every party gained, but not the terms of any outcome. Everything rests on modelled or composite preferences, so the results say something about the method, not about what a real negotiation would produce.
This is a demonstration. The party positions in it are modelled for the simulation and do not represent the position of any government, of Simpol or of Smartsettle.
Spread the word
The demonstration is only as credible as the range of people who take part. If you know someone with first-hand knowledge of one of these parties, please send them this page.
Questions: info@simpol.org
Simpol · simpol.org | Smartsettle Resolutions Inc. · smartsettleresolutions.com
Comments