STRATEGY

Our short term plan is to try to get several demonstrations organized, one after the other. People have a lot of well justified concern about the dangers of AI that need to be focused and mobilized. We may be able leverage two or three smaller demonstrations in quick succession into larger follow on demonstrations. Our goal is to coordinate with any other group who advocates a Pause to keep the issue in the public mind and build momentum for a Pause.

 In practical terms, Congress needs to pass laws regulating the AI industry in the same way any other industry is regulated that deals with potentially dangerous technology. We need a new federal office set up to regulate AI companies in the same way the U.S. Nuclear Regulatory Commission regulates the nuclear energy industry in the United States.  AI companies should have to get approval for new models and significant changes to current models before releasing these things publicly.

In the slightly longer term, we need to work for a world Pause treaty. It is in the interest of everyone everywhere for us to Pause AI development at our current level. But no nation will want to join a world treaty if they feel other countries may cheat. That means the treaty must include very strong enforcement and verification provisions. Once again, there is a successful international nuclear regulation precedent we can follow in the International Atomic Energy Agency. We need an international treaty that establishes a worldwide AI control organization with strong, even unprecedented (but narrow) enforcement powers over AI development.  Tracking and controlling the use of every existing high powered computer chip is huge undertaking that just so happens to be well with in the capability of our current AI.

Preventing unauthorized AGI development would have an unfortunate but relatively small effect on certain civil liberties such as the number of super computer GPUs one person or company could own without registering with the appropriate national and international AI control authorities. The number of average citizens impacted by this regulation would be small to none, since it is only very high end supercomputing chips that are a concern. We advocate that mechanisms be set up that allow every super computer chip to be tracked at all times over the internet. Large data centers would also have to be monitored to insure they were complying with the anti-proliferation regulations. This would allow the authorities to know where every chip was and who was using it. Again, AI is ideally suited to perform much of this work. There other things that need to be done as well, and can be done. A Pause is technically feasible. Those who argue that it won’t work are, in almost every case, people who don’t want to see it work. But it can work and it is our best choice right now.

To get a world treaty on AI cooperation will take much work and education in every country. Our midterm strategy is to set up local chapters of the World Pause Coalition in every country as quickly as possible by coordinating with as many existing NGO’s worldwide as possible. Using donated worker time from existing organizations would potentially speed up the organizing process by 1 or even 2 orders of magnitude. This is important because the Overton window for an international treaty is limited.