You can think about maximizing the positive impact of a policy by evaluating it along two axes: effectiveness and feasibility. Say your goal was to reduce carbon emissions. An immediate global ban on fossil fuel extraction would eradicate emissions overnight, but is all but completely unenforceable. No economy could withstand the shock; no major emitter would comply. On the other hand, voluntary self-reported emission pledges are easy to agree with—but precisely because they are impossible to enforce. Any country and company could happily sign while continuing to emit. In other words, a policy that is effective but unfeasible is bad; a policy that is feasible but ineffective is also bad.
The 1963 treaty banned nuclear tests in the atmosphere, in space, and underwater, but not underground. The reason is pure verification design. Existing means, monitoring stations picking up radioactive debris and, later, satellites and seismic arrays, could detect atmospheric tests worldwide without any inspection inside the other country. Those means could not reliably distinguish underground tests from earthquakes at the time, and such tests would have required on-site inspection the Soviets would not grant. So the treaty covered exactly the environments that national technical means could police and left out the one they could not.
In the following exercise, you will develop intuitions with this effectiveness-feasibility tradeoff framework in the context of ASI development. Should governments agree to a more moderate slow-down or a more drastic pause? Should you compromise ambition or real-world enforceability?
Sort the following policy buckets on the feasibility x effectiveness matrix.
Design for the hardest case. Verification strong enough to support a full pause — chip registries, compute metering, inspection rights — supports every weaker bucket for free. The reverse is not true. That is why this track studies verification against the pause, even if what gets signed first is transparency.
The desk is a JavaScript surface. If this line stays, its scripts did not load — a reload usually fixes it.