Rogue AI developers as hostis humani generis: under an international AI treaty, should they be subject to international concurrent jurisdiction? — LessWrong
An international treaty banning superhuman AI will, necessarily, require the arrest of rogue AI developers who violate the treaty. But who should be authorized to conduct these arrests? Should there be some sort of international concurrent jurisdiction, allowing at least some countries to conduct arrests of defectors within other countries’ territory?
The default approach - allowing countries to make such arrests only within their own borders - may quickly cause the treaty to fall apart. If national governments operate under the thesis that anyone who achieves superhuman AI (artificial superintelligence, or ASI) will immediately have dominion over the rest of the world, any government has reason to defect and begin racing to build ASI if they even suspect any other country is also defecting (for instance, by choosing not to arrest their own citizens who are defecting and racing to ASI). As a result, bad intelligence collection by a single country could lead to the collapse of the treaty, even with nobody actually defecting beforehand.
Another approach is inspired by the hostis humani generis doctrine applied to pirates during the Age of Sail, in which any seagoing country had the right to interdict pirates and try the pirates in their own admiralty courts, regardless of the pirates’ nationality. Under this system, any country would be permitted to conduct arrests of rogue AI developers within any other such country, even though this would mean operating unilaterally within another country’s territory. This solves the previous approach’s core problem, but introduces many new problems of its own. One country might decide to fabricate evidence, or use spurious evidence, as justification to arrest all of another leading country’s known AI researchers. Forces of one country operating on another country’s territory unannounced, or with short notice, could easily lead to a completely unnecessary shooting war that a better-structured treaty would avoid entirely. The original hostis humani generis doctrine applied to pirates at sea; arresting another country’s pirates in international waters is a much less fraught matter than arresting another country’s AI researchers within that country itself.
As both of those approaches fail, an alternative is needed. I propose that a broad coalition of middle powers, independent of the US and China, stand up and maintain a well-trained, well-resourced force capable of making such arrests across borders. All countries would be bound by treaty to tolerate this force’s operations within their own borders, under certain criteria. This coalition itself would appoint a smaller set of mediator countries, with long and trusted records of fair dealing in fraught geopolitical circumstances, to adjudicate evidence and determine whether the middle powers’ force may proceed with any particular action.
An international treaty banning superhuman AI will, necessarily, require the arrest of rogue AI developers who violate the treaty. But who should be authorized to conduct these arrests? Should there be some sort of international concurrent jurisdiction, allowing at least some countries to conduct arrests of defectors within other countries’ territory?
The default approach - allowing countries to make such arrests only within their own borders - may quickly cause the treaty to fall apart. If national governments operate under the thesis that anyone who achieves superhuman AI (artificial superintelligence, or ASI) will immediately have dominion over the rest of the world, any government has reason to defect and begin racing to build ASI if they even suspect any other country is also defecting (for instance, by choosing not to arrest their own citizens who are defecting and racing to ASI). As a result, bad intelligence collection by a single country could lead to the collapse of the treaty, even with nobody actually defecting beforehand.
Another approach is inspired by the hostis humani generis doctrine applied to pirates during the Age of Sail, in which any seagoing country had the right to interdict pirates and try the pirates in their own admiralty courts, regardless of the pirates’ nationality. Under this system, any country would be permitted to conduct arrests of rogue AI developers within any other such country, even though this would mean operating unilaterally within another country’s territory. This solves the previous approach’s core problem, but introduces many new problems of its own. One country might decide to fabricate evidence, or use spurious evidence, as justification to arrest all of another leading country’s known AI researchers. Forces of one country operating on another country’s territory unannounced, or with short notice, could easily lead to a completely unnecessary shooting war that a better-structured treaty would avoid entirely. The original hostis humani generis doctrine applied to pirates at sea; arresting another country’s pirates in international waters is a much less fraught matter than arresting another country’s AI researchers within that country itself.
As both of those approaches fail, an alternative is needed. I propose that a broad coalition of middle powers, independent of the US and China, stand up and maintain a well-trained, well-resourced force capable of making such arrests across borders. All countries would be bound by treaty to tolerate this force’s operations within their own borders, under certain criteria. This coalition itself would appoint a smaller set of mediator countries, with long and trusted records of fair dealing in fraught geopolitical circumstances, to adjudicate evidence and determine whether the middle powers’ force may proceed with any particular action.
Full analysis here: https://connorsscratchpad.substack.com/p/rogue-ai-developers-hostis-humani