Maybe AI Will Happen Outside US/China

I'm interested in the claim important AI development (in the next few decades) will largely occur outside any of the states that currently look likely to lead AI development. I don't think this is likely, but I haven't seen discussion of this claim.^[1] This would matter because it would greatly affect the environment in which AI is developed and affect which agents are empowered by powerful AI.

Epistemic status: brainstorm. May be developed into a full post if I learn or think more.

I. Causes

The big tech companies are in the US and China, and discussion often assumes that these two states have a large lead on AI development. So how could important development occur in another state? Perhaps other states' tech programs (private or governmental) will grow. But more likely, I think, an already-strong company leaves the US for a new location.

My legal knowledge is insufficient to say how well companies can leave their states with any confidence. My impression is that large American companies largely can leave while large Chinese companies cannot.

Why might a big tech company or AI lab want to leave a state?^[2]

• Fleeing expropriation/nationalization. States can largely expropriate companies' property within their territory unless they have contracted otherwise. A company may be able to protect its independence by securing legal protection from expropriation from another state, then moving its hardware to that state. It may move its headquarters or workers as well.
• Fleeing domestic regulation on development and/or deployment of AI.

II. Effects

The state in which powerful AI is developed has two important effects.

1. States set regulations. The regulatory environment around an AI lab may affect the narrow AI systems it builds and/or how it pursues AGI.
2. State influence & power. The state in which AGI is achieved can probably nationalize that project (perhaps well before AGI). State control of powerful AI affects how it will be used.

III. AI deployment before superintelligence

Eliezer recently tweeted that AI might be low-impact until superintelligence because of constraints on deployment. This seems partially right — for example, medicine and education seem like areas in which marginal improvements in our capabilities have only small effects due to civilizational inadequacy. Certainly some AI systems would require local regulatory approval to be useful; those might well be limited in the US. But a large fraction of AI systems won't be prohibited by plausible American regulation. For example, I would be quite surprised if the following kinds of systems were prohibited by regulation (disclaimer: I'm very non-expert on near-future AI):

• Business services
  • Operations/logistics
  • Analysis
  • Productivity tools (e.g., Codex, search tools)
• Online consumer services — financial, writing assistants (Codex)
• Production of goods that can be shipped cheaply (like computers but not houses)
• Trading
• Maybe media stuff (chatbots, persuasion systems). It's really hard to imagine the US banning chatbots. I'm not sure how persuasion-AI is implemented; custom ads could conceivably be banned, but eliminating AI-written media is implausible.

This matters because these AI applications directly affect some places even if they couldn’t be developed in those places.

In the unlikely event that the US moves against not only the deployment but also the development of such systems, AI companies would be more likely to seek a way around regulation — such as relocating.