Hi Daniel, thank you for that thoughtful response. I agree with this articulation:
“If a particular middle power doesn't believe that ASI/RSI/LoC are real serious possibilities in the near term, then for the reasons you mentioned, it'll be really hard and unlikely for them to do what I recommend, even if (say) the PM himself totally does believe. Like you say, he might believe himself, but if other politicians, the media, the public don't believe, then maybe he'll feel like he doesn't have enough support to go through with it.“
I think the final sentence is critical and too often overlooked in the discourse, which too often focuses on ‘our poll says XX% of the population doesn’t want superintelligence” or similar, and then stops there, rather than articulating what trade offs now exist for politicians or Governments to act.
And I like your #2 recommendation, applied to one Middle Power at a time. I worry that the companies themselves will note this is happening, and simply play Middle Powers off against each other/call on US to exert leverage, but that shouldn’t stop a nation from approaching this strategy.
I would also note that many Middle Powers have signed up to Pax Silica’s most recent offering, the SF Declaration, which would indicate to me that they are more interested in aligning with US views on this than perhaps passing some of the above. My read is that these governments accept AI is a powerful technology and is coming with more down the track, and have responded by moving closer to Washington, so they are now bandwagoning (siding with the stronger power to share its gains) which would also make retreat more costly and less likely.
I think a couple of further points.
“In politics, it's normal for people to say something is an existential threat, even when it totally isn't. Alas. Because once every century or so, something comes along that really actually is an existential threat, and our political leaders are slower to react than they should be.”
Yes indeed, which is I think fairly critical to the framework you apply here. Let us take your super plague outbreak example and apply it to the UK. You fairly argue that a PM may close borders and worry about US response afterwards, prioritising the health of the population over the geopolitics. I just don’t see this as playing out like this for AI risk, and arguably we would struggle to see a recent example of this being the case in general (Covid is a useful example here, look how long Britain took to close borders, even watching their peers close theirs - and we literally watched fatalities go up and up.) So I think without literally deaths you wouldn’t see this approach work. This goes to my earlier points: you will have a dozen experts, the best in their respective fields, briefing the PM/President. They will likely have shades of differing views. When we close borders for plague, we hope we get no new arrivals with plague. When we burn US AI bridge, what do we hope to show in return to our population, and what actual impact would it have on the direction of the technology being built in SF?
The second thing missing is even deeper: the psychological profile of the leader. I appreciate this is hard to model in general terms, but again, let’s use Britain as an example. During the pandemic, Britain’s Prime Minister was Boris Johnson. He was criticised for having an approach to governing called ‘Cakeism’: in simple terms, having your cake and eating it. This manifested in usually waiting and wearing until various options ran out, then picking the last few remaining options. Various Prime Ministers have since taken this approach with different forms. This may be because making clinical trade offs is difficult and unpopular, and by their nature, politicians who rise to this position have to understand the criticality of importance and how to keep various camps content or motivated.
Thank you for your time!