Very well written horror story! Props :)
1Believe what?
That's kind of the point... people reject premises all the time. And I don't mean in the "this premise seems unrealistic" sense. I mean more in the "I refuse to participate in this thought experiment!" sense. Even when the point wasn't to give a lesson about the shape of reality but to give a lesson about the shape of the reader's mind and how they respond to the thought experiment.
People just hate inspecting their own mind with a passion. It's also common for people not to trust you when you suggest a thought experiment if they can't see where it's going. It's very easy to get an anger reaction this way (literally raised voice, tense muscles, raised heartbeat type of reaction).
For concreteness, let’s say that the world model requires a trillion (“N”) bits to specify, the intended head costs
10,000 bits, and the instrumental head costs 1,000 bits. If we just applied a simplicity prior directly, we expect to spend N + 1,000 bits to learn the instrumental model rather than N + 10,000 bits to learn the intended model. That’s what we want to avoid.
Not sure if I'm misunderstanding this, but it seems to me that if it takes 10,000 bits to specify the intended head and 1000 bits to specify the instrumental head, that's because the world model - which we're assuming is accurate - considers humans that answer a question with a truthful and correct description of reality much rarer than humans who don't. Or at least that's the case when it comes to the training dataset. 10,000 - 1000 equals 9,000, so in this context "much rarer" means 2^{9,000} times rarer.
However,
Now we have two priors over ways to use natural language: we can either sample the intended head at random from the simplicity prior (which we’ve said has probability 2^{-10,000} of giving correct usage), or we can sample the environment dynamics from the simplicity prior and then see how humans answer questions. If those two are equally good priors, then only 2^{-10,000} of the possible humans would have correct usage, so conditioning on agreement saves us 10,000 bits.
So if I understand correctly, the right amount of bits saved here would be 9,000.
So now we spend (N/2 + 11,000) + (N/2 − 10,000) bits altogether, for a total of N + 1,000.
Unless I made a mistake, this would mean the total is N + 2,000 - which is still more expensive than finding the instrumental head.
I think this is reasonable, which is why we include the more extreme measures including recognition of the right to self-defense. I personally would be surprised if we could throw this together so quickly that none of the conditional deterrence measures ever need to be activated...
In darker timelines, I think the more extreme economic measures could slow down the superpower AI programs and give time for middle powers to get more serious with their military deterrence, which has a good chance of being effective imo.
This is a good point. We didn't have time to address this in the first version of the proposal, but there are potentially some mitigations that can be implemented here, like very heavy penalties for defecting from the agreement.
At the end of the day though... you just must get deep buy in about the x-risks of ASI among middle powers (including softer ones like the possibility of a permanent US and China singleton).
More superficial motivations could be easy to break, but I think it would be difficult to tempt a country where the relevant decision makers think the best case scenario for ASI is for one's state to be completely dismantled by a US singleton (effectively if not literally).