Wyomingian Senators in the Year 3000

transcript

So I'm interested in the state of Wyoming. I learned a lot of things about the demonyms for different states preparing this talk. It turns out people from Wyoming are called Wyomingians, which was not previously relevant to me because I didn't know anyone from Wyoming.

A fun fact is that currently Wyoming has as many senators as California does, despite the fact that it has about 150 times fewer citizens. I'm interested in understanding whether in the year 3000 this will continue to be true and whether the basic Wyomingian advantage will persist.

Here's my basic hot take. It seems plausible to me that a core dynamic of how the future goes is that right now people have much worse options for locking current power structures in place.

It seems to me like historically there are very few examples of anyone handling this kind of problem well. Roughly speaking, there are just not very good systems of coordination between states that allow them to lock in current power structures given predicted power shifts in the future. And this is basically just because it's hard to find some unbiased party who's going to make decisions on your behalf in the future.

Athens and Sparta would love to set up some kind of governing body that's going to prevent them from using violence to back out of this agreement, and that's going to settle how decisions are made on questions of interest to both Athens and Sparta. One of the key steps here is that, as part of signing up to this agreement, it has to be difficult for both parties to get out of it, even as the underlying hard power shifts. Because otherwise you can't get out of your core problem where once Athens is much more powerful, they can use their hard power to renegotiate.

But it seems plausible that this is very different once you have powerful AI. Because you can basically make an agreement where I say to my counterparty: both of us are going to make it so that our AIs refuse to engage in violence.

One thing I'm assuming throughout all of this is that people have AIs that they correctly believe are aligned with their interests. Seems kind of implausible — it could happen, might not happen. That's what I'm going to talk about here.

So I'm basically imagining a world where we have AIs that are enforcing that multiple actors act in ways determined by a governing body. AI feeds into this in two ways. First, it's more possible to have AIs that you are comfortable allowing to enforce that you do not have the ability to do violence. And second, you're more comfortable delegating authority to a federal body because you're now more able to have AIs control that federal body, and you can be more confident that it will make decisions that you like in the future.

So AI is making these two steps cheaper. It's making it so you're less afraid of allowing a third party to remove your ability to do violence. And you're also less afraid to centralize decision-making power somewhere else.

So I want to make this basic prediction. In cases where people predict that hard power is going to shift in the future, then the parties who think they are going to relatively lose out are going to push much harder on trying to cause durable mechanisms for power sharing, where the power is shared according to current power divisions. Whoever expects to lose should push for that resolution right now. You do some reasoning and bargaining to figure out relative power, and then just set that as a long-term power ratio.

To get back to the great state of Wyoming: a prediction I want to make is that it seems quite likely that Wyoming and other states are strongly motivated to gang up against states that are going to become relatively more powerful in the future, to make it harder for those other states to secede. Wyoming has a good thing going where they can extract a bunch of pork from other states and they can enact regulations that affect other states. They don't want to lose that. I think that by default they would be right to fear that they're going to lose that in the future, and I think it's reasonably likely that they should try to lock in their power right now.

There are a few different ways you could do this. You could make it so it's harder for other states to escape from control by the Senate. But you could also do something where you more or less dissolve the Senate and do a one-time payment to Wyomingians so that they have an equivalent share in whatever institution you're using to collectively make decisions.

There are very few good historical analogies for this kind of problem. We have lots of examples of cases where there's some democratic deliberative body that has representatives from different groups with power allocated in a particular set of ways, and then some of those groups become much more relatively powerful over time. There are just very few analogies that quite have all the moving pieces here.

Some classic ones people will point out: France made it onto the UN Security Council sort of by coincidence even though France was really a waning power at the time. And China made it on for semi-random reasons. But this only led to France having some limited amount of hard power — it's not actually that efficacious overall. The US tried to implement a bunch of international institutions right after World War II when it was at the peak of its power, something like half of total world output. That worked to some extent, but I don't think it actually worked that well.

Basically, there are not very many historical cases where it was possible to fix current power structures in a way that wasn't incredibly inefficient by delegating to some source of power that was going to remain static over time.

I'm sure you're wondering about specifically what happens to the Wyomingian senators. It seems to me that even if the power of the Senate is maintained and even if Wyoming's power through the Senate persists until the year 3000, the actual senators themselves end up with substantially less discretionary power — just because it seems quite likely that the Wyomingians are able to use AI assistance to have a much better understanding of what the senators are doing. So I think the actual role seems reasonably likely to become substantially less attractive.

That's everything I want to say. Thanks.