AI, centralization, and the One Ring

By Owen Cotton-Barratt @ 2024-09-13T13:56 (+15)

This is a linkpost to https://strangecities.substack.com/p/ai-centralization-and-the-one-ring

People thinking about the future of AI sometimes talk about a single project ‘getting there first’ — achieving AGI, and leveraging this into a decisive strategic advantage over the rest of the world.

I claim we should be worried about this scenario. That doesn’t necessarily mean we should try to stop it. Maybe it’s inevitable; or maybe it’s the best available option — decentralized development of AI may make it harder to coordinate on crucial issues such as maintaining high safety standards, and this is a major worry in its own right. But I think that there are some pretty serious reasons for concern about centralization of power. At minimum, it seems important to stay in touch with those. This post is deliberately a one-sided exploration of these concerns.

In some ways, I think a single successful AGI project would be analogous to the creation of the One Ring. In The Lord of the Rings, Sauron had forged the One Ring, an artifact powerful enough to gain control of the rest of the world. While he was stopped, the Ring itself continued to serve as a source of temptation and corruption to those who would wield its power. Similarly, a centralized AGI project might gain enormous power relative to the rest of the world; I think we should worry about the corrupting effects of this kind of power.

Forging the One Ring was evil

Of course, in the story we are told that the Enemy made the Ring, and that he was going to use it for evil ends; and so of course it was evil. But I don’t think that’s the whole reason that forging the Ring was bad.

I think there’s something which common-sense morality might term evil about a project which accumulates enough power to take over the world. No matter its intentions, it is deeply and perhaps abruptly disempowering to the rest of the world. All the other actors — countries, organizations, and individuals — have the rug pulled out from under them. Now, depending on what is done with the power, many of those actors may end up happy about it. But there would still, I believe, be something illegitimate/bad about this process. So there are reasons to refrain from it[1].

In contrast, I think there is something deeply legitimate about sharing your values in a cooperative way and hoping to get others on board with that. And by the standards of our society, it is also legitimate to just accumulate money by selling goods or services to others, in order that your values get a larger slice of the pie.

What if the AGI project is not run by a single company or even a single country, but by a large international coalition of nations? I think that this is better, but may still be tarred with some illegitimacy, if it doesn’t have proper buy-in (and ideally oversight) from the citizenry. And buy-in from the citizenry seems hard to get if this is occurring early in a fast AI takeoff. Perhaps it is more plausible in a slow takeoff, or far enough through that the process itself could be helped by AI.

Of course, people may have tough decisions to make, and elements of illegitimacy may not be reason enough to refrain from a path. But they’re at least worth attending to.

The difficulty of using the One Ring for good

In The Lord of the Rings, there is a recurring idea that attempts to use the One Ring for good would become twisted, and ultimately serve evil. Here the narrative is that the Ring itself would exert influence, and being an object of evil, that would further evil.

I wouldn’t take this narrative too literally. I think powerful AI could be used to do a tremendous amount of good, and there is nothing inherent in the technology which will make its applications evil.

Again, though, I am wary of having the power too centralized. If one centralized organization controls the One Ring, then everyone else lives at their sufferance. This may be bad, even if that organization acts in benevolent ways — just as it is bad for someone to be a slave, even with a benevolent master[2]. Similarly, if the state is too strong relative to its citizens then democracy slides into autocracy — the state may act in benevolent ways for the good of the people, and still be depriving them of something important.[3]

Moreover, even if in principle the One Ring could be used in broadly beneficial ways, in practice there are barriers which may make it harder to do so than in the case of less centralized projects:

The corrupting nature of power

The One Ring was seen as so perilous that wise and powerful people turned down the opportunity to take it, for fear of what it might do to them. More generally, it’s widely acknowledged that power can be a corrupting force. But why/how? My current picture[4] is that the central mechanism at play is insulation from prosocial pressures:

I sometimes think about this power-as-corrupting-force in the context of AI alignment. It seems hard to specify how to get an agentic system to behave in a way that is well-aligned with the intent of the user. “Hmm,” goes one train of thought, “I wonder how we align humans to other people?”. And I think that the answer is that in the sense of the question as it’s often posed for AI systems, we don’t do a great job of aligning humans. 

We wouldn’t be happy turning over the keys to the universe to any AI system we know how to build; but we’d also generically be unhappy doing that with a human, and suspect that a nontrivial fraction would do terrible things given absolute power.[5]

And yet human society works: many people have lots of prosocial instincts, and there is not so much effort spent in the pursuit of seriously antisocial goals. So it seems that society — in the sense of lots of people with broadly similar levels of power who mutually influence and depend on each other — is acting as a powerful mediating force, helping steer people to desires and actions which are more aligned with the common good. 

All of this gives us reasons to be scared about creating too much concentration of power. This could weaken or remove the pro-social pressures on the motivations of the actor(s) who hold power.[6] I believe the same basic argument works for organizations or institutions as for individuals. Moreover — and like the One Ring — an organization which has (or is expected to gain) lots of power may attract power-seeking individuals who try to control it.

The importance of institutional design

If someone does create the One Ring, or something like it, the institution which governs that will be of utmost importance. The corrupting nature of power means that this is always going to be a worrying situation. But some ways for institutions to be set up seem more concerning than others. This could be the highest-stakes constitutional design question in history.

This is its own large topic and I will not try to get to the bottom of it here, but just note a few principles that seem to be key:

What if there were Three Rings?

How much of the issue here is about the very singular nature of the One dominant project, vs centralization more generally into a small number of projects?

I think that multiple projects could meaningfully diffuse quite a lot of the concern. In particular there are two dynamics which could help:

There would still be worry about the possibility of collusion between the small number of projects moving things back to something resembling a One Ring situation. And broadly speaking, Three Rings might still represent a lot of centralization of power. 

There may be other ways to decentralize power than increasing the number of projects. Perhaps a single centralized project could train the most powerful models in the world — but instead of deploying them directly, it licenses fine-tuning access to many companies, who then sell access to the models. But the more there are meaningful single points of control, the more concerned I feel about One Ring dynamics. Creating a single point of control is the core difficulty of a single centralized project.[7] In this example, I would hope for great care and oversight of the decision-making process that keeps the project licensing fine-tuning access to many companies on equal footing.

Why focus on the downsides?

This post isn’t trying to provide a fair assessment of whether it’s good to forge the One Ring. There are a number of reasons one might decide to do so. But there are many incentives which push towards people accumulating power, and hence push against them looking at the ways in which that might be problematic. This applies even if the people are very well intentioned (since they’re unlikely to imagine themselves abusing power). I worry some about the possibility of people doing atrocious things, and justifying those to themselves as “safer”.

I would like to counteract that. I’ll have much more trust in any decision to pursue such a project if the people who are making that decision are deeply in touch with, and acknowledge, the ways in which it is a kind of evil. The principle here is kind of like “in advance, try to avoid having a missing mood”. This would increase my trust both in the decision itself (it’s evidence that it’s the correct call if it’s chosen after some serious search for alternatives which avoid its problems), and in the expected implementation (where people who are conscious of the issues are more likely to steer around them).

This is also the reason I've chosen to use the One Ring metaphor. I think it's a powerful image which captures a lot of the important dynamics. And my hope is that this could be more emotionally resonant than abstract arguments, and so could help people[8] to stay in touch with these considerations even if their incentives and/or social environment encourages thinking that a centralized project would be a good idea.

Acknowledgements: Thanks to Max Dalton for originally suggesting the One Ring metaphor. Thanks to Max Dalton, Adam Bales, Jan Kulveit, Joe Carlsmith, Raymond Douglas, Rose Hadshar, TJ, and especially Tom Davidson for helpful discussion and/or comments.

 

  1. ^

    I’m not pinning down the exact nature of these reasons, but I’ll note that they might have some deontological flavour (“don’t trample on others’ rights”), some contractualist flavour (“it’s uncooperative to usurp power”), or some virtue-ethics-y flavour (“don’t be evil”).

  2. ^

    I am grateful to a reviewer who pointed out the similarities between my concerns about illegitimacy and Pettit’s notion of freedom as nondomination; the slave analogy is imported from there.

  3. ^

    I’m interested in ACS’s research on hierarchical agency for the possibility of getting more precise ways to talk about these things, and wonder if other people should also be thinking about topics in this direction.

  4. ^

    Formed from a mix of thinking this through, and interrogating language models about prominent theories.

  5. ^

    Perhaps there are some humans who would persistently function as benevolent dictators, even given absolute power over a long time period. It is hard for us to tell. Similarly, perhaps we could build an AI system which would in fact stay aligned as it became more powerful; but we are not close to being confident in our ability to do so.

  6. ^

    We might hope that this would be less necessary if we were concentrating power in the hands of an AI system that we had reason to believe was robustly aligned, relative to concentrating power in human hands. But it may be hard to be confident in such robust alignment.

  7. ^

    Although this may also have advantages, in making it easier to control some associated risks.

  8. ^

    Ultimately, it may be only a few people who, like the sons of Denethor, are in a position to decide whether to pursue the One Ring. I have little fear that they will fail to perceive the benefits. It seems better if, like Faramir, they are also conscious of the costs.