American democracy currently operates far below its theoretical ideal. An ideal democracy precisely captures and represents the nuanced collective desires of its constituents, synthesizing diverse individual preferences into coherent, actionable policy.

Today's system offers no direct path for citizens to express individual priorities. Instead, voters select candidates whose platforms only approximately match their views, guess at which governmental level—local, state, or federal—addresses their concerns, and ultimately rely on representatives who often imperfectly or inaccurately reflect voter intentions. As a result, issues affecting geographically dispersed groups—such as civil rights related to race, gender, or sexuality—are frequently overshadowed by localized interests. This distortion produces presidential candidates more closely aligned with each other's socioeconomic profiles than with the median voter.

Traditionally, aggregating individual preferences required simplifying complex desires into binary candidate selections, due to cognitive and communicative limitations. Large Language Models (LLMs), however, introduce a radical alternative by processing detailed, nuanced expressions of individual views at unprecedented scales.

Instead of forcing preferences into narrow candidate choices, citizens could freely articulate their concerns and solutions in natural language. An LLM can rapidly integrate these numerous, detailed responses into a clear and unified "Collective Views" document. Previously, synthesizing a hundred individual perspectives might have required five person-hours; specialized LLMs can now accomplish this task in minutes. Parallel implementations could aggregate millions of voices within an hour, transforming a previously unimaginable task into routine practice.

Such rapidly generated collective statements create a powerful mechanism for accountability, making government responsiveness directly measurable against clearly articulated public preferences. Transparency naturally constrains representatives' ability to diverge unnoticed from voter priorities.

Moreover, LLM-generated collective views could directly shape legislative drafting, significantly reducing lobbyist influence and governmental inefficiency. Continuous, dynamic engagement through AI enables real-time policy-making aligned closely with public sentiment, redefining democratic responsiveness.

This is the first in a possible series of posts exploring practical AI solutions to realize democratic ideals at scale. Subsequent posts could cover:

  1. Aggregation: Prototyping AI systems that clearly synthesize individual views into a cohesive Collective Will statement.
  2. Accountability: Holding up a Collective Will next to actual government activity (e.g., budget allocation) to highlight discrepancies.
  3. Action: Outlining concrete strategies to translate a Collective Will into effective legislative outcomes.
New Comment
16 comments, sorted by Click to highlight new comments since:

LLMs may enable direct democracy at scale

I suppose they could, and from now on I'll consider this to be one of the many significant dangers of governance by AI. 

Part of the problem with direct democracy is that it provides no unambiguous mechanism for leaders to exercise judgment about the relative importance of different individuals' preferences and concerns, or the quality of their information and reasoning. A great many of America's, and the world's, most important advances and accomplishments in governance have happened in spite of, not because of, public sentiment. As I see it, many of the failings of American democracy in recent decades are rooted in the fact that we now demand a level of transparency that precludes the kind of quiet, private dealmaking and negotiation that used to enable elected officials to handle matters where their constituents' claimed preferences are inconsistent, incoherent, misguided, or otherwise not good for the country as a whole. 

Is the less transparent human alternative often dangerous and misguided too? Absolutely. Could a truly virtuous (possibly sovereign) AI do better than any set of humans at setting up governance structures that facilitate human flourishing like never before? Also yes. But there are many ways to get the aggregation mechanism even just slightly wrong that could then turn into catastrophic or unrecoverable mistakes. Right now, if anyone proclaims any intention to try to implement such a thing, I'm quite confident they have no idea how to do so in a way that avoids such dangers.

your desire for a government that's able to make deals in peace, away from the clamor of overactive public sentiment... I respect it as a practical stance relative to the status quo. But when considering possible futures, I'd wager it's far from what I think we'd both consider ideal.

the ideal government for me would represent the collective will of the people. insofar as that's the goal, a system which does a more nuanced job at synthesizing the collective will would be preferable.

direct democracy at scale enabled by LLMs, as i envision it and will attempt to clarify, might appeal to you and address the core concern you raise. specifically, a system that lets people express which issues they care about in a freeform way would, i think, significantly reduce the amount of misguided opinions. today, individual participation mostly looks like voting for candidates who represent many opinions on a vast range of issues—most of which we almost definitely don't care about—instead of allowing us to simply express our feelings about the issues which actually affect us.

in other words, when politicians hold rallies and try to persuade us, they're usually asking us to care about things as distant from our real lives as Russia is from Alaska. and those opinions become the most misguided ones: the bigotry we harbor toward people we don't know, the technophobic or blindly optimism attitudes (either extreme) we have toward technology we don't understand, and the moralism that makes us confident projecting our voice across domains where we're ignorant.

all this to propose: a system representing the collective will as a synthesis of individual wills expressed freeform. i'd encourage anyone interested in this issue to work on it ASAP, so we can coordinate around extinction risks before they're here.

I agree that AI in general has the potential to implement something-like-CEV, and this would be better than what we have now by far. Reading your original post I didn't get much sense of attention to the 'E,' and without that I think this would be horrible. Of course, either one implemented strongly enough goes off the rails unless it's done just right, aka the whole question is downstream of pretty strong alignment success, and so for the time being we should be cautious about floating this kind of idea and clear about what would be needed to make it a good idea.

 

There's a less than flattering quote from a book from 1916 that "Democracy is the theory that the common people know what they want and deserve to get it good and hard." That pretty well summarizes my main fear for this kind of proposal and the ways most possible implementation attempts at it would go wrong.

"Democracy is the theory that the common people know what they want and deserve to get it good and hard."

Yes, I think this is too idealistic. Ideal democracy (for me) is something more like "the theory that the common people know what they feel frustrated with (and we want to honor that above everything!) but mostly don't know the collective best means of resolving that frustration.

a system that lets people express which issues they care about in a freeform way

We already have that: the Internet, and the major platforms built on it. Anyone can talk about anything.

allowing us to simply express our feelings about the issues which actually affect us.

If the platform is created, how do you get people to use it the way you would like them to? People have views on far more than the things someone else thinks should concern them.

"If the platform is created, how do you get people to use it the way you would like them to? People have views on far more than the things someone else thinks should concern them."

>

If people are weighted equally, ie if the influence of each person's written ballot is equal and capped, then each person is incentivized to emphasize the things which actually affect them. 

Anyone could express views on things which don't affect them, it'd just be unwise. When you're voting between candidates (as in status quo), those candidates attempt to educate and engage you about all the issues they stand for, even if they're irrelevant to you. A system where your ballot is a written expression of what you care about suffers much less from this issue.

the article proposes a governance that synthesizes individuals' freeform preferences into collective legislative action.

internet platforms allow freeform expression, of course, but don't do that synthesis.

I highly recommend checking out the work being done in the collective deliberation / digital democracy space, especially the vTaiwan project. People have been thinking about scaling up direct democratic participation for a long time, and those same people are starting to consider exactly how AI might play a role. 

In particular, check out this collaboration between the creators of Polis (a virtual platform for scaling up citizen engagement) and Anthropic, or my distillation of a DeepMind project to scale citizen assemblies. There's a lot happening in this space right now! 

awesome thx

I do think it's helpful that managers now have a reliable way to summarize large amounts of comments, instead of making some poor intern with Excel try and figure out "sentiment analysis" to "read" thousands of comments without having to pay for a proper data scientist, and I wonder if that's already had some effects in the world.

My problem with democracy is that most people are stupid. More precisely, they are relatively good at saying whether they are happy or unhappy, starving or fed, etc. They can give a relatively reliable "thumb up" or "thumb down" feedback to the government. But if you ask them about the specific steps that should be taken to make them better fed, etc., many of the popular suggestions will be quite insane.

For example, people can have a legitimate complaint about healthcare being inaccessible for them, and yet the suggestion many would propose will be something like "government should spend more money on homeopathy and spiritual healing, and should definitely stop vaccination and other evil unnatural things".

This seems inevitable to some degree. We can't realistically expect that everyone will be an expert on everything. And we want to allow feedback such as "the experts are hurting us, and we want that to stop", but we don't really want the non-experts to micro-manage the experts.

The traditional solution is to give people only the "thumb up/down" kind of feedback. For example, in capitalism, you can choose to buy or not to buy a certain kind of product or service, but you can't e.g. go to a restaurant and demand that they start cooking meals according to your recipe, no matter how good that recipe sounds to you. But you can start your own restaurant that will cook according to your recipes, and then the customers will vote on the result.

Even this approach has many problems, because some things seem good in short term, but are harmful in long term, and these often get a "thumb up" from the non-experts. The more tasty food can be less healthy. The economic problems or successes of today may be a consequence of things that the previous government did.

The problem with democracy is externalities. People who vote for stupid things don't only hurt themselves; they equally hurt their neighbors. It would be better to set up things in a way that when people insist on doing the wrong thing, they hurt themselves more than they hurt the others. But this itself is a subject to voting, and many people oppose it.

...shortly, I am worried that this would allow democratic micro-management, i.e. the rule of stupid.

That said, I am not sure what is the optimal level of democracy, but I am suspicious of the assumption that more is necessarily better.

For example, people can have a legitimate complaint about healthcare being inaccessible for them, and yet the suggestion many would propose will be something like "government should spend more money on homeopathy and spiritual healing, and should definitely stop vaccination and other evil unnatural things".

Yes. This brings to mind a general piece of wisdom for startups collecting product feedback: that feedback expressing painpoints/emotion is valuable, whereas feedback expressing implementation/solutions is not.

The ideal direct-democratic system, I think, would do this: dividing comments like "My cost of living is too high" (valuable) from "Taxes need to go down because my cost of living is too high" (possibly, but an incomplete extrapolation). 

This parsing seems possible in principle. I could imagine a system where feedback per person is capped, which would incentivize people to express the core of their issues rather than extraneous solution details (unless they happen to be solution-level experts).

feedback expressing painpoints/emotion is valuable, whereas feedback expressing implementation/solutions is not.

Yep. Or, let's say that the kind of feedback that provides solutions is worthless 99% of time. Because it is possible in principle to provide a good advice, it's just that most people do not have the necessary qualification and experience but may be overconfident about their qualification.

I find it ironical that popular wisdom seems to go the other way round, and "constructive criticism" is praised as the right thing to do. Which just doesn't make sense; for example I can say that a meal tastes bad, even if I don't know how to cook; or I can complain about pain without being able to cure it.

Imllementing direct democracy is vastly simpler than implementing a LLM. It's called sortition. Randomly select people to form the legislature instead using elections. Random people are a better direct, descriptive representation of the public. Voila, it's not actually direct democracy, just like relying on LLMs is not actually direct democracy, yet it's vastly more simple to implement. 

Like with any agent, whether AI or random or elected, all agents will suffer from principle -agent problems and problems of corruption.

I'm currently working on similar stuff. I wanna build open source embedding search right now. Feel free to schedule a call if you find this or this interesting.

[+][comment deleted]10
Curated and popular this week