What does it mean for AI to be democratic?

Some pushback on a specific fuzzy idea with some ominous implications

I love democracy. I have what’s basically religious zealotry for the form of liberal democracy most of the wealthy world lives under. Citizens having control and oversight of their leaders is a rare treasure in human history, and one that AI threatens to erode. I have never read a popular modern critic of democracy in the broad sense who I didn’t see as, at best, a complete joke.

I’ve noticed that in lots of conversations about current AI models and the general AI buildout, it’s becoming more and more popular to say that AI isn’t happening "in a democratic way. There are times when I strongly agree. I’m very worried about the ways AI could enable the concentration of power, or impose huge (including existential) risks that citizens never signed up for and would hit the brakes on if they could.

But there’s a specific way of arguing that AI isn’t “democratic” that’s gaining traction, and that I find ominous and antithetical to what makes a good democratic society. It’s usually dressed in the language of progressive politics, but underneath, I think it carries authoritarian, homogenizing and anti-democratic implications.

Pluralist vs homogenizing democracy

There are two visions of democracy hiding behind our talk about it, one I think is good and the other I think is bad, and you can probably tell which is which by the labels I give them.

The goal of pluralist democracy is to give people with wildly different beliefs about the good life and what’s really valuable control and veto points over how they’re governed, without letting them steamroll each other into all living under one contentious ethical or religious vision of the world.

Most liberal democracies like the US manage to balance pluralism with majority rule. Despite being a majority Christian country, and that majority having voting power over who represents them, the US government is not able to impose Christianity on everyone else. People have a lot of democratic control over contentious issues of government, but they often can’t use this to steamroll other people into very specific visions of the good life. Christians are restrained by governing norms from forcing non-Christians from attending church.

The goal of homogenizing democracy is to discover the true, good ethical values and beliefs about the world, with the faith that the majority either publicly or secretly believes them, and then to structure all society around those true, good ethical beliefs, and push ideological minorities into going along.

Here’s a more ominous way to frame it: “There is an authentic community of rooted, powerless, good people, who know what’s good and right in life and agree on it. They’re up against unrooted powerful people who don’t know what’s good in life and want to override it for their own gain. The good people are in the majority. Democracy means discovering these good values all good people hold, and then representing and acting on it, steamrolling the bad powerful people into going along with these good values.”

I read this as ominous because it’s an especially useful cudgel for forcing people into your specific vision of ethics and the good life. If you invalidate another value system by saying it’s not what all the good, powerless people believe, this gives you more permission to steamroll the people who hold that value system in the name of resisting power. Many Christian nationalists will frame themselves as representing the good, simple, rooted majority beset by powerful cabals who don’t understand their values, and use this to argue that underneath these differences is only a raw struggle for power that the good people need to win. Thus to them, seizing the government to impose Christianity on religious minorities is actually a form of democracy. It’s simply letting the values of the good majority be represented in how the country is governed. Stuff like this is why I worry about accepting a narrative that lets homogenizing democracy sneak in unexamined. Democracy at its best weighs the pluralistic desires of radically different people. At its worst it excludes those desires from the debate entirely, on the grounds that anyone who disagrees with the majority is illegitimate and not worth making concessions for.

And it can creep in under the guise of noble-sounding rules, like “a town should have control over how its resources are used.” For example, consider the claim that a small town should have complete democratic control over what gets built there. This sounds nice in the abstract, but is actually a way for homogenizing democracy to slip in. Imagine what happens if 80% of the town is Catholic, and 10% is Muslim. The Muslim population gets together and says they would like to buy some of the town’s land to build a mosque. Will this mosque get built if it’s voted on by the town? Well, 80% of the town believes that the mosque is part of a false religion that sends people to hell. It seems unlikely that the mosque will get built. By giving citizens democratic control over everything that gets built in their town, we’ve allowed them to steamroll pluralist values and homogenize the area into the contentious beliefs of the majority of people who happen to live there. This is bad!

Some people imply that in a “truly” democratic society, all value disagreement would ultimately be dissolved through deliberation. Communities would come together and deliberate until they discovered the most virtuous or liberatory way to live. I think deliberation like this is a great model for specific decisions the government makes, but it shouldn’t be allowed to seep as far into everyday life, because taken to the extreme it implies everyone must eventually be made to agree on every contentious value.

So while I think that at its best, liberal democracy is the transcendent crowning achievement of human history, talk about democracy can also be used to justify authoritarian homogenizing impulses and needs to be carefully examined. When people invoke democracy, they’re usually smuggling in general rules about how society should be governed, and those rules’ political implications need to be drawn out. The specific model of democracy where the goal is always to discover the good values that need to be imposed on everyone, that all good people either publicly or secretly believe (or would believe if they could be led to the light), and that anyone who disagrees with is not worth considering in deliberation, is a recipe for destroying the delicate balance that allows people with wildly different beliefs to live together and change their values freely.

People in democratic societies have strong reasons to try to frame their opponents’ opinions as outside the sphere of legitimate democratic consideration. An effective way to do this is to frame their opponents as representing “the opinions of the powerful”meaning illegitimate, out-of-touch views trying to undermine the good people. This is often a way of trying to banish different views from the process of democratic deliberation. I’m worried that this framing comes up a lot in the AI debate, used by people who think they’re behaving as the defenders of democracy, but are often actually stifling authentic difficult democratic deliberation between radically different values.

Which visions of democratic AI are pluralist vs homogenizing?

I think AI poses a unique technological threat to pluralist democracy. Society currently exists in a delicate, kind of miraculous balance where most people are able to add a lot of economic value, and have lots of ways to defect from important systems if their political power is threatened. We are deeply interdependent. AI could, under some models, erode our economic interdependence and leave powerful people with less incentive to respect the will of everyday voters. It’s very possible that democracy is a lucky passing phase in human history, built around a specific economic regime that’s existed since the Industrial Revolution that might go away if we enter a new and different mode of economic production. See here for more.

AI also poses lots of other risks, either indirect risks that people will use it to create dangerous powerful new weapons, or more speculative direct risks of AI itself overriding what we want. Just like I think everyday people should have democratic control over whether their neighbors have access to nuclear weapons, I think people have a democratic right to expect the general flow of technological progress to not descend into a specific basin where society enters an incredibly unstable equilibrium of extremely powerful easily accessible weapons.

The dynamics of automation with AI also threaten to concentrate wealth and power in the labs themselves. OpenAI and Anthropic are now some of the fastest growing companies in history. Many of their members have become billionaires, and some local political fights are becoming proxy battles between them over rival visions of AI safety and governance policy. If you expect AI capabilities to continue to progress, this poses a danger of a unique form of corporate power consolidation we haven’t seen in America in a long time, if ever. This also threatens political pluralism, and people have a right to expect that their representatives respond to this new threat. How? I don’t really know, but I’m willing to entertain lots of different answers from libertarians or socialists or anyone in between. I worry our learned political impulses might be as useful in the new era of AI as Aristotle would have been in the Industrial Revolution.

Lots of concerns like this I think are legitimate threats to pluralist democracy, and people have the right to democratically demand forms of intervention that I wouldn’t normally consider for more normal technology. More broadly, a new era of intense automation will demand democratic deliberation between people with radically different values on the shape of the new world that they want to live in together.

I think AI will over the next ten years demand an “all hands on deck” response from political philosophers, economists, and everyday people, and authentic pluralist democratic deliberation needs to happen.

While I’ve been happy with some aspects of how the left’s thinking has evolved on this, there are specific strands of intensely anti-democratic thinking that seem to be attempting to take over the debate and define what democratic AI even means. Here I worry a lot that people are using the language of democracy to push for homogenization and authoritarianism.

A clear example comes up a lot in the data center buildout. A very common line among people claiming to represent democracy in the debate is to say “No one is asking for these! They only benefit billionaires!” I think if I were able to separate these people from the debate, take them out for some nice tea somewhere, have a long conversation with them about something else, and then politely ask “Just being real, do you think there are a decent number of everyday people who think about AI in a different way than you whose demand for more is incentivizing the data center buildout?” I think they would say yes, but that these people are mistaken. We could go on and have a nice conversation from there. I worry that the initial claim that “no one is asking for this!” is basically a way of pushing the vast numbers of people who are in fact asking for this out of the difficult process of deliberation that should happen with any new technology.

There are a lot of people who want to use AI. Chatbots are consistently the most downloaded apps. Half of all web developers polled say they use AI daily. About a quarter of Americans say they use chatbots daily.

Source

More people say AI helps than hurts most things they use it for.

Source

However, Americans also believe AI is more likely to make society worse.

Source

So what do we do with this? How should we think about this from a democratic perspective?

If a quarter of people like a product, two quarters hate it, and one quarter is indifferent, should that product be allowed to exist?

Homogenizing democracy would say no. The plurality has spoken. People with different opinions from the plurality should be steamrolled.

This is obviously wrong, yet I see people imply basically this in lots of debates about data centers. People will often say that people using AI who like it are both a small minority, and so out of touch with what’s good and right that their AI use can’t be allowed to use any resources at all. It can’t be allowed a drop of water or an acre of land or to emit any CO2. When I argue that people who like chatbots need to have their own demands represented in the market and in democratic deliberation, I’m often met with a knowing sneer, like “Oh, this guy hasn’t gotten the memo. These people using AI aren’t really ‘the people.’ They’re alien outsiders and don’t get access to the sphere of deliberation. Real democracy means making sure these people are completely disempowered and are never able to use any community’s resources.” To anyone who actually thinks about political philosophy at all, this is a bizarre and sad mangling of the democratic impulse. It’s an intense assertion of anti-democratic power for one specific faction in the debate. For any other unpopular product, we’d recognize how ridiculous this is.

One way to steamroll behavior you don’t like is to argue that it imposes lots of externalities on other people, so it’s actually an attack on their own freedom. Despite data centers not really imposing more externalities on locals than any other normal industry (from what I can tell), a huge media bubble has developed where wild stories are told about data centers forcing whole towns to evacuate, of infrasound ruining people’s lives, of water polluted and taps running dry, most of which evaporate the moment you poke at them. People seem to have a strong desire for these stories to be true. There was a recent New York Times opinion column on how the data center backlash may be a “winning issue for Democrats,” framed almost entirely around the sense of community people feel in anti-data center protests rather than any solid issues with them. I worry a lot of this is a sublimated authoritarian lurch to control the behavior of people doing something you don’t like. Stories about data centers being bad feels good because it hands you a cudgel you can use on people you think are out-of-touch losers. There’s a lot happening here, but I don’t see this as a good sign for authentic democratic deliberation about the future of AI. It seems much more like an impulse to homogenize everyone into a specific way of thinking about AI that they wouldn’t assent to on their own.

This also shows up in very confident claims that AI cannot, by definition, ever produce anything of value at all. At some point this looks a lot more like ignoring the vast numbers of intelligent people saying they’re getting something of value by using it. Critics don’t have to agree with them, but the confident metaphysical claims about the inherent value of AI output start to look more like a lurch to stifle debate by casting people who disagree out beyond the realm of legitimate conversation.

Separate from this debate, another instance of the homogenizing impulse is the idea that individual models themselves should be made to more democratically represent the opinions of the majority. There’s a lot of talk about “for whom” chatbots are designed, implying that they have lurking biases that would be fixed by making them more democratic. I don’t doubt that chatbots (like anything) carry lots of blindspots and misunderstandings of the experience of people very different from the people who designed them and who wrote the data they trained on, but I don’t at all think democracy specifically will make this problem better, and this implication that private products should all represent the democratic will of the good people I think is another way of smuggling in the homogenizing impulse.

Current AI models would become way more conservative if their values were voted on

In my 20’s I went through a huge Civil War phase. Anyone who does the deep dive comes away kind of shocked at the state of public discourse about the Confederacy. While public understanding has improved, I think a lot of people still really don’t know just how bad slavery was or how central of a motivation it was for the South in the war. This is reflected in public polling. More Americans have a positive than negative opinion of Jefferson Davis, for example:

Maybe this is an artifact of most people just not knowing who he was, or confusing him with Thomas Jefferson. Regardless, this is a pretty clear case where I’m very happy that the main AI models deviate from the winning view of Jefferson Davis among the public.

Claude:

ChatGPT:

Gemini:

Maybe someone could program an AI model that more accurately represents the views of the median American. This AI would be unsure about evolution, and in general much more conservative across the board than the average person demanding that AI’s democratically represent the beliefs everyday people hold.

Someone who holds a simplistic, homogenizing view of democracy often needs to invent a story where everyday people all secretly agree with them. If you imagine that the good rooted majority of people all know your values are correct, this gives you grounds to steamroll the few out of touch people who disagree with you. Because they’re out of touch, their divergent views have no claim to political respect or protection.

But the views of everyday people are complicated and often way out of line with progressive visions of the world. There are many values I share with my fellow coastal elite city dwellers that I’m happy to see reflected in AI models. These values show up less outside of cities, and even less in other countries. They are often not the views of either the American or global majority.

I think a lot about this exchange from this 80,000 Hours interview with Mustafa Suleyman, the co-founder of DeepMind:

Rob Wiblin: Yeah. While you were involved with DeepMind and Google, you tried to get a broader range of people involved in decision making on AI, at least inasmuch as it affected broader society. But in the book you describe how those efforts more or less came to naught. How high a priority is solving that problem relative to the other challenges that you talk about in the book?

Mustafa Suleyman: It’s a good question. I honestly spent a huge amount of my time over the 10 years that I was at DeepMind trying to put more external oversight as a core function of governance in the way that we build these technologies. And it was a pretty painful exercise. Naturally, power doesn’t want that. And although I think Google is sort of well-intentioned, it still functions as a kind of traditional bureaucracy.

Unfortunately, when we set up the Google ethics board, it was really in a climate when cancel culture was at its absolute peak. And our view was that we would basically have these nine independent members that, although they didn’t have legal powers to block a technology or to investigate beyond their scope, and they were dependent on what we, as Google DeepMind, showed them, it still was a significant step to providing external oversight on sensitive technologies that we were developing.

But I think some people on Twitter and elsewhere felt that because we had appointed a conservative, the president of the Heritage Foundation, and she had made some transphobic and homophobic remarks in the past, quite serious ones, that meant that she should be cancelled, and she should be withdrawn from the board. And so within a few days of announcing it, people started campaigning on university campuses to force other people to step down from the board, because their presence on the board was complicit and implied that they condoned her views and stuff like this.

And I just think that was a complete travesty, and really upsetting because we’d spent two years trying to get this board going, and it was a first step towards real outside scrutiny over very sensitive technologies that were being developed. And unfortunately, it all ended within a week, as three members of the nine stood down, and then eventually she stood down, and then we lost half the board in a week and it was just completely untenable. And then the company turned around and were like, “Why are we messing around with this? This is a waste of time.”

Rob Wiblin: “What a pain in the butt.”

Mustafa Suleyman: “Why would we bother? What a pain in the ass.”

Rob Wiblin: It was a very striking story to me, reading that in the book. People complain, I think correctly, that decisions of enormous global importance, historical importance, are potentially going to be made inside these AI labs — and the kinds of people who work at these labs are a very small fraction of the people in the world in terms of their political views, the values that they have, the things that they’ve studied, the kind of information that they happen to know. So it would be good if we could get a wider cross-section of the human population involved in scrutinising or having some input on these.

But to share power with the general public in the Global South, or even just outside big cities in the US or UK, will inevitably involve giving influence to people with views that I imagine are very offensive to Google staff — probably more offensive than Kay Coles James, who had to resign from the board for having more conservative, traditional views on gender. So based on that experience, it just seems like it’s very unlikely to happen, when we might just be flat out trying to get acceptance for having more input from a broader cross-section of educated people in the UK into Google. Like, that’s going to be the most that people will tolerate.

Mustafa Suleyman: Yeah, totally. This is part of the problem, right? I mean, 40% of people in the US believe that trans rights are moving too quickly; 30% believe that abortion should be made illegal; 30-odd% are against gay marriage.

Rob Wiblin: And then think globally, right?

Mustafa Suleyman: Right, exactly, then think globally. So I think we have to just learn to sit down with people who we fundamentally disagree with. That goes for China and the Taliban and all these people who hold these views — because if we can’t do that, then we’ve really got no chance of actually hearing one another out and changing one another’s views and making progress. I think in the last two or three years, I feel like we’ve really taken a few backward steps in that direction, and it’s super problematic because it just demonizes the other and then we just end up hating on one another.

And it’s very frustrating for us, because we put in a huge amount of effort to make that happen. Before that, when we were acquired, we made it a condition of the acquisition that we have an ethics and safety board. That in itself was a first step towards this kind of broader public effort. Then after the ethics and safety board, we actually tried to spin DeepMind out as a global interest company: one that was legally governed by the requirement to consider all of the stakeholders when making decisions. So it was a company limited by guarantee. And then the charter definition had an ethics and safety mission for AGI development; we actually had the ability to spend vast amounts of our income on scientific and social mission.

So it was a really creative and experimental structure. But when Alphabet saw what happened with that board, they basically just got cold feet. That was the bottom line. They saw what happened there and they were just like, “This is totally nuts. The same thing’s going to happen for your global interest company. Why do that?” And then eventually we pulled DeepMind into Google, and in a way DeepMind was never independent — and isn’t independent now; obviously now it’s completely part of Google.

Rob Wiblin: Yeah. The core of trying to address representativeness is that you will be ceding power to people who don’t share your values, and if people are not willing to make that compromise, then it’s not going to happen.

I’ll just flat out say that I have zero interest in sitting with what the Taliban has to say about what values current AI models should reflect. The Taliban can just not use the big chatbots, or build their own more in line with their values. I don’t feel any sense of democratic injustice at all that their beliefs are not influencing the behavior of a private product. The only worlds where the opinions of the Taliban become relevant at all for AI are the really wild speculative situations where AI comes to determine basically everything that happens on Earth. I do see some risk of that happening, but few of the people worried that AI doesn’t represent everyone on Earth equally seem to share this particular worry. If AI rises to run all society, I will circle back with the Taliban.

I’m sure the AI models get plenty of important facts wrong about the lives and experiences of people in very different situations from their designers. But the solution here doesn’t at all look like making AI more representative of the average person in the world, because the average person also carries horrible biases against lots of other people. One way to fix this could maybe be selecting specific expert representatives of different under-represented communities or ways of living and having those experts impose specific ways of responding on the AI. Maybe different worldviews could make their own LLMs, each representing those specific views. But I worry that making these private products “more democratic and representative of everyday people” when they already hold more liberal views than most people do would mostly just lead to more wrong or hateful models.

Lurking under this idea that AI models themselves should be “more democratic” is sometimes the idea that all products in society should reflect the correct opinions of the good majority, and never reflect the opinions of the bad minority. It’s again a form of homogenization. I think a better solution here would be for radically different ideological communities to each have powerful AI models of their own, none claiming to represent the democratic will of all good people, something impossible in a complex pluralist world. But I also am just not especially bothered that AI models don’t reflect a lot of the very conservative values of everyday people that I think are just incorrect. I’m not eager for CreationistGPT.

Conclusion

I think authentic democratic deliberation about AI is going to be deathly important over the next ten years, as AI threatens to concentrate economic power and unleash new risks on the world that we’re not ready for. I don’t want this debate to lead to anyone using the banner of democracy to override actual democratic deliberation. We need to acknowledge that there’s already both a lot of demand for and a lot of hatred of AI, and welcome both as full participants in the debate about the future. We should be wary of anyone declaring that barring huge numbers of people from the debate is a form of “democratic resistance to AI.” This threatens to mostly burn the commons of the debate and alienate people from more subtle ways of resisting and democratically governing the unique risks AI poses.