AI, platform, crowd
Aviv Ovadya has a vision for supercharging deliberative democracy with the help of AI.
Welcome to the first proper issue of Futurepolis! As a reminder, these posts will have a two-part structure: a chunk of text and a bunch of links. The chunk of text will be things like an interview with someone in the future-of-governance world, a commentary on some recent news, or a book review. This is where I get to hash out some thoughts or highlight people who are doing interesting work. Each time, I’ll include a TL;DR, since I know we’re all busy people. The bunch of links will be, well, a bunch of links to interesting things that have popped up in the news recently. I hope you enjoy it and find it useful.
Aviv Ovadya is a researcher who, per his bio, “works at the intersection of AI, platforms, democracy, and deliberation.” More concretely, that means he works on ways to give citizens a say in making the rules for the tech giants that affect all our lives. But increasingly he’s thinking about how to apply the same techniques beyond tech platforms, to problems of democratic and corporate governance in general. He’s about to launch the AI & Democracy Foundation, so I caught up with him to get a handle on his thinking and what his new organization is going to do.
The TL;DR: Aviv thinks we can use representative deliberation—a.k.a. citizens’ assemblies, deliberative polling, and the like—much more widely than we do now. In part that will happen by using AI to make these processes both more representative and more efficient. And he believes this can lead to nothing less than a “new constitutional order” in which certain matters are routinely decided by groups of citizens rather than by politicians or company CEOs.
This interview has been edited for clarity and conciseness.
In a paper last year you identified four phenomena that are now making it possible to think about democracy in new ways: Representative deliberations, AI augmentation, democracy-as-a-service, and platform democracy. Can you briefly say what each one is?
So first there’s representative deliberations, which… wow, it’s really hard to summarize quickly.
Let’s assume my readers know what citizens’ assemblies are.
OK, so representative deliberations refers to things like citizens’ assemblies, juries, or panels, or deliberative polling. They're all processes that involve a representative segment of the public deliberating together in order to come up with some recommendations to policymakers.
Next is AI augmentation of representative deliberations. Because these deliberative processes are hard. They require significant skill to facilitate. They’re costly, and take up time, and that also limits the kinds of people who can participate and the kinds of places that can afford them. But we can use AI to help address a bunch of these bottlenecks, and even design entirely new kinds of processes.
For instance, a citizens’ assembly might have 100 people at its core, as it does now, but then you might also have a 10,000-person representative public that provides much broader input into that core assembly, with AI systems facilitating that input. Or you could use AI to do modeling of, say, the environmental impacts of a development, and let the citizen deliberators use those models to explore what would happen if they took this or that decision. So AI augmentation can be valuable both for making decisions with more legitimacy and for making better decisions.
The third phenomenon is called democracy-as-a-service, or deliberation-as-a-service. There's a lot of work involved in representative deliberations. So there could be a service provider that can help support governments in running these processes. That service provider could be outside the government, so it could work across different governments, share knowledge, develop best practices, and achieve cost efficiencies.
So basically like a third-party contractor for various forms of representative deliberation.
Yeah, much like an election provider supplies voting machines and the processes around them.
And then the fourth thing is platform democracy, which is when you're doing all this representative deliberation but it's for platforms like Facebook, Google, or Twitter [now X—GL].
Which I think of as a platform’s users getting a chance to influence what its policies will be, rather than just the platform’s executives and owners.
And not just its users. A platform like Facebook affects the political economy of the entire country, because of the incentives it creates for political leaders to ensure that what they say is being seen, including by spreading misinformation or taking more extreme positions. So you might be asking the general population for input on Facebook’s policies, because even someone who never uses Facebook sees posts from it on their cable news, and those influence the power structures that affect their day-to-day lives.
So the underlying thesis here is that you can use representative deliberation not just for governments, but also for decisions around things like: what should the recommender systems, which determine what we see when we open up Facebook or Twitter—what should they be rewarding? What kinds of content, what kinds of political actors? Right now they may be implicitly deciding that sensationalist content should get more attention. That's a result of algorithmic choices which may be unintentional in many cases. But having some intentionality about this is actually pretty important, because it really does affect the structure of our attention economy, and therefore of our political economy.
What are some examples of the kinds of decisions that have been taken with platform democracy? And how well has that worked?
It's still somewhat early. One of the challenges with both governments and companies is that when there are leadership transitions, or as in the case of Meta, when there’s a giant dip in the stock price, these things can get derailed. Before Elon Musk bought Twitter, they were actually going to convene a citizen assembly-type process as a pilot to help make decisions around things like ranking and Trending Topics. After Musk took over that team was laid off.
But Meta has now conducted a series of citizen assembly-lite processes and deliberative polls to help inform really challenging trade-offs, for example around privacy versus security. OpenAI has also been experimenting with lightweight deliberative processes, just in the last year or so. And Anthropic ran a process that impacted at least one specific clause of their AI constitution.
It’s good to know the platforms are at least open to the idea.
By the way, another big benefit of representative deliberations is that they are jurisdiction-agnostic. When you have an election, you’re choosing someone who will only govern one particular jurisdiction. And you need to reach everyone in that jurisdiction just for them to be able to vote. When you're talking about representative deliberations, though, you only need to talk to a small subset of a population. So that population can be a group of users, it can be a geographic region, it can be the entire planet. You just need to find a way to get a representative group of them. That opens the door to global and transnational political processes.
And that is really important. When you're thinking about platforms or AI, their “jurisdiction” is much of the planet—just like with climate, for instance. I think it's healthy to have some deliberation when you're introducing new things that could alter an ecology, like, say, genetically modified mosquitoes to combat the spread of malaria. Similarly, when you're talking about introducing AI systems that are openly available, then there are legitimate concerns about how they might be misused—risks like misinformation, child sexual abuse material, bioweapons, or terrorist attacks. And these risks are planet-wide. So you want to have very broad buy-in for regulation, because it's useless if even just one country with significant AI power decides to do its own thing.
Then, on the flipside, let's say you do decide at some point that some AI models shouldn't be made openly available. Well, then you're giving a scary level of power to a handful of private companies that control them. How do you set the guardrails? How do you have not just the companies decide that but the people decide that? And then, if these are things that cross national boundaries, how do you govern them? Deliberation provides a way to do that.
And then you just have the fact that AI moves so fast, and our current government systems don't move that fast. So anything that we can do to speed up decision-making while maintaining quality seems important for us to invest in.
It's interesting that you talk about this as speeding things up, because I think of deliberative processes as pretty slow. You convene a bunch of people. They learn about the subject. They deliberate. They come to some recommendations. That all takes time. Then they present those recommendations to lawmakers, who then have to decide what they’re going to turn into legislation. So how is that a quicker process?
Well, one of the things that slows governments down now is the lack of political will and legitimacy around a decision. If you're able to identify public agreement around a particular set of decisions through a deliberation, you can move forward in a way that you wouldn't otherwise have been able to move.
Right, Ireland used citizens’ assemblies to create legitimacy for legalizing same-sex marriage and abortion, which were issues that had been too politically hot to touch before.
Yes. deliberative processes can unblock things that were previously blocked because they can help us find understanding and common ground— and create political will behind that common ground.
Finally, what are you hoping that your new AI & Democracy Foundation will achieve?
There’s already a vibrant ecosystem of deliberative process providers. There are governments working at the highest levels around the world with those processes for informing legislation and regulation. But there's a lot of work to be done to improve those processes to make them more broadly accessible, and less costly to run.
That’s especially true if we want to move to a new constitutional order where deliberation is core. To do that we need real investment in addressing the design and operational challenges.
So one of the things that our foundation will do is map out what it would look like to have a truly mature ecosystem, and even to move from deliberative inputs, i.e. recommendations for policy, to binding deliberative outcomes—binding on governments, but also AI companies. The organizations that control and develop these powerful AI systems are among the most significant powers in the world. The ecosystem is not yet at the point where I would be, like, “Oh, this should completely replace OpenAI’s board.” I think it is possible to get to that point.
When you say a “new constitutional order,” do you mean that every political decision, everything previously done purely by legislators, should have a deliberative element to it?
Not necessarily. I mean that in the same way that we have different branches of government or different functions within an organization, there should be a clear set of issues where we need the legitimacy and buy-in from the public that deliberative processes provide. There are perverse incentives in our existing political and corporate decision-making processes. For those cases, where our existing processes are going to continue to produce the same kinds of failures, you could have these essentially disinterested people, with no direct profit or political incentive, selected by lottery to be representative of the public, who are the ones actually making the decision. If you do that you can remove or overcome those perverse incentives that politicians and corporate leadership have.
So by a constitutional order, I mean a system where, in cases where deliberative processes are used, their decisions can be binding for certain kinds of decision. Right now they’re not at the level of quality yet where you could feel comfortable handing over that power. So part of this is about improving those processes to the point where that becomes a possibility.
Links
How Colombia’s participatory democracy got torpedoed. The consulta popular, a form of local referendum enshrined in the 1991 constitution, became a way for local communities to block extractive mining and oil drilling projects. Then the courts backed the mining companies. (Ash Center)
A democracy road trip. Back Roads and Better Angels: A Journey Into the Heart of American Democracy, by Bloomberg columnist Frank Barry, is a 560-page doorstop of a road-tripping memoir. Here’s a 20-minute podcast episode where Barry discusses what he learned about what divides the country, and what can bring it together. (Bloomberg News)
Institutions under attack. It’s not just in the US, either; worldwide, populists are seeking to politicize courts and disempower independent government agencies. (Noema)
Paris gets its first law from a citizen’s assembly. The city government adopted 20 recommendations from a citizen bill on homelessness. A case study in how to get the political buy-in that’s crucial for these processes to work. (DemocracyNext)
Sudanese want the internet more than they want food, water, and physical safety. A fascinating nugget from a survey done in preparation for a deliberative polling exercise later this year. (Stanford Deliberative Democracy Lab)
Left and right unite against political violence. A heartwarming story about overcoming partisan divides in a small town 200 miles from where Donald Trump was shot. (Christian Science Monitor)
Who decides what's censorship? Tech barons and governments are duking it out for the power to set rules on speech. Expect more clashes to come. (Tech Policy Press)