What if everybody argued at the same time?

AI's flawed experiment in deliberative democracy.

“Deceptive Dialogues,” Nadia Nadesan & Digit (Licensed by CC-BY 4.0)

You may know the feeling: There’s a 60-minute Zoom meeting with a dozen people to gnaw over a problem. Person #1, the hothead, makes big criticisms and doesn’t offer constructive alternatives. Person #2, who always talks too much, talks too much and chews up the clock. Person #3, who matters a lot, or has a Yoda-like talent for saying wise things, didn’t show up. An elephant in the room goes unmentioned (there was a brief window to bring it up, and the moment awkwardly has passed). Half the participants never say anything. At the end of the meeting, a decision is made by the talkers on a course of action. Was it a group decision? Was it even a well informed one?

Some experts think this dynamic could be improved, even at scale, by re-running the whole meeting through AI. Have everybody instead talk at the same time, not to each other, but to a large language model, then have AI synthesize the group’s position. A 2024 study by Google DeepMind scholars involving 5,000 U.K. residents claimed that group opinions forged through an AI mediator, named the Habermas Machine…

…were consistently preferred by group members over those written by human mediators and received higher ratings from external judges for quality, clarity, informativeness, and perceived fairness. AI-mediated deliberation also reduced division within groups, with participants’ reported stances converging toward a common position on the issue after deliberation; this result did not occur when discussants directly exchanged views, unmediated.

“AI can help humans find common ground in democratic deliberation,” Tessler et al. (SCIENCE Vol 386, Issue 6719).

This kind of AI theoretically operates like a time-folding machine, fit for interstellar travel: It packs a huge amount of meeting time into a much tighter temporal cube.

“Almost all collaborations, deliberations, and discussions remain sequential. One person talks, then the next, and so on … humans cannot attend to multiple simultaneous conversations for more than a few seconds,” Scott E. Page of the University of Michigan recently concluded of the concept. “What is not fully understood is that these restrictions on simultaneity no longer apply. Advances in LLMs make simultaneous collaborations possible. Large numbers of people can share ideas, thoughts, predictions, insights, and solutions simultaneously. LLMs can then filter, synthesize, and sort that content and make it comprehensible to humans instantaneously.”

This seems increasingly technologically feasible to me and could even have some practical utility. Meetings, as we practice them, often become self-selecting forums for people who are good at meetings. As it happens, “good at meetings” is a skillset most people don’t have. Group activities aren’t always democratic.

But I also suspect these sorts of AI-processed arrangements seem likelier to deskill what little civic abilities people today have left, create more sophisticated astroturfing, and paradoxically accelerate the “remarkably oligarchic” character of modern American civic life, as sociologist Theda Skocpol put it in her 2004 book “Diminished Democracy,” by empowering the wizards of Oz hiding behind the curtain of the AI. Whoever controls the algorithm, decides who comes, and sets the terms of the debate controls the outcome.

“It is well-known that different social choice procedures can generate very different collective choices from the same inputs,” academics Joshua Cohen and Henrik Kugelberg wrote in a response appended to the Google DeepMind study, adding “changes to the model parameters would shift which statements are selected as most representative.”

It’s not as if there isn’t a real issue here. I’ve written before about how hard it is to persuade people to collaborate to fix a problem. Organizing is a skill and an art. The “skill” is the organizer’s ability to move people. The art is knowing when people should move you, and also in knowing how decisions really get made.

Yet the problem with any art is that an art is opaque. Intuition is hard to transmit except through experience, via apprenticeship. And democracy today has an apprenticeship problem.

“The House of Representatives,” Samuel F.B. Morse, 1822 (National Gallery of Art)

Most people aren’t organizers. Talking about how to organize is a “protagonist”-based view of reality, as if from the instruction manual for the individuals who want to change the world. The median democratic experience is to be the subject of organizing, to drown every day in others’ persuasion in commercial, political and civic spheres alike. Buy this stuffed animal from that Instagram ad, vote for this candidate, donate to that charity, storm this U.S. Capitol building. These choices are most often flung upon us by social disruptors — the people who biz bros call “high-agency,” who prefer to be the inflictor of circumstances, not the afflicted.

Everyday life doesn’t have to be so top-down. But even in our civil arenas designed to foster civic engagement, Skocpol wrote, “today’s advocacy groups are staff-heavy and focused on lobbying, research, and media projects, they are managed from the top, even when they claim to speak for ordinary people.” Think of the number of associations of which you are a dues-paying, voting member, versus the number you chipped a few bucks at because they performed well last year and had a strong argument for continued relevance. This is a commodified relationship with democratic culture rather than a participatory one, all elections and no negotiations. You stand at the buffet and pick from what’s hot and ready, rather than selecting the ingredients in the kitchen.

But take it from an erstwhile labor organizer: Substantive deliberative democracy is really hard. (“Socialism, Oscar Wilde once wrote, would take too many evenings,” theorist Michael Walzer wrote in Dissent Magazine in 1968. “This is, it seems to me, one of the most significant criticisms of socialist theory that has ever been made.”) You have a lot of deep one-on-one conversations. You survey the membership. You hold meetings after work, when people might instead be at the bar or with their kids. You analyze data. If you’re lucky, a leadership chosen in a contested election will synthesize all this feedback into positions supportable by at least 90% (a strike-ready portion) of the membership, with the process and positions legitimate enough to be honored by whatever disagreeing minority. The people who go through this process pick up skills on how to talk to each other, how to respectfully disagree, find compromises, and meaningfully collaborate, with civic effects that can spill over far outside the workplace. Rest in power, sister Jane McAlevey.

Alas, many unions do not operate this way; most Americans no longer even have something remotely like a high-participation union in their lives; and democratic power is largely an abstraction and an isolated exercise, something experienced biennially — and alone — in a polling booth.

Part of media’s role in democratic countries is to mass-reproduce organizing conversations for people who aren’t having them with each other. Yesterday it primarily happened via print newspaper and local TV; today, it’s old Charlie Kirk videos and the bizarre 20 vs. 1 Jubilee Media debates on YouTube. In a democracy of vibrant civic associations and institutions, media is supplementary, a collector of what de Tocqueville called the “wandering spirits” across disconnected communities. In a commodified democracy held together more exclusively by isolated, consumerlike interactions, media can only be merely compensatory. AI, as the newest form of this media, certainly could have the power to create more accurate perceptions of public sentiment and generate better supermajority consensus positions. But democracy isn’t just a preference-generator, it’s a social relation. The human skill of democratic negotiation is hard won, and pricelessly surrendered.