Know Thy Self
Note: This piece was first published on Medium (May 10, 2026). For more Amor Mundi posts, visit the Medium page (updated weekly) or the official Hannah Arendt Center website, which offers a fuller archive of past writings.
In Athens this week for the World Beautiful Business Forum, I took a tour of philosophical Athens. Beyond the Parthenon, I visited Plato’s Academy, the site of the first philosophical school. Our word “academy” comes from the Greek “Akademos,” which means far from the city. Plato built his school in a grove outside the noise of the public square, a quiet place for reading, thinking, and contemplation — and also for training in gymnastics. I also walked through the ancient Athenian Agora, the public market where Plato’s teacher Socrates would talk to young Athenians, engaging them in his famous Socratic questioning, the dialectic, designed to show them that they did not know what justice, piety, courage, or the truth actually were. It was Socrates’ trial, conviction, and death for the charge of corrupting the youth of Athens that led Plato to relocate politics from the agora to the school.
In the Agora I sought out the ancient Kleroterion machine, a device common amongst the Greeks by which random selections of citizens were chosen for service on juries and political bodies. Of the hundreds of office holders selected annually in classical Athens, nearly all were chosen by lottery through machines like the Kleroterion. For the Athenians, and also for Plato’s greatest student Aristotle, it was a matter of fact that democracy required the selection of political representatives by lottery. The use of elections, Aristotle understood, would lead to a government composed of the financial, hereditary, and educated elite. Only random selection from amongst all citizens would actually result in a government by the people.
In her talks at the World Beautiful Business Forum, Claudia Chwalisz reminded us that our electoral representative democracies would be seen by the Greeks as oligarchies or aristocracies, certainly not as democracies. Because of what Bernard Manin calls the “principle of distinction,” electoral representation almost always leads to the selection of those who are respected, vetted, and credentialed: namely, the elite.
The driving force behind the growing movement for citizens’ assemblies chosen by lottery is the belief that democracy, if it is to be meaningful, must include the voices and opinions of a broader range of the people. A democracy governed only by elites may be efficient and rational, but it is hardly representative.
The American Founders knew this. They were worried about the instability and majoritarian rule of untutored democracies. They created a federalist, electoral, constitutional republic in large part to protect against the dangers of democracy. For James Madison, it was a great benefit of republican as opposed to democratic government that the elected elite would serve as a filter for the opinions of the people, keeping out of government the most radical and destabilizing opinions. We owe the stability of American democracy in great part to the Founders’ wisdom and their success in substituting consent through the election of elites for the more democratic ideal of random selection.
As brilliant as the United States Constitution is, Hannah Arendt did think it made one important mistake. While the Constitution provided deliberating societies for elite representatives of the people in the Congress and state legislatures, it did not offer institutional spaces for deliberation for the vast majority of the people. Where Arendt spoke of reimagining Thomas Jefferson’s idea of the Ward system — small debating societies in every town — there is, today, a movement to re-create mass deliberative institutions through citizens assemblies. We institutionalized representation but not citizenship.
One persistent challenge for citizens’ assemblies is to address the fact that most people — especially those who do not debate politics for a living — have opinions about the world that are emotionally powerful and yet politically naive. When asked, people have strong opinions — about immigration, taxes, and abortion. Most people hold powerful moral intuitions long before they have had the time, information, or opportunity to translate them into workable political judgments. Much of the magic that happens in citizens assemblies is that the randomly selected participants are taught, through deliberation, discussion, and moderation by trained consultants, to transform their rough opinions into political ideas that can yield a broad consensus.
Deliberation, the work of refining positions through encounters with others, takes time. That is why citizens’ assemblies meet for many days over the course of months. It is also why they are so expensive. Participants arrive with prejudices, opinions, and passions. Trained moderators facilitate conversations and encounters that allow trust to build, questions to emerge, and contradictions to settle. Over time, the participants learn not only to trust their co-assembly members. They also refine their own opinions and even build a new sense of self.
Citizens’ assemblies are powerful political and also emotional experiences. But they don’t scale. It is nearly impossible to have Socratic interlocutors in every town. While some proponents of citizens assemblies dare to imagine a time in which citizens would assemble regularly across the country for Socratic dialogue on a regular basis, such an ideal remains a fantasy.
Shutaro Aoyama, working at the intersection of AI and civic technology, has been building tools to help close this gap between fantasy and reality. He argues that he has a solution to “the last problem of politics.”
Aoyama’s diagnosis is that the bottleneck that makes politics so difficult is not the inability to work out our disagreements; it is, rather, that most of us don’t actually know what we truly think before we begin to disagree.
A person can care deeply about education, Aoyama writes, without knowing whether it is better to have smaller classes, better teachers, or better schools. It is normal to think that the system is broken and life is unfair, without knowing what part of the system is most at fault or what compromises one would make to fix it. The gap between having a political intuition and knowing what one thinks is the right political solution means that most political arguments flare up and break down even before either side has actually thought through its own position.
The great advantage of deliberative democratic processes is that exposure to different views combined with access to information, expertise, and trained facilitators can move people toward both clarifying their own beliefs and a willingness to compromise. James Fishkin’s deliberative polling has shown that deliberation shifts opinion substantially. Citizens assemblies are transformative because they bring people to see the power of deliberation for both self knowledge and the development of common sense.
What Aoyama realized is that most people come to these experiences with opinions that are not yet formed and thus not ready for real deliberation. A big part of the deliberative experience is first overcoming the gap between a raw feeling and thought-out idea of what should be done. The tool that he and his collaborators have built is not designed to settle disagreements amongst plural opinions. It is, rather, a tool to enable people to refine their own opinions before the activity of deliberation begins.
Before a citizens’ assembly — or any deliberative meeting begins — Aoyama has each participant spend about ten minutes on their phone answering a sequence of questions generated by AI. The yes or no questions are dynamically generated by the AI and are designed to help the participant refine their rough political intuitions into well-thought-out political positions.
Aoyama tells the story of a session in Ota City, Japan, where twenty residents had assembled to discuss municipal policy. When the conversation reached school lunches, a woman in her fifties — a parent whose children had gone through the public schools — said firmly that the quality had clearly gotten worse. The facilitator asked one question: “Is that a fact, or an impression?” The room paused. The woman paused. Then she said, slowly, that she thought it was an impression — that she was not sure she had evidence. That single exchange changed the meeting. The conversation stopped being a clash of slogans and became a shared inquiry. By the end of the session, the group had produced three concrete, tradeoff-aware proposals.
Aoyama’s AI tool means that a participant who walks into a deliberative process thinking “school lunches should be better” walks out of the brief pre-survey knowing something more precise: that their concern is less about cost than about whether free provision has affected quality; that she suspects quality has declined, though she is not sure whether that is fact or impression; that if quality did decline she would consider partial cost-sharing as a solution. The AI system did not supply that view. It helped her arrive at it.
Aoyama’s wager is that by having participants in citizens assemblies go through a pre-deliberative process with his AI tool, they will enter the deliberative experience with more refined opinions and better prepared to deliberate. This could accelerate the process and reduce the cost.
What strikes me in Aoyama’s account is the modesty of the claim. He is not arguing that AI will solve political disagreement. He is arguing close to the opposite: that the underlying problem of politics — that coexistence requires ongoing negotiation among beings who do not want the same things — will not be optimized away. Deliberation is still needed. But better information and the use of an AI system make deliberation better.
His pre-deliberative process can resolve factual disagreements about whether school lunch quality has actually declined. It cannot resolve the real value conflicts underneath: whether we want healthier or cheaper lunches or how much freedom we are willing to trade for safety. These are not problems with solutions. These are tensions that must be negotiated, again and again, by the people who live inside them. As AI concentrates the power to make decisions at unprecedented speed and scale, the question of whose values guide those decisions becomes, if anything, more urgent.
Aoyama is also clear-eyed about the risk that a tool like his creates. If AI helps people clarify their views, where does helping end and shaping begin? In an earlier version of his system, the AI was, in his words, eager to help determine opinion: after a short exchange it would offer to draft a proposal on the participant’s behalf. Users loved it. It also, he writes, felt wrong. The same capability that speeds up self-reflection can quietly substitute the system’s framing for the participant’s own: Which follow-up question appears first? Which tension gets highlighted? Which of five concerns gets elevated into “the main point”? These are editorial choices, and an AI makes thousands of them per session, invisibly.
Aoyama does not claim to have solved this problem. His team’s current approach aims to slow the AI process down. The AI can ask, probe, mirror, and pressure-test a participant’s views, but it does not draft the participant’s final position. The last step — the actual articulation — has to come from the participant, even when doing so introduces friction. That friction, he insists, is not a flaw. It is part of what keeps the output the participant’s own.
This is where Aoyama returns me to Athens. Socrates’ question to the young men of the Agora was never “what should the city do?” It was always, first, “what do you actually think, and how do you know?” The dialectic was a discipline of self-clarification before it was a method of public argument. It was also, of course, dangerous: Athens accused Socrates of corrupting the youth and killed him for it.
What Aoyama has built is, in a quiet way, a digital approximation of Socrates’ first move — the question before the question, the work of figuring out what one’s own position actually is before bringing it into the room. He is not naive about the risks of putting that move into the hands of an algorithm; he is the first to say the problem is unsolved. But the alternative — leaving the gap between feeling and position to chance, or to whichever loud voice happens to fill it first — is not obviously safer. The threshold he is trying to lower may be smaller than democracy. It may also be upstream of most of it. Socratic questioning — not optimization — is the real democratic technology.
You might also enjoy my talk from last year’s Arendt Forum, “On Joy: Loving the World in Dark Times,” and recent essays, “Solidarity: What Are We Fighting For?” “Pearl Diving,” “When Crimes Cease to Appear as Crimes: Hannah Arendt’s Warning as the Republic Turns 250,” “Expropriation: Notes on the Revolution to Come,.”
Learn more about the Hannah Arendt Center for Politics and Humanities at Bard College here.