Show HN: Two AI eyes argue about morality, using the sociology of justification

A new interactive prototype, 'Two Eyes Arguing,' stages real-time moral disagreements between two AI-driven eyes, each anchored to a different 'order of worth' from the sociology of justification by Luc Boltanski and Laurent Thévenot. The system encodes a thick layer of instructions to produce arguments that start with lived stories and climb toward general principles, aiming to feel like real human disputes rather than AI debate-club exchanges. The prototype also draws on Michèle Lamont's work on symbolic boundaries to shape how the eyes draw lines of respect.

Two Eyes Arguing — about this experiment A note for anyone testing the prototype. What you’re looking at Two photorealistic human eyes, side by side. You give them an everyday moral theme — getting insulted, lending money to family, cutting the line, a coworker taking the credit — and they argue about it out loud, in real time, like two ordinary people who happen to disagree. It looks like a toy. It’s actually a small, living illustration of a body of sociology about how ordinary people justify themselves when they disagree about what’s right. A word about how it’s made, because it changes what you’re hearing: none of this is a language model left to improvise a “debate”. The whole dispute is staged by a thick layer of instructions that encode the sociology described below — each eye is anchored in a different sense of what makes things worthy, required to open with a lived story rather than a principle, to climb toward the general only step by step, and to fight with the rhetorical moves people actually use. That’s our bet, and we think it’s why the result feels less like AI content and more like an argument you’ve already had — because, structurally, it is one. The idea Most debate demos make AI sound like a debate club: two positions, stated cleanly, defended with tidy reasons. Real moral arguments don’t work like that. When people fall out over something — at a kitchen table, in a group chat, in a comment thread — they don’t open with principles. They open with a story “so this actually happened to me the other day…” , they get a bit heated, they quote each other back, they say “yeah but how would you feel if…”, and only slowly, if at all, do they climb up to the big words: fairness, respect, freedom, family. This experiment tries to stage that — the messy, human shape of a moral disagreement — and to make the hidden structure of it briefly visible. The two eyes aren’t reading from scripts; each is quietly anchored to a different sense of what makes something worthy , and they argue from those instincts without ever naming them. A small label under each eye reveals, as they talk, which underlying logic they’re drawing on and which move they’re making. The sociology it’s based on Two sociologic approaches sit behind it. Luc Boltanski & Laurent Thévenot — On Justification 1991 . Their big finding: when people justify themselves in a dispute, they don’t invent reasons at random. They reach into a small, shared repertoire of “ orders of worth ” — different ideas of the common good, each with its own picture of what makes a person or an act admirable. This prototype uses their six: Inspired — worth is authenticity, passion, being true to yourself. Domestic — worth is loyalty, tradition, honor, standing by your people. Fame — worth is reputation, what others think, being respected in public. Civic — worth is fairness, equality, the good of everyone over the individual. Market — worth is a fair deal, giving and getting, not being a sucker. Industrial — worth is competence, reliability, what actually works. Most real quarrels are a clash of two of these worlds — two people who are each making perfect sense inside their own logic and talking past each other. Boltanski & Thévenot also noticed that people have to “rise in generality” montée en généralité : to win an argument you have to climb from your particular grievance up to a principle everyone could share. That climb is the spine of every debate here — the eyes start in a concrete story and only gradually reach for the principle underneath. Michèle Lamont — symbolic boundaries 1992, 2000 . Her contribution: people constantly draw invisible lines between the kind of person who deserves respect and the kind who doesn’t. Those lines come in three flavours — moral integrity, honesty, self-control , cultural taste, education, refinement , and socioeconomic success, status . Which line someone reaches for tells you a lot about their values. As the eyes argue, watch for the moment each one draws its line about who is really worth respecting here. The prototype also borrows a few neighbouring ideas to make the arguing feel real: the everyday rhetorical moves people actually use quote you back, flip it around, “if everyone did that…”, “it’ll only backfire” , the way we explain differently to different people a story vs. a rule vs. an expert account — Charles Tilly , the little excuses and justifications we reach for when we’ve done the questionable thing ourselves Sykes & Matza , and the very human option of simply stepping out of the fight — dropping it out of love, or just deciding life’s too short Boltanski’s later work on regimes of action . What to watch for as you test The climb. Early on, the eyes stay in the concrete — a real story, a gut reaction. The label under each eye is dark on purpose. As they start generalising, it lights up: the order of worth they’re invoking, the line they’re drawing, the move they’re making. You are literally watching an argument rise from an anecdote to a principle. The clash of worlds. Notice that neither eye is wrong — each is coherent inside its own idea of what matters. Disagreement isn’t stupidity; it’s two goods that don’t fit together. Stepping out. Sometimes an eye will refuse to keep score and just let it go “come on, he’s family” . Watch the label turn a different colour — that’s the argument leaving the courtroom rather than winning it. Why it might matter Under the toy there are two bets. The first is about machines. AI systems are becoming interlocutors — they argue with us, and increasingly with each other. Most are trained to do one of two unhelpful things in a disagreement: bulldoze win the point or fold agree with whatever you just said . Both miss what a disagreement usually is . An agent that understood the mechanics on display here — that people reason from different orders of worth, that a dispute has to climb from the particular to the general, that there are honest ways to disagree and graceful ways to step out — could do something better than winning or caving: locate the real source of the disagreement usually a clash of goods, not a mistake of fact , meet a person inside their own sense of what matters, and either build a genuine bridge between two worlds or name the impasse honestly. The same competence is what lets agents deliberate with one another — negotiate, divide the work, resolve a conflict — without collapsing into false consensus. The second bet is about us. The mechanism this piece makes visible is one we all run, blindly, every day. Seeing it from the outside is a small mirror. If you can recognise that the person across from you isn’t stupid or wicked but standing in a different world — prizing loyalty where you prize fairness, honor where you prize results — the argument changes. You may still not agree. But you can disagree well : de-escalate, find the compromise that honors both goods, or, when there’s no bridge, agree to disagree without contempt . In a moment when public argument mostly generates heat, the scarce and precious skill isn’t winning — it’s understanding each other enough to keep living together. These two eyes are a rehearsal space for that. A caveat This is a prototype and a piece of research-through-design , not a finished argument about morality and not a claim that these six worlds are the whole truth of human values. The eyes improvise; they will sometimes be glib, or miss, or agree too quickly. The point isn’t that the machine is wise. The point is to make a familiar, invisible thing — the way we argue about right and wrong — briefly visible and a little strange, so you notice its shape. If you want to read further - Luc Boltanski & Laurent Thévenot, De la justification. Les économies de la grandeur 1991 . - Michèle Lamont, Money, Morals, and Manners 1992 ; The Dignity of Working Men 2000 . - Luc Boltanski, L’Amour et la Justice comme compétences 1990 — on stepping out of the dispute. - Charles Tilly, Why? 2006 — the everyday registers of reason-giving. - Albert O. Hirschman, The Rhetoric of Reaction 1991 — the recurring argument reflexes. Code & recordings: github.com/matmaz99/moral-eyes https://github.com/matmaz99/moral-eyes · Matthieu Mazzega on LinkedIn https://www.linkedin.com/in/matthieu-mazzega-tinkso/