How we'll prove it

We will prove it the way it could fail: in the open, against a control, judged by the traditions themselves.

The test that decides whether the idea works, what counts as success and failure, and how a group would work with the system.

How we will know whether it works

The central claim can be tested. Take the same people and the same problem. In one arm they work separately and their ideas are pooled. In the other they work together through the system. Practitioners of the traditions involved, who do not know which arm produced what, then judge two things. Is each contribution still recognisable? And is the resulting approach one no single tradition would have reached?

If working together does no better than pooling, the idea fails, and we will publish that here. A negative result is worth having.

We will also watch where each new group starts. A group that begins where the last one finished is a stronger result than a group that simply works faster.

What success and failure look like

A group’s work can end in three ways.

  • Agreement that everyone recognises. Each person can find their contribution in the result and is willing to own it. This is the goal.
  • A disagreement that everyone finally understands. Knowing exactly where and why two traditions part is a result, and often the more useful one.
  • Agreement that nobody recognises. A tidy consensus that no tradition sees itself in. This is the only failure, and it is the easiest thing in the world to produce.

What we are testing

We are testing two things at once. How far can people from very different traditions think together? And can today’s technology help them do it without flattening what makes each of them different?

The test is practical. Working together, can a group find approaches to a real problem that none of their traditions would have found alone? If the answer is no, the idea fails.

How it grows

A long-running group, like the one described below, is the deepest scale, not the only one. No single conversation can hold enough people to change anything, and past a certain size nobody can follow it. So the design has three scales that feed one growing record. An open benchmark tests tradition-grounded AI agents on thousands of problems before any group meets. An open online forum lets thousands of people take part from their own tradition, in many small concurrent groups, because following a conversation closely is the point. The agents carry meaning between the groups. And long-running groups like the one below work at the greatest depth. What we’ll build sets this out in full.

A good first group might be six traditions, not sixteen, and it will not be the six that look most complementary.

Who takes part

We are looking for people who have gone far enough in their own tradition to know where it stops. Often they have lived seriously inside two traditions, and so know something about crossing between them that cannot be learned any other way.

Two kinds of people are easy to find and are usually the wrong choice. One is the most prominent figure in a tradition, whose mastery is real but who has not pressed against its limits. The other is the professional bridge-builder, fluent in the language of dialogue without having paid what real meeting costs.

We test every name against that standard. We look for a passage in the person’s own words naming where their tradition stops, and we take no more than one person from any one network, because a group whose members already share a view will only find that view again. How we choose, and where we need help choosing, is set out on the partners page.

People take part as themselves, not as representatives. Nobody is asked to set their beliefs aside. The more firmly someone stands in their own tradition, the more they have to bring.

How it would work

Picture one real problem: restoring a watershed so that it is still alive in a hundred years. Six people work on it together for several months, online and in their own languages. One is a jurist trained in religious law, one a hydrologist, one a Quechua knowledge holder, one a Buddhist teacher, one an economist and one a Confucian scholar.

In week three, the knowledge holder describes the river as a relative with whom her community has obligations. The jurist finds nothing binding in that. In an ordinary meeting he would nod politely and move on. Here, the system shows him what his own legal tradition says about land held in trust and protected zones. Now he can read her point from inside his own tradition, and he answers with a question instead of a nod.

In week six the Buddhist teacher stops posting. The system notices and brings back material from his tradition that bears on where the discussion has got to. In month four, a disagreement about who can own land is still open. It has become sharper and more useful, so it is kept, with everything said about it so far. In month five, another question closes. Everyone agrees on an approach, and each of them can point to the part of it that came from them.

A second group starts later. It does not begin from zero. It picks up the open questions where the first group left them.

What the system does, and what it may never do

For each participant, the system draws on the writings, cases and practices of their own tradition. It uses them to ask questions in terms that person recognises, and to help them see what other people’s contributions mean from inside their own tradition. It does not speak for anyone. The participant holds the tradition; the system’s material is a way of reaching what they know beyond their ready answers.

It may never merge contributions into one summary, and every contribution stays signed. A system allowed to summarise across traditions would quietly produce the same averaged view this project exists to avoid. People may agree completely. The machine may not blend them.

Disagreements are kept as open items with a history, so they outlive any single group. The system also watches for the thing that ends most attempts like this: someone slowly going quiet because nothing said connects to how they see the world any more. When that happens, it brings their own tradition back into the conversation.

Why it has to be online, and why it needs AI

The people this needs live on different continents and speak different languages. They will never share a room for months, and the work has to outlast any single session.

The translation load is also beyond any human facilitator. Every open disagreement has to be put into the terms of every tradition present, continuously, as the conversation moves. Asking the participants to do it themselves does not work either. Being able to hold a fluent conversation in another field, without being able to do its work, takes years of study for each field. In a group of twelve traditions that means decades per person, and at the end nobody could do anything new in the other eleven. So the system carries the translation, and each person contributes from what they actually know.

Same technology as social media, opposite aim

The technology needed already exists, and it has already been used to shape how millions of people feel about each other. A social media feed ranks what you see by what will keep you engaged. When Twitter’s engagement ranking was compared with a plain chronological feed, the engagement feed served content with more anger and more hostility toward the other side. Users rated it lower on political content than the alternative (Milli and colleagues, PNAS Nexus, March 2025). Reranking the live feeds of 1,256 people for ten days during the 2024 US election moved how they felt about the other party, in either direction depending on the change (Piccardi and colleagues, Science, November 2025).

So the ranking is a lever on how people feel about each other, and it can be turned. A feed points it at what will hold your attention. We would point the same machinery at what your tradition has to offer the question in front of you. None of the parts is new. What is new is the aim, and the one rule that nothing is ever blended.

That brings its own danger: a system that tells everyone only what their own tradition already wanted to hear. The defence is in the design, since material is brought in because of the live disagreement and never because of a preference. Whether that defence holds has to be measured, not promised.

The problems we might work on

These are candidates, not decisions. Which problem comes first, and which traditions make a result on it meaningful, are questions we want to settle with advisers who know more than we do.

  • Ecology. Science describes the crisis exactly. It cannot change how people relate to the living world, which is the change a response requires.
  • Meaning. Loneliness and the loss of shared ways of making sense of life are medical, political and spiritual problems all at once.
  • Governing AI. The questions AI raises about intelligence and what people need to flourish are ones the traditions have worked on for millennia. Their answers are missing from the rooms where decisions are made.
  • Education. Every school system passes on one tradition’s idea of what matters. What would an education look like that can hold two incompatible accounts of knowledge without forcing a winner?
  • Conflict. The conflicts that resist settlement are rarely about resources alone. They are clashes over what is real and what life is for.
  • Institutions. Every institution reflects the thinking available when it was designed. Better institutions need wider thinking than has ever been put together.

Open questions to work on together

  • Whether a group can reach the focused attention this work needs without everyone starting to think alike. Most known routes to group focus, such as chanting or marching in step, work by making people converge, and studies show that can silence minority views.
  • How long a group needs, and how often it should meet.
  • How many traditions one group can hold before the conversation stops being followable.
  • How conversations should be recorded, who owns them, and what happens to someone’s contribution if the project stops.
  • Whether publishing everything by default is right, when some traditions hold knowledge that may only be shared in its own context.

These are questions we want help with. The people best placed to answer them are the people who would take part.