AI for the public good
AI that widens how humanity thinks, built in the open with the people it serves.
A short brief for AI labs, technology companies and foundations that support AI for the public good. It covers what WhatIfWe sets out to do, where it meets your priorities, and what we would build together.
In brief
Planetary problems need an intelligence wider than any one discipline, culture or tradition. WhatIfWe’s mission is to create the space for it: AI infrastructure where science, indigenous knowledge and the world’s religious and philosophical traditions work on the same real problems and reach approaches none of them could reach alone.
The design has three parts. Tradition-grounded AI agents bring each way of knowing into the conversation and carry every contribution into the terms of the others. An open global forum lets people take part from their own tradition and in their own language. One shared record keeps every agreement, open disagreement and new approach, so each piece of work starts further on. The WhatIfWe Open Benchmark (working paper) tests the agents first: can a model reason from inside a tradition, and do more than borrow its vocabulary?
The timing matters. A 2026 study of language models on ethical dilemmas from Indigenous contexts found that they adopt a tradition’s persona on the surface while defaulting to individualistic conclusions (Merrikhi, preprint). Another found that an average American was a closer match to Indian cultural values than any model trained for India, and that better prompting did little to change that (Agarwal and colleagues, 2025). Meanwhile the cost of running a language model at a given quality falls about tenfold every year (a16z, November 2024). The technology that is narrowing how the world thinks can be pointed at widening it. What we’ll build sets out the design in full.
Where it meets your priorities
| Priority | What WhatIfWe contributes |
|---|---|
| AI for the public good | Infrastructure for shared work on planetary problems, with an open record held by a non-profit |
| Pluralistic alignment | A public benchmark of whether models reason faithfully from inside many traditions, judged by practitioners of those traditions |
| Democracy and deliberation | A layer that lets deliberation platforms such as vTaiwan and Polis work across different frames of thought |
| Epistemic diversity | A direct answer to the evidence that AI narrows how people think: a system designed to widen it |
| Indigenous data sovereignty | Contributions held on terms set with each community, learning from Indigenous-led programmes such as Abundant Intelligences |
| Collective intelligence research | An open, controlled test of whether working together across traditions reaches further than pooling separate ideas |
What we would build together
- AI labs: model access, compute and evaluation expertise, and a serious public test of whether your models can reason from inside a tradition.
- Researchers in multi-agent systems, evaluation and collective intelligence: the design of the benchmark, the agents and the forum.
- Deliberation platforms: a pilot of the cross-tradition layer alongside tools already working with very large numbers of people.
- Foundations and technology philanthropy: support for the open benchmark first, because it needs only published texts and anyone can reproduce it, then for the forum and the long-running groups.
Partners describes every role, including the people inside the traditions whose judgement the whole project rests on.
Questions partners ask
What exists today?
The argument, the research and the design. Sixteen accounts written from inside each tradition, a survey of what others have built and where each one stops, the design of the system, and a list of people to talk to, tested against our own standard. The open benchmark comes next, and the system follows it.
How does this relate to existing deliberation platforms?
Platforms such as vTaiwan, Polis and DeepMind’s Habermas Machine help large groups find agreement within one shared frame. WhatIfWe adds the layer that lets genuinely different frames, scientific, indigenous, religious and philosophical, work on the same problem and produce something new together. It is designed to work alongside those platforms.
Who holds what participants contribute?
A non-profit foundation, as a steward. Every contribution stays signed and traceable to the person who made it, and what a community shares is held on the terms that community sets. The mission and these commitments are written into the foundation’s founding documents.
What do the traditions and communities receive?
Recognition, a say, and a place in the work. Each contribution carries its source, so it stays theirs. Each community sets the terms on which its knowledge is held. And as the record grows, the people who build it choose together which problems it takes on next.
What structure will WhatIfWe have?
Two linked entities, following precedents such as Mozilla, Wikimedia and Patagonia. A non-profit foundation holds the mission, the knowledge the traditions contribute, and the relationships with communities and funders. A technology company builds and runs the platform and is the vehicle for research and innovation funding. From the start the foundation holds the voting control of the company, and part of the company’s profits return to the foundation, so the mission is protected by design. Where each is incorporated will follow the partner network, and we welcome shaping the structure with founding partners, including hosting by an existing institution while the foundation is formed.
How will you know whether it works?
Through an open, controlled test. The same people and the same problems are worked two ways: separately, with the ideas pooled, and together through the system. Practitioners of the traditions involved judge, without knowing which is which, whether each contribution is still recognisable and whether the result is something no single tradition would have reached. We publish the results, whatever they show.
Why should an AI lab take part?
Because the question at its centre matters to everyone who builds models: can AI reason faithfully from inside many ways of knowing? A public benchmark judged by the traditions themselves is evidence that internal evaluation cannot produce on its own.