Why it matters
"Humane AI" now appears in benchmark papers, EU and US policy language, product principles, and funding calls. Each community grounds it differently: in accountability frameworks, in care ethics, in engagement metrics, in transparency obligations. The result is a term that everyone can claim and no one can be held to.
This matters because the stakes are no longer academic. Regulators are writing chatbot-safety law, evaluation teams are scoring frontier models on wellbeing, and companies are shipping "humane" features, all against definitions that do not interoperate. A field that cannot say what it means by its central term cannot measure progress toward it.
The deeper problem is buy-in. Without a shared measure, no team knows where it is falling short, and no measure holds unless the people it is applied to accept the definition underneath it. That means policy people, product people, the public, philosophers, and researchers all have to recognize themselves in the same sentence. No single community can write that sentence for the others, and the frontier labs will not always choose what is good for people; what is good should not be theirs alone to decide.
Starting material
Short framing talks open the session. Each presents an existing framework in ten minutes, as material to synthesize rather than positions to defend. The hard part they share is translation: "what do we want AI to do, and what is humane" are philosophical and sociological questions, and a usable definition has to carry them into a quantitative expression that people across communities trust.
Format
This is a working session, not a mini-conference. The room is capped at sixteen so that every participant contributes to what ships.
| 13:00 | Welcome and framingWhat a shared definition needs to do, and what version 1.0 will and won't claim. |
| 13:15 | Framing talksFour frameworks in ten minutes each, presented by organizers and invited participants: ART, HumaneBench, policy language, measurement & verification. |
| 14:00 | Synthesis, round oneSmall groups in conversation: where the frameworks agree, where they differ, and where they go silent. Output: candidate dimensions. |
| 14:45 | Break |
| 15:00 | Synthesis, round twoGroups draft definition language against the dimensions. Facilitators consolidate live. |
| 15:40 | Plenary: finalize the language, name the route to adoptionRead the draft aloud and agree what ships as v1.0. Then each table names three bodies or people who could act on it, and one person in the room who can make the introduction. That list ships with the definition. |
| 16:00 | Close |
What the session produces
Three artifacts, all public, all released under CC BY 4.0 in the workshop repository.
Contribution does not end at the door. The position statements are the first contributor base, and the repository stays open to the communities each participant represents.
Who should participate
We are selecting for people who hold a working definition of humane AI and have hit its limits. A definition that lasts needs buy-in from five constituencies: researchers, philosophers, policy people, product people, and the public. The room is built from three seat types, roughly a third each, that between them cover all five.
The public does not submit position statements, so civil society, youth advocates, and funders who represent users' interests are welcome in any seat and weighted toward the policy seat. Participation is in person; there is no remote option.
What to submit
One page. The statement is the input to synthesis, so its job is to state a definition clearly and show where it breaks, not to argue for it as the winner. Answer three questions:
- Your definition.How you or your community currently define humane AI, in your own words or by reference to a framework you use.
- Where it comes from.The tradition, regulation, product context, or evidence base it rests on.
- What nuances have you observed?Where it has been tested, stretched, or broken in practice? Where is it ambiguous or unmeasurable? How does it hold up when it travels into another field, whether policy, research, or industry?
Statements are non-archival and non-anonymous. Plain prose in any format; no template, no page style requirements, no references needed unless they help. Write it in an afternoon.
How we build the room
The call is open to anyone whose work touches this question. The organizers will read every statement and invite sixteen people to the working session. This is a working-group selection process, not peer review or paper ranking. Statements are not scored against one another; we are composing a small group with complementary perspectives, experiences, and contexts.
We will consider:
- A real definition with a real challenge. Preference for specific cases where a definition has been tested, stretched, or broken in practice.
- Complementary perspectives. We want genuinely different definitions and approaches. A strong submission may not be selected if a similar perspective is already well represented.
- Balance across contexts. We seek a mix of constituencies, regions, institutions, and organizational sizes.
- Readiness to contribute. Every seat is a working seat: writing, listening, questioning, and building together.
- Willingness to apply the work. Preference for people who can take v1.0 back to their team, organization, community, or lab and test it in practice.
With the author's consent, every submitted statement will be published alongside v1.0, so the contributor base extends beyond the room and the synthesis remains traceable to its inputs.
What this is and is not
This is not
- A paper track. Nothing is archived, reviewed, or cited as a publication.
- A venue to present your framework as the answer. Frameworks enter as material to synthesize.
- A general AI ethics discussion. The output is one definition of one term.
- A product showcase or a demo session.
- A debate with a winner, or a final word. v1.0 ships with its open points named and a path for anyone to propose changes.
This is
- Three hours of generative conversation and drafting, with sixteen people who define the term differently on purpose.
- An open-source definition in a public repository, with named contributors, named adopters, and changes by pull request.
- A first attempt at interoperability between research, policy, and product language.
- Open to anyone with a definition and a failure case.
- Free to attend for selected participants.
Submit a position statement
Answer the three questions below, or paste a link to a one page statement you have already written. Submitting sends it straight to the organizers; you will see a confirmation here and hear back by 9 October.
Organizers
Room and co-location generously arranged by the AIES 2026 organizing team at Umeå University and Malmö University.