Beyond the Chat BoxEssay 09 / 10
Collaborative AI
ESSAY 09

A Conversation Has More Than One Owner

Shared AI spaces need visible context ownership, permissions, and attribution.

A conversation with an assistant can look private even when the work is collaborative. A team may share a document, meeting, codebase, or project workspace. The assistant may have access to context supplied by several people, but it is not always clear whose information it can use or disclose. The idea of a single user issuing commands breaks down in shared environments.

Shared work, different permissions

Imagine a team reviewing a product launch plan. One person asks for a discussion summary. Another has added private notes. A third has access to customer data that cannot be shown to everyone. The assistant must distinguish shared artifacts from each participant's private context. Access to a workspace is not blanket permission to reveal everything any participant can access.

Make ownership visible

A collaborative interface could label information as shared, personal, restricted, or generated. A summary should show which sources contributed and allow participants to challenge an attribution. People should be able to say, “That was my suggestion,” or “This note was not part of the shared discussion.” These labels turn permissions into something people can understand at the point of use.

Decisions need attribution

The system should distinguish a suggestion, a proposal, and a decision. It should not convert tentative discussion into a team commitment. Before sending a message, changing a roadmap, or assigning work, the assistant may need approval from a role or individual with authority.

Where this can fail

The assistant may leak private context, attribute a statement to the wrong person, or mistake silence for consensus. Disagreement should remain visible rather than being smoothed into a single confident summary. Logs should preserve source attribution and the permission context used.

Questions for research

  1. How do users understand source and permission labels in AI-generated summaries?
  2. Which attribution errors cause the greatest loss of trust?
  3. How should an assistant handle disagreement between participants?
  4. Which actions require consensus, owner approval, or individual confirmation?

Design hypothesis

Collaborative AI should treat context ownership and permission as part of the interface, not as invisible backend rules.

Research starting points

  1. Clark (1996), Using Language, Cambridge University Press.
  2. Amershi et al. (2019), CHI. https://doi.org/10.1145/3290605.3300233
  3. NIST AI Risk Management Framework. https://www.nist.gov/itl/ai-risk-management-framework
← Back to all ten essays