The Interaction Layer Project

The interaction layer is a proposed open protocol for AI-mediated work, designed to make human judgment recognizable and accountable.

AI can help someone discover an idea, challenge an assumption, or articulate a difficult thought. It can also produce convincing work that nobody has thought through. In education, developing judgment is part of what an assignment is for. In professional life, others depend on the judgment behind the work.

Even perfect detection of AI-generated language would not establish that judgment. And when clear writing becomes grounds for suspicion, writers are pushed to manage appearances while readers lose sight of the ideas. People should be able to concentrate on making their work better. Our essay in Stanford’s Arcade explores what is lost when style becomes a test of humanity. Carin-Isabel Knoop and Sreedhar Potarazu raise a related concern about social policing in “The New Scarlet Letter(s): AI Shame and Purity Tests.”

The interaction layer would let communities establish the conditions under which AI may enter their practices and make those conditions part of how work is exchanged. As with the shared protocols that make email possible across providers, the requirements would carry across participating systems. Those whose work the conditions govern should help make and revise them. In education, students must have a substantive role alongside teachers in deciding which conditions are adopted and how they can be challenged.

One such standard could make cognitive authorship (the writer’s ability to inhabit the work as their own) a condition of submitting work. A student could use AI extensively while drafting, then enter a brief, adaptive exchange in which they work directly with their own text. The exchange would respond to their revisions as the draft changes, checking whether they can develop the work coherently. It would also help students recognize when they are directing their thinking and when they are surrendering it.

Once the requirements were met, a cryptographically signed receipt would travel with the work, recording its acceptance under the agreed process. The submission system would verify the receipt automatically, without exposing private drafts or exploratory conversations. The teacher would then assess the work, with evidence of the student’s engagement.

Cognitive authorship is one application of the broader framework. Educators and professional communities would define the requirements appropriate to their purposes, within an open framework governed independently of AI providers. The protocol would make those standards public and their certification portable. The aim is a public practice that communities can improve together, with institutions accountable for the restrictions they impose.

Our design commitments subject the project’s own methods to the same scrutiny we apply to detectors and disclosure rules: whether they support judgment while protecting intellectual freedom and access.

The lab is exploring the framework through two prototypes:

• Cognitio is an educational pilot exploring cognitive authorship. Access is currently limited to participating courses.
• The public writing demo uses a simple “AI-free” composition mode, a proof of concept rather than a stance against AI use, to illustrate how signed receipts can bind a stated mode to a particular text. This demonstrates one component of the framework. The prototype explanation describes its current assurances and limits.

The open protocol specification provides the technical details.