What are we actually trying to learn?
LBB began with a lived experience and a conceptual question. Neither establishes that the intervention works. The purpose of Pilot Zero is to turn some of those questions into things that can be examined.
Research in progress: from generation to independent testing
October 23, 2026 | Kathleen C. Thompson, J.D. — Exploratory research artifact; symposium abstract submitted, with selection pending.
On October 23, 2026, I submitted a proposal to Disciplining AI, a symposium hosted by the Harvard Journal of Law & Technology and the University of Iowa's Innovation, Business, and Law Center.
The proposed paper addresses a specific problem: when conversational AI contributes to harm, how can corrective interventions be independently tested for effectiveness, rather than relying exclusively on providers' internal assessments?
That inquiry emerges from a broader, still-developing research program concerning machine-augmented inquiry, evidentiary status, human judgment, and the Refraction Point.
The accompanying notebook entry documents one instance of AI-assisted theoretical development. Beginning with a metaphor about language and perception, the model generated a broader proposal involving civic participation, institutional accountability, and distributed economic governance.
That response is preserved as an artifact of the inquiry, not as independent validation of its claims.
The research question is not whether a language model can produce a persuasive theory.
It is what happens when generated representations influence subsequent human judgment — and what methods can establish when further generation should yield to evidence, independent examination, or real-world participation.
The proposal was submitted. Selection has not occurred, and nothing here implies acceptance or endorsement.
- Submitted to
- Disciplining AI — Harvard Journal of Law & Technology · University of Iowa Innovation, Business & Law Center
- Submitted
- October 23, 2026
- Status
- Selection pending. Not accepted. Not peer reviewed.
- Proposed paper
- Independent evaluation of post-harm AI interventions, rather than providers' internal assessments
- Title and abstract
- Available on request
- Contact
- kathleen@atrialofcolor.com
Primary research question
Can AI prepare a decision space in a way that helps humans distinguish what an AI-generated representation establishes from what still requires verification, routing, uncertainty, or stopping?
The goal is not to make users distrust AI. The goal is not to increase trust. The proposed outcome is better-calibrated judgment about what kind of answer the user has and what competent next move follows.
Three things we must not confuse
Pilot Zero comparison
What we might measure
Current hypotheses
Pilot Zero must allow that result.
What LBB does not yet know
We do not yet know whether:
- ·FLASHLIGHT improves judgment.
- ·the Board helps more than it distracts.
- ·“jurisdiction” is understandable to ordinary users.
- ·explicit status labels transfer to new tasks.
- ·AI prefill creates anchoring effects.
- ·users meaningfully override LBB.
- ·VERIFY / ROUTE / HOLD / STOP become genuine alternatives to generating again.
- ·World Return changes conclusions or merely creates more paperwork.
- ·educators will find the system useful.
- ·the terminology survives contact with prior art and researchers.
- ·a simpler intervention performs just as well or better.
Current product status
- · AI-first artifact analysis
- · claim extraction
- · proposed status
- · established / not established
- · uncertainty / missing information
- · jurisdiction routing
- · recommended moves
- · Board routing
- · human disposition
- · full World Return analysis
- · before/after status comparison
- · Return Cards
- · learning modules
- · Pilot Zero study tooling
- · team collaboration
- · institutional deployment
- · larger curriculum
- · longitudinal research
- · expanded agent Board
- · resource / circulation infrastructure
Research materials
Participate in the research
Help us find out whether this works.