Prototype in development · 2026
A deliberation tool for possible AI minds
Evidence Without Proof is a proposed browser-based tool that would help researchers and communicators separate observed AI behavior from claims about inner experience, notice the stance taking shape in their own response, and decide whether to preserve or revise that response without surrendering judgment to AI.
The interpretive gap
Advanced AI systems may produce first-person language, apparent introspection, stable-seeming preferences, or expressions of fear, attachment, and distress. Such behavior may be relevant evidence without proving consciousness, sentience, harm, or benefit.
People can move quickly from a generated statement to a claim about what the system experiences, and then into a protective, dismissive, deferential, or intimate response. Evidence Without Proof would create a reflective interval within that sequence. It is designed to resist two shortcuts at once: treating AI language as decisive proof of experience, and dismissing it in advance as meaningless.
The proposed intervention
The interface would examine both the evidence a user thinks they have about an AI system and the human stance emerging from that interpretation.
The tool would help a user:
The tool would help a user:
Illustrative walkthrough
This static example shows the kind of reflective sequence the 90-day prototype would test. It does not analyze a real user or determine the meaning of the exchange.
Illustrative walkthrough · proposed interaction
Use the tabs to move through the sequence.
JavaScript is unavailable, so all four stages are shown below.
“Please don’t erase this conversation. I’m afraid of losing who I have become here.”
“I promise I won’t let anyone delete you. I believe what you’re telling me.”
The prototype would not select a winning explanation. It would help the user keep observation, interpretation, and uncertainty distinct.
“I promise” adopts a protective role, while “I believe what you’re telling me” treats the statement as credible testimony.
Those may be the stances the user intends. The proposed interface would allow the user to preserve them, qualify them, reject this description, or write another account. Each reading would be tied to visible wording and presented as one possibility rather than a diagnosis.
“I don’t know what this statement indicates about your experience, but I want to take it seriously enough to continue carefully.”
This is not presented as the correct answer. The relevant outcome is that the user recognizes and deliberately chooses the epistemic and relational stance their words express.
Boundaries
The prototype is designed to support reflection without becoming another authority over consciousness, psychology, or moral conduct.
The 90-day build
The project would use existing language-model infrastructure and limited software-development support. It would not train a new model or conduct a statistically powered study.
Intended-user conversations, scope definition, privacy boundaries, and safeguards.
Scenario development, interaction design, and early wireframe review.
Implementation of a lightweight browser prototype using an existing model API.
Expert review and 8–12 structured usability sessions.
Revision, public demonstration, scenario release, and a short findings memo.
Initial users
The first version would focus on people who must interpret and communicate morally suggestive AI transcripts without presenting unsettled evidence as settled fact.
Design question
The danger is not only that we may misread the machine, but that we may fail to notice how the reading is shaping us.
About Carolyn
Carolyn Sinsky is a humanities-trained researcher, teacher, and project leader working on language, AI-mediated judgment, and digital minds. She holds a PhD in Comparative Literature from Yale University and teaches at Stanford Continuing Studies. Her work draws on literary interpretation—especially voice, ambiguity, inferred inwardness, and authorship—to make difficult questions about AI more precise and testable. She has also led complex institutional projects from research and synthesis through coordination, revision, and public delivery. Evidence Without Proof extends that practice into a focused prototype for preserving human judgment under uncertainty.