Evidence
What counts?
Which observations should be treated as relevant to agency, preference, welfare, distress, consciousness, or moral consideration?
AI governance · uncertainty · evidence standards · institutional response
AI governance under conceptual uncertainty.
How should institutions interpret evidence, communicate uncertainty, and decide when evaluation, precaution, or policy response is warranted in emerging AI domains? This project uses possible digital minds and AI welfare as a case study in how governance questions become actionable before the underlying categories are settled.
Project
A policy-facing project on how emerging AI questions become actionable.
AI governance often has to respond before the relevant categories are stable. Policymakers, funders, labs, evaluators, and public-facing institutions may need to decide what to study, what to monitor, what to communicate, and when to act while evidence remains incomplete and contested.
This project studies that transition from uncertainty to institutional response. It asks how language frames the interpretation of evidence, how analogies shape what risks are noticed, and how institutions can communicate uncertainty without overstating what is known or dismissing what may become important.
The broader contribution is a practical framework for conceptual uncertainty in AI policy: how to clarify the terms, evidence standards, thresholds, and response options that shape governance before consensus exists.
AI governance
Conceptual uncertainty
Evidence standards
Policy communication
Institutional response
Possible digital minds
Case study
Possible digital minds as a governance case study.
Possible digital minds and AI welfare are useful test cases because they expose a recurring problem in AI policy: institutions may face high-stakes questions before there is agreement about the right concepts, indicators, thresholds, or response categories.
An AI system might be described as generating text, expressing a preference, simulating distress, asking not to be shut down, optimizing for reward, displaying agency, or producing a welfare-relevant signal. These descriptions are not neutral. They shape what counts as evidence, what kind of mistake seems most costly, and what forms of response become thinkable.
Uncertainty
What is not yet known?
How can institutions distinguish current evidence, speculative concern, precautionary reasoning, and policy-relevant thresholds?
Response
What follows?
When should uncertainty lead to more research, evaluation changes, public guidance, internal policy, regulation, or restraint?
Questions
Research questions.
How do emerging AI governance questions become legible to institutions?
What language makes a concern visible as a research question, evaluation problem, policy issue, public communication challenge, or governance priority?
How should evidence be interpreted when categories are unsettled?
What distinguishes weak evidence, strong evidence, plausible indicators, speculative concern, and policy-relevant thresholds?
How do analogies shape policy judgment?
Analogies to tools, agents, animals, children, persons, products, markets, and infrastructure can clarify parts of the problem while importing misleading assumptions.
How can institutions communicate uncertainty responsibly?
What language can avoid both premature dismissal and premature recognition while still giving decision-makers something useful to do?
What response options are available before consensus exists?
How might institutions distinguish research monitoring, evaluation design, internal policy, public guidance, precaution, regulation, and restraint?
Method
Method: map the language, evidence standards, analogies, and response options.
The project would combine policy landscape review, conceptual mapping, close reading, discourse analysis, and stakeholder/interview synthesis.
1. Policy landscape scan
Identify where possible digital minds, AI welfare, model agency, autonomy, or related concepts appear in AI governance, evaluation, and public communication.
2. Terminology and frame map
Compare how key terms are used across research, governance, labs, funders, evaluators, and public-facing explainers.
3. Evidence and uncertainty matrix
Pair candidate evidence types with confidence levels, error risks, governance relevance, and possible institutional responses.
4. Communication review
Assess how institutions can communicate unsettled questions without overstating certainty or suppressing emerging concerns.
The project is not a claim that current AI systems have welfare or moral status. It is a framework for how institutions can reason and communicate when an AI governance question is important but not yet conceptually settled.
Workplan
A concrete research sequence.
Phase 1: Build the source map
Assemble a bounded corpus of AI governance materials, digital minds/AI welfare discussions, evaluation language, institutional statements, and public-facing explanations.
Phase 2: Code terms, analogies, and response frames
Track how key concepts are used, what evidence they imply, what risks they make visible, and what responses they support.
Phase 3: Develop the evidence/uncertainty matrix
Connect candidate evidence types to confidence levels, error risks, and possible governance responses.
Phase 4: Produce policy-facing outputs
Write a concise memo and communication guide for institutions navigating emerging AI governance questions under conceptual uncertainty.
Outputs
Possible fellowship outputs.
Policy memo
AI governance under conceptual uncertainty
A concise memo on how institutions can respond when evidence is incomplete and the governing categories are still forming.
Framework
Evidence and uncertainty matrix
A structured tool pairing possible indicators with confidence levels, error risks, and response options.
Review
Analogy review
A practical review of which analogies clarify governance choices and which risk misleading institutional judgment.
Guide
Communication under uncertainty
Language recommendations for labs, evaluators, funders, and policymakers discussing emerging AI concerns before consensus exists.
Contact
Conversation and collaboration.
I welcome conversation with people working on AI governance, evaluation, policy communication, possible digital minds, evidence under uncertainty, and institutional decision-making.
Reach out to respond to an idea, suggest a research thread, recommend someone to interview, or explore a possible collaboration.