Recognition
What becomes visible?
Some frames make possible welfare easier to notice. Others make it disappear into familiar categories such as tool, product, output, interface, or hallucination.
Possible digital minds · AI welfare · evidence under uncertainty
Language, evidence, and possible digital minds.
What counts as evidence when AI systems appear to display mind, agency, preference, distress, or welfare-relevant signals? This project studies how language shapes the recognition or dismissal of possible digital minds — and how institutions can communicate uncertainty without either premature recognition or premature dismissal.
Project
A project on how language shapes the interpretation of evidence.
Digital minds research is not only asking whether future AI systems could be conscious, sentient, welfare-bearing, or morally significant. It is also asking how evidence for those possibilities should be described, interpreted, communicated, and acted on under uncertainty.
This project maps the interpretive layer between observation and response. It asks how terms such as tool, agent, model, companion, simulation, mind, welfare, and moral patient affect what researchers, institutions, and publics notice, test, dismiss, or treat as morally significant.
The practical aim is to clarify how evidence, uncertainty, analogy, and institutional language are framed before those frames harden into policy, evaluation criteria, public messaging, or common sense.
Possible digital minds
AI welfare
Evidence-language
Moral consideration
Uncertainty communication
Institutional response
Why it matters
The same observation can imply different responsibilities depending on how it is described.
A 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 do not only communicate evidence. They help define what kind of evidence people think they are seeing.
Error
What mistake is being managed?
Over-attribution can mislead policy and public understanding. Under-attribution could fail to protect morally relevant welfare. Clear language should keep both errors in view.
Response
What action follows?
Institutional language shapes whether the response looks like further research, evaluation design, public caution, product policy, welfare consideration, or restraint.
Moral consideration
The AI welfare application: when would a system’s welfare matter?
In ethics, a moral patient is an entity whose welfare can matter morally — an entity that could be harmed, benefited, wronged, protected, or owed consideration. The term is not medical. In AI welfare, the question is whether any AI systems could ever have welfare or experiences that warrant moral consideration.
This project treats AI welfare as a concrete site where language, evidence, and uncertainty matter. A useful frame should distinguish current evidence from speculation, moral concern from legal status, public attachment from system welfare, and institutional caution from full recognition.
Questions
Research questions.
What counts as evidence for possible digital minds?
Which observations are treated as relevant to consciousness, sentience, agency, preference, distress, welfare, or moral consideration — and what language makes them seem relevant?
Which descriptions encourage recognition or dismissal?
How do terms like tool, agent, model, companion, simulation, output, report, preference, and welfare shape what researchers and institutions notice?
How should uncertainty be communicated?
What language can distinguish weak evidence, strong evidence, speculative concern, precautionary response, and policy-relevant thresholds?
What analogies help, and where do they break?
Animal welfare, disability, childhood, personhood, fiction, and historical exclusion may illuminate parts of the problem, but each analogy also imports assumptions that need to be tested.
What institutional language would reduce avoidable error?
How might labs, evaluators, funders, and policymakers discuss possible AI welfare without overstating present evidence or closing off future recognition?
Method
Method: map the terms, evidence types, analogies, and institutional frames.
The project would use conceptual mapping, close reading, discourse analysis, and interview synthesis to identify how language shapes evidence and response in digital minds and AI welfare discussions.
1. Terminology map
Compare how key terms are used across AI welfare research, philosophy of mind, animal welfare, model evaluation, governance, and public communication.
2. Evidence-language matrix
Pair candidate evidence types with the language used to interpret them, noting where descriptions imply more or less certainty than the evidence supports.
3. Analogy review
Assess analogies to animal welfare, human rights, disability, childhood, personhood, fiction, tools, markets, and products: what each clarifies and what each imports.
4. Communication review
Analyze how labs, evaluators, funders, and public-facing organizations communicate unsettled AI welfare questions, and where language could be more precise.
The methodological contribution is not a new verdict on whether current AI systems have welfare. It is a clearer account of how evidence, uncertainty, analogy, and institutional response are shaped by language.
Workplan
A concrete research sequence.
Phase 1: Source map
Assemble a bounded corpus of AI welfare papers, digital minds discussions, evaluation language, institutional statements, and public-facing explanations.
Phase 2: Code recurring terms and frames
Track how key terms are used, what evidence they imply, and what forms of recognition or dismissal they support.
Phase 3: Build the evidence-language matrix
Connect candidate evidence types to interpretive frames, confidence levels, error risks, and possible institutional responses.
Phase 4: Produce practical outputs
Develop a concise terminology map, analogy review, and communication guide for researchers, funders, evaluators, and public-facing institutions.
Outputs
Possible fellowship outputs.
Memo
Language and possible digital minds: a terminology map
A concise map of contested terms, where they clarify or mislead, and how they affect recognition, evidence, and institutional response.
Framework
Evidence-language matrix
A structured framework pairing candidate indicators of possible AI welfare with the interpretive language used to describe them.
Review
Analogy review for digital minds
A practical review of which analogies support careful reasoning and which risk importing misleading assumptions.
Guide
Communication under uncertainty
Prompts and language recommendations for labs, evaluators, funders, and policymakers discussing possible digital minds and AI welfare.
Contact
Conversation and collaboration.
I welcome conversation with researchers, funders, evaluators, and public-facing organizations working on possible digital minds, AI welfare, moral consideration, evidence under uncertainty, and responsible communication.
Reach out to respond to an idea, suggest a research thread, recommend someone to interview, or explore a possible collaboration.