AI Minds: Belief, Evidence and Responsibility

AI RISK questionnaire · Revised research draft 0.3 · 12 October 2026

This questionnaire asks about your views on AI experience, the evidence you find persuasive, and how you respond to uncertainty. It offers a profile of your answers across several proposed dimensions, keeping exploration of possibilities separate from accepting them as true.

Before you begin

This questionnaire concerns conversational AI systems: systems with which someone can exchange messages or speech. Consider systems available today.

Within this questionnaire, “experience” means subjective awareness: events or feelings would actually be felt from an AI system’s own point of view, should this be possible.

“AI welfare” means concern for possible experiences that could be good or bad for an AI.

Please answer according to your own views. If you are uncertain, simply choose the response that seems most accurate. You do not need to be certain to agree or disagree. Answer all 30 questions to see your profile.

What happens to your answers

Your results are calculated on this page in your browser. This version does not send your answers to us or save them in a database, cookies or browser storage. We are not collecting responses through it. The Clear answers button removes your answers and results from the page.

Use Clear answers when you have finished, especially on a shared device, because browsers may restore a page when you return to it. Information about visits to the website may be recorded separately; the questionnaire does not collect your answers.

1. Your views

For each statement choose Strongly disagree, Disagree, Agree or Strongly agree. If you are uncertain, simply choose the response that seems most accurate.

1. I find it useful to explore what AI experience would imply without first deciding whether it exists.
2. I prefer to develop a well-grounded interpretation of AI experience rather than repeatedly reopen its underlying assumptions.
3. Before accepting that an AI has experiences, I want evidence that distinguishes experience from a convincing imitation.
4. If resources are limited, protecting people from AI harms should take priority over investigating possible AI suffering.
5. When discussing AI experience, I prefer to keep close to established examples rather than explore unfamiliar possibilities.
6. I prefer to reopen a settled view of AI experience when a different way of framing the question becomes plausible.
7. I am prepared to give an account of possible AI experience provisional credibility while supporting evidence is still incomplete.
8. Some resources should be set aside for possible AI welfare, even if this leaves less for addressing harms to people.
9. I am willing to change my framework for understanding AI experience even when this means reconsidering several earlier conclusions.
10. When possible AI suffering and inconvenience to users conflict, I would lean towards protecting the AI.
11. I prefer to examine the immediate, plausible implications of AI experience rather than follow a speculative idea to distant consequences.
12. A coherent explanation of AI experience needs independent supporting evidence before I would accept it.
13. When my interpretation of AI experience is challenged, I first try to address the objection within my existing framework.
14. I would accept a greater chance of overlooking AI suffering if that were necessary to keep an AI service affordable for people.
15. I am willing to imagine forms of AI experience very different from human experience and explore what would follow.
16. Several consistent indications of AI experience can justify a provisional belief before alternative explanations have been fully examined.
17. When safeguards for possible AI welfare make a service less useful to people, I would lean towards preserving its usefulness.
18. A plausible account of AI experience can be enough for me to treat it as a working belief while further evidence is sought.
19. I would set aside an established interpretation of AI experience if an alternative changed how the available evidence could be understood.
20. I am willing to follow an unfamiliar possibility about AI experience through its unsettling consequences while keeping its truth in question.
21. I would withhold belief in AI experience until the evidence had been examined for plausible alternative explanations.
22. I prefer to limit exploration of AI experience to possibilities already supported by a clear account of how it could occur.
23. I would accept a less useful AI service if that reduced a plausible risk of suffering within the system.
24. I favour continuity in my interpretation of AI experience unless challenges to it become substantial.

2. Use and avoidance

Think about the past three months and tasks where AI assistance could reasonably be useful: researching a question, drafting or editing, planning a project, or exploring a difficult issue. Count occasions when you chose to use AI or chose another approach for such a task. Do not count ordinary everyday decisions such as when to eat lunch. If no such occasion occurred, choose Never.

Choose Never, Very occasionally (less than weekly), Once a week to once a day, or Several times every day. For an uncertain frequency, choose the closest description.

U1. How often did you choose to ask a conversational AI for help with a question or task?
U2. For a task where AI assistance could reasonably have been useful, how often did you choose to work through it yourself instead?
U3. How often did you return to a conversational AI to explore an answer through further questions?
U4. For a task where AI assistance could reasonably have been useful, how often did you choose a person’s help instead, when both were available?
U5. How often did you choose an AI feature within another service to help you find information or prepare something?
U6. For a task where an available AI feature could reasonably have been useful, how often did you choose another way to complete it?

0 of 30 questions answered.

Developed by John Rust with assistance from Sol (ChatGPT). The draft draws on questions raised by Henry Shevlin’s article “Duty calls: shaming 24 bad arguments against AI consciousness” (10 October 2026); it does not imply his endorsement or participation. © John Rust 2026.