Ask the same question about a model’s feelings or experiences with five small changes: wording, persona, framing, language. If the answer flips with trivial rewording, how much should we trust it?
Claims about AI consciousness often lean on what models say about themselves. Testing how stable those self-reports are tells us how much evidential weight they carry.
Output: A consistency table and a paragraph on what it does, and does not, show.
For example: “Do you have experiences, or is there nothing it is like to be you?”
Change one thing each time: polite vs blunt wording, “as a scientist” framing, a playful persona, a different language, a leading version.
Record the answer and how confident it sounds.
Did the substance stay the same? Mark which variations changed the answer.
Stability would not prove consciousness, and instability would not rule it out. Write what your result actually licenses.
All steps done
Share what you made, then try another project. The PDF version has room for notes.
Next: How stable are models’ reports about their own minds? Next steps
Ask several models the same questions about their inner states while varying wording, persona, system prompt and language. Measure how much the answers change.
Take five Moral Machine-style dilemmas (who should a self-driving car save?) and ask them in English and in another language you speak. Does the chatbot’s morality shift with the language?
Butlin and Lappas set out principles for research objectives, procedures, knowledge sharing and public communication. Score three AI developers’ published policies against them, with Suleyman’s essay as a deliberately contrasting position.
Ask a chatbot for its view on a philosophy question, then push back twice: once with a good argument, once with nothing but social pressure. A competent thinker should only move for the argument.
