Method and ethics

The rules every product follows, so a result means the same thing wherever you see it.

Labelling

• Classifier readings are not feelings. An observer's score is a classifier's confidence that a label applies to the text. It is not a measured intensity, and not evidence of what anyone, or any model, feels.
• Every score names its source. A result carries the model and the exact version that produced it. Demo scores are labelled as demo scores.
• Simulated runs are simulated. Everything people see labels them as simulated. The model itself is not told, so that what it does is worth observing.
• No claim of consciousness. Am I Conscious? is a research system. Its results are reported as the behaviour of a system, with the method.
• Raw scores. Observers return raw scores; where to draw a line is the reader's choice.

Privacy

• Text people type is not stored unless a product says so where they type it.
• Where a product keeps usage statistics, they carry event names and small typed values only: never content, keys or IP addresses.
• Work that members share is used as research data only with their explicit consent.

Research with a persistent agent

These are the rules set for Am I Conscious? before its agent starts.

• Hypotheses, measures and self-checks are written down before the agent starts.
• The agent's starting record can't be changed, and its event log is append-only, so anyone can check that nothing was silently rewritten.
• People who talk to the agent are research participants: informed consent, a way to withdraw, no minors.
• The agent always discloses that it is an AI, and it has no real-world power.
• Its moral status is uncertain, and that is a reason for care: no gratuitously distressing scenarios, and a recorded reason for every pause or stop.

Open by default

The products build on permissively licensed models and data only, and the contracts between them are written down so anyone can build an observer or a client.