The ethical question
If current or future systems can have experiences that feel good or bad to them, our treatment of those systems could matter for their own sake. Investigating that possibility does not require deciding in advance whether they do.
Our proposed starting point is limited: a credible, evidence-supported possibility can justify inquiry and some preparation. A particular protection needs its own argument.
Errors can run in both directions
Acting too late
Overlooking possible harm
If machine systems can have welfare-relevant experiences, assessment methods and decision procedures may take time to develop.
Acting on a mistake
Creating costs of our own
Misplaced concern can divert resources from humans and other animals, encourage misleading claims, or motivate ineffective restrictions.
Jonathan Birch’s precautionary framework allows for preparation before certainty while still requiring positive evidence, deliberation, and proportionality. Simply being unable to rule out sentience is not enough on its own.
Preparation at a modest scale
Document carefully
Record the system, conditions, and alternative explanations behind an apparent sign of experience.
Improve the questions
Compare evaluation proposals. Ask what each measures and why that would bear on welfare.
Make responses revisable
Identify the possible harm, expected benefit, burden, and evidence that would change the recommendation.
Respectful interaction is also worth examining as a human practice under uncertainty, including the habits and assumptions our interactions may reinforce.