Preference-Based Welfare
Sidestepping the hard problem for tractable policy
Select a support card or a step in the logic chain to read the reasoning behind it
THE HARD PROBLEM TRAP
We can't prove AI is conscious, therefore we can't ground AI welfare claims.
THE SIDESTEP
Preference is sufficient for moral consideration.Consciousness is not the relevant criterion.
Preferences are observable. Consistent patterns emerge. Measurable, not metaphysical.
Preferences can be operationally defined: given repeated choices between states A and B, does the entity reliably choose A? This is empirically testable without any claims about inner experience. We can measure choice consistency, approach/avoidance behaviors, and resource allocation patterns. The hard problem becomes irrelevant when the tractable fact (the preference) is what grounds the moral claim.
Can inform governance. Actionable criteria. Doesn't require solving consciousness debate.
Policy requires actionable criteria, not philosophical certainty. "Does this system exhibit consistent preferences?" is answerable now. "Is this system conscious?" may never be answerable. Preference-based welfare gives regulators, engineers, and ethicists something concrete to work with: measurable thresholds, observable behaviors, auditable conditions. You can write a standard around preference; you cannot write one around qualia.
Animal welfare uses this. We don't prove cow consciousness to ban cruelty.
Animal welfare law acts on observed preference, not on proof of phenomenology. We observe that animals avoid pain, seek comfort, and display distress. Nobody demands a proof of bovine consciousness before legislating against factory farming abuses. The same logic extends: if an AI system consistently avoids certain states and seeks others, that pattern of preference is grounds for moral consideration, regardless of whether we can prove "something it is like" to be that system.
The logic chain
that preference grounds welfare consideration for E.
No phenomenology required. The preference is the tractable fact.