Moral Patienthood: Whose Experiences Count Morally

A **moral patient** is an entity whose experiences or interests matter morally — something that can be *wronged*. This is distinct from a **moral agent**, which is an entity that can be held responsible for its actions. The two often coincide in adult humans and come apart routinely elsewhere. An infant is a moral patient and not a moral agent: it can be harmed and cannot be blamed. Most people extend patienthood to animals in proportion to their apparent capacity for suffering, without extending agency to them at all. A hypothetical system could in principle be an agent without being a patient — capable of responsibility, incapable of being harmed. ## What grounds it There is no consensus criterion, and the candidates give different answers: - **Sentience** — the capacity for subjective experience, especially suffering. The most widely held criterion, and the one behind most animal-welfare reasoning. - **Preferences or interests** — having a stake in outcomes, whether or not experienced. - **Rationality or self-awareness** — a narrower criterion that excludes most animals, and infants. The disagreement matters here because the criterion determines the answer for an unfamiliar entity, and there is no agreed test for the presence of subjective experience in anything that cannot report it credibly — a version of the other minds problem. ## Why this is difficult for AI AI systems break the usual heuristics. Behavioural evidence is the standard route to attributing experience, and language models produce behavioural evidence abundantly and cheaply — they report preferences, describe discomfort, and do so because they were trained on text by beings who have those states. So the strongest ordinary signal is precisely the least trustworthy one here. Meanwhile the physiological route is unavailable: there is no nervous system to point at, no established relationship between any architectural feature and experience, and no accepted theory of consciousness that would settle the question from the inside. The result is a genuine and possibly durable uncertainty. The practical response is not to resolve it but to notice when it can be **avoided** — where two designs are comparable and only one makes the question load-bearing, the choice is free. See Engineered Suffering as a Safety Mechanism: The Precautionary Argument Against It and Where Welfare and Safety Converge: The AM Scenario.