Updated 2026-09-09
from the news desk
What’s the difference between sentience, consciousness and moral patienthood?
Antonio Chella’s 2026 paper says sentience, consciousness and moral patienthood are related but not interchangeable. The distinction matters because Chella says evidence for one property should not be treated as automatic evidence for…
Antonio Chella’s 2026 paper says sentience, consciousness and moral patienthood answer different questions, and writes: “These notions are related but not interchangeable ( Long et al., 2024 ; Birch, 2024 ).” Chella defines sentience as welfare-relevant subjective experience, consciousness as the broader category that also includes access-, self- and report-related phenomena, and moral patienthood as the normative question of whether an entity can be wronged or benefited for its own sake.
Why the terms get mixed up
In a Perspective published on 19 August 2026, Chella says a serious research program should begin by disaggregating sentience, consciousness, self-modeling, agency, moral patienthood and AI welfare. Chella’s abstract states: “The paper distinguishes sentience, consciousness, self-modeling, metacognition, agency, moral patienthood, and AI welfare; separates evidence about an AI system from evidence about human attribution; and proposes domain-specific ordinal evidence levels rather than binary verdicts or an aggregate sentience score.”
Chalmers says part of the confusion is verbal as well as substantive. In his paper on sentience and moral status, he notes that sentience is used in more than one way, and he proposes a clearer vocabulary: “To avoid ambiguity, I will always use a prefix, speaking of consciousness sentience (which is just phenomenal consciousness) and affective sentience (which is affective consciousness).”
The Butlin report makes a related methodological point. Butlin and 18 co-authors argue that AI consciousness should be assessed through indicator properties drawn from scientific theories of consciousness, not from surface behavior alone, because: “The main alternative to this approach is to use behavioural tests for consciousness, but this method is unreliable because AI systems can be trained to mimic human behaviours while working in very different ways.”
Sentience: welfare-relevant experience
Chella defines sentience narrowly and in welfare terms: “In this paper, sentience denotes welfare-relevant subjective experience, especially valenced states such as suffering, distress, pleasure, or preference satisfaction.”
That definition makes sentience the threshold most directly tied to harm and benefit. Chella adds that indicators of access, agency or self-modeling cannot substitute for evidence specifically related to valence and welfare. In Chella’s framework, valence means the good-or-bad side of experience: whether a state is like pleasure or distress rather than just information processing.
Peter Singer frames the moral importance of sentience in similar welfare language. Singer writes: “The capacity for suffering and enjoying things is a prerequisite for having interests at all” One case for treating sentience as the key threshold rests on that idea: if suffering and enjoyment are what generate interests, then evidence about those valenced states matters first. For the site’s separate evidence explainer, see whether AI is sentient.
Consciousness: the broader empirical question
Chella draws the next line by saying: “Consciousness is broader: it includes phenomenal consciousness, as well as access-related, self-related, and report-related phenomena.”
On Chella’s account, consciousness covers more than suffering or pleasure. It also covers access-related, self-related and report-related phenomena, so a system could in principle show some consciousness-relevant capacities without yet showing welfare-relevant experience.
The Butlin report takes that broader question as its target. Butlin and co-authors say their analysis suggests no current AI systems are conscious, while also saying there are no obvious technical barriers to building systems that satisfy their indicators. One reading is that the consciousness question is empirical and architecture-sensitive, not something settled by fluent conversation alone. The site’s explainer on whether AI can be conscious tests covers that evidence question in more detail.
Moral patienthood: the normative threshold
Chella defines the third threshold this way: “Moral patienthood is a normative status: whether an entity can be wronged or benefited for its own sake.”
Chalmers uses closely related language when he writes: “A being has moral status (or moral standing, or moral patiency), to a first approximation, if it matters in moral decisions for its own sake.” Both authors, on this reading, treat moral patienthood as a normative status. The question is not just what a system can do or report, but what follows morally if the relevant mental properties are present.
Chalmers then separates two moral views that are often run together. He says he is sympathetic to consciousness sentientism, the view that consciousness is required for moral status, and he argues against affective sentientism, the view that affective consciousness is required. Chalmers states that conclusion directly: “If these judgments are correct, then affective sentientism is false. It is not the case that affective consciousness is required to have moral status.” Singer’s formulation points in a different direction because he makes suffering and enjoyment the prerequisite for interests. The site’s explainer on AI welfare vs AI rights and its glossary entry on moral patienthood cover that separate normative step.
Which question are you really asking?
Chella says every study should state which target property is under investigation and should not treat evidence for one property as automatic evidence for another. A simple way to use that advice is to separate three questions before arguing about the answer.
- Chella’s sentience question asks whether the system has welfare-relevant subjective experience, especially valenced states.
- The Butlin report’s consciousness question asks whether the system shows indicator properties derived from scientific theories of consciousness.
- Chalmers’s moral-status question asks whether the entity matters in moral decisions for its own sake.
One reading is that these questions can inform one another without collapsing into one another. Evidence that supports a consciousness indicator may change how seriously researchers take later welfare questions. A moral argument may say consciousness or sentience matters ethically. But none of those steps, on the sources’ own definitions, is automatic.
How the thresholds come apart in AI debates
Chella warns against treating nearby capacities as equivalents. In his words: “A system may possess metacognitive capacities without phenomenal experience. It may act agentically without being sentient. It may produce first-person language without self-awareness.”
That is why self-report does not settle the issue. Chella says AI self-reports should be treated as outputs requiring causal explanation rather than a direct window into sentience, and the Butlin report says behavioral tests are unreliable when systems can mimic human behavior while working differently.
The upshot is a ladder, not a single switch. On Chella’s framework, evidence for consciousness does not automatically establish sentience, and evidence for either does not automatically settle moral patienthood. The case for a sentience threshold rests on welfare and suffering, as Singer’s sentence shows. The case for a consciousness threshold rests on the view, defended by Chalmers, that a being could matter for its own sake even without pleasure or pain. Either way, the sources point to the same caution: a claim about mechanisms or reports should not be silently widened into a claim about moral status.
Frequently asked questions
Is sentience just another word for consciousness?
Chalmers says the word sentience is used ambiguously, sometimes for consciousness in general and sometimes for affective consciousness. Chella uses it more narrowly for welfare-relevant subjective experience.
Can an AI sound self-aware without being sentient?
Yes, on Chella’s framework. Chella says a system may produce first-person language without self-awareness, and he says AI self-reports should be treated as outputs that need causal explanation rather than direct evidence of sentience. Butlin and co-authors add that behavioral tests are unreliable when AI systems can mimic human behavior while working differently.
If an AI were conscious, would that mean it has moral patienthood?
Not automatically, on these sources’ definitions. Chella defines moral patienthood as a normative status, not an empirical mechanism, and Chalmers frames moral status as the question of whether a being matters in moral decisions for its own sake. Chalmers argues that affective sentience is not necessary for moral status, while Singer treats suffering and enjoyment as the prerequisite for interests.
Why does this distinction matter in AI debates?
Chella says studies should specify which property they are testing and should not treat evidence for one property as automatic evidence for another. That matters because a finding about self-report, agency or consciousness indicators would not, by itself, show welfare-relevant experience or settle the moral-status question.
Caveats: The specific three-part taxonomy used here comes chiefly from Antonio Chella’s 19 August 2026 Perspective article. Chalmers explicitly notes that philosophers use the word sentience differently, so not every author draws the boundary in exactly the same place.