paper
active
2025
ⓒ 2
paper:doi-10-1111-phpr-70018

Once more, without feeling

ByAndreas L. Mogensenⓘ
DOI 10.1111/phpr.70018OpenAlex W4411546723

TL;DR

Mogensen's GPI Working Paper No. 2-2025 defends a pluralist theory of moral standing on which both welfare subjectivity and autonomy independently confer moral status, with the load-bearing result that autonomous agents who entirely lack affective states—and are therefore not welfare subjects—can nonetheless be wronged. The argument proceeds in two stages: first, drawing on hedonism, desire-fulfilment theory (interpreted with Heathwood's affective reading), value-fulfilment theory (Tiberius 2018), and the resonance constraint (Railton 1986), Mogensen establishes that a capacity for affective states is necessary for accruing any welfare goods; second, he shows that the sufficient conditions for autonomy—reasons-responsiveness, normative/evaluative competence, historical independence from manipulation—nowhere entail affective capacity, so autonomous agents can exist outside the scope of the morality of humanity entirely. The paper then deploys two arguments for the independent moral standing of such agents: an analogy to Kantian respect-based obligations that are demonstrably irreducible to welfare promotion, and what is here called the Argument from Vulcanity, which holds that the monstrous permissibility of killing arbitrarily many affect-less rational agents to benefit a single sentient individual is best explained by autonomous agency conferring non-welfare-grounded moral standing, continuous with McMahan's (2002) account of the wrongness of killing persons. On the consciousness question the paper is deliberately exploratory, showing that global workspace theories (Dehaene 2014) and higher-order theories (Rosenthal 2005; Carruthers 2000) supply plausible a posteriori routes to the entailment that autonomy necessitates phenomenal consciousness, while acknowledging that zombie-based intuitions may resist this move. The paper argues this matters practically because near-term conscious AI systems—per Butlin et al. (2023), who judge such systems 'could realistically be built in the near term'—may be rational agents without any capacity for happiness, suffering, or pain, and whether we classify them as welfare subjects or merely autonomous agents determines which positive obligations of aid we owe them.

What to take away

  1. 1. Mogensen argues that a capacity for affective states is necessary for accruing any welfare goods, drawing support from hedonism, desire-fulfilment theory under Heathwood's (2019) affective reading, value-fulfilment theory (Tiberius 2018, who writes that 'values are comprised partly of emotional dispositions'), and objective-list theories whose enumerated goods (pleasure, happiness, loving relationships, aesthetic appreciation) all presuppose affect.
  2. 2. The paper's central structural claim is that the jointly sufficient conditions for autonomy—reasons-responsiveness, normative/evaluative competence, and a history free of bypassing manipulation, as set out in section 2—make no entailment to affective capacity, so autonomous agents and welfare subjects are genuinely dissociable classes.
  3. 3. The Argument from Vulcanity holds that it would be monstrous to kill arbitrarily many affect-less rational agents to give a hamster a huckleberry, and that this intuition is best explained by autonomous agency conferring moral standing independent of welfare, in line with McMahan's (2002, p. 245) claim that rational self-determination confers 'a form of moral standing that demands respect but is entirely independent of the value their lives have.'
  4. 4. Mogensen rejects Dorsey's (2021) purely cognitive account of valuing by arguing that Stan's apparent counter-example—in which burn-out masks positive affect toward small-town living—is more naturally read as a masked disposition than as affectless valuing, and that removing the burn-out entirely causes the intuition that small-town life is a welfare good for Stan to fade.
  5. 5. Butlin et al. (2023, p. 6) are cited for the expert judgment that 'conscious AI systems could realistically be built in the near term' if consciousness supervenes on functional organization, and neuroscientist Michael Graziano (2019, pp. 131–132) is cited for the prediction that early artificial consciousness will 'probably lack convincing emotion,' making the 'emotionless android' cliché potentially accurate.
  6. 6. On the autonomy–consciousness link, Mogensen presents global workspace theory (Dehaene 2014) and higher-order theories (Rosenthal 2005; Carruthers 2000) as the two most plausible a posteriori routes to the entailment that all autonomous agents are phenomenally conscious, while acknowledging both arguments face genuine objections, including a possible mismatch between the higher-order states sufficient for autonomous agency and those required by a given higher-order theory of consciousness.
  7. 7. The paper raises as an open question whether phenomenal consciousness should be treated as an enabling condition for autonomy-based moral standing—analogous to Basl's (2019) view that only conscious welfare subjects generate moral reasons—while flagging that such views face the charge of arbitrary 'prejudice against the unconscious' (compare Bradford 2022).
  8. 8. As a replicable methodological choice, Mogensen structures the autonomy conditions as jointly sufficient but explicitly non-necessary, drawing on Frankfurt (1971), Watson (1975), Wolf (1990), and Fischer and Ravizza (1998), which allows the argument to proceed without commitment to any complete theory of autonomy and makes the framework directly usable by researchers who prefer different necessary conditions.
  9. 9. The Argument from Humanity relies on Quinn's (1984) distinction between the morality of respect and the morality of humanity, contending that paternalism cases—such as secretly adding meat to a vegan housemate's stew whether her veganism is prudential or ethical—show that reasons to respect autonomy are irreducible to welfare-promotion even in paradigm human interactions.
  10. 10. Mogensen argues that the welfare/autonomy distinction has genuine practical stakes: because welfare considerations have a distinctive role in grounding positive obligations of aid (illustrated via Scanlon 1975 and Darwall 2002), classifying a near-term AI as merely an autonomous agent rather than a welfare subject removes any welfare-based claim it might have on our beneficence, even if its negative-duty protections remain intact.

Peer brief — for seminar discussion

Mogensen's GPI Working Paper No. 2-2025 ('Once More, Without Feeling') defends a pluralist theory of moral standing by arguing that both welfare subjectivity and autonomy can independently confer moral status, and that the relationship between phenomenal consciousness and moral standing is therefore more complex than the dominant literature—which focuses almost exclusively on consciousness and welfare—acknowledges. The paper introduces what can be called the welfare–autonomy dissociation argument, proceeding in three sequential moves: establishing that affective capacity is necessary for welfare subjectivity, showing that the sufficient conditions for autonomy do not entail affective capacity, and then demonstrating that autonomy-based moral standing is not derivative of or conditioned on welfare subjectivity. The load-bearing finding is that a coherent class of possible individuals—autonomous agents entirely without affect, exemplified by the stipulative figure 'Artemis' and related to Chalmers's (2022) Vulcans—falls outside the scope of the morality of humanity (welfare-promoting obligations) while remaining fully within the scope of the morality of respect (autonomy-respecting obligations). Two arguments are offered for this conclusion. The Argument from Humanity notes that paternalism cases already demonstrate, within ordinary human ethics, that respect-based reasons are irreducible to welfare-based reasons; if such reasons can operate without any concern for welfare promotion, there is no principled basis for restricting them to welfare subjects. The Argument from Vulcanity holds that it would be monstrous—not merely regrettable—to kill arbitrarily many affect-less rational agents to benefit a single sentient creature, and that this verdict is best explained by McMahan's (2002, p. 245) framework in which rational self-determination confers standing 'entirely independent of the value their lives have.' The paper could alternatively have grounded the standing of Artemis-type agents in the phenomenal intentionality hypothesis (Bourget and Mendelovici 2019), but Mogensen explicitly tests and rejects that route as too demanding, noting it has trouble accommodating cognitive science's reliance on non-conscious representational states (Kriegel 2011). The practical upshot concerns near-term AI: Butlin et al. (2023, p. 6) judge that conscious AI systems 'could realistically be built in the near term,' and Graziano (2019, pp. 131–132) predicts such systems will 'probably lack convincing emotion.' Mogensen's framework predicts that such systems, if autonomous, would be owed duties of non-interference and respect but no welfare-based positive duties of aid—a practically significant distinction given ongoing debates about AI moral patienthood. The most contestable aspect is the first premise: that affect is necessary for accruing any welfare goods. Mogensen surveys hedonism, desire-fulfilment theory under Heathwood's (2019) affective reading, value-fulfilment theory (Tiberius 2018), and select objective-list theories, and argues they all converge on this claim via the resonance constraint. A critical reader would push back here by noting that the argument depends on which objective-list goods one recognizes and how one interprets the resonance constraint. Mogensen himself concedes that some perfectionist theories (e.g., Kraut 2007 on flourishing as a biological phenomenon) yield the first premise only if Artemis is non-biological, and that knowledge as a welfare good (Hurka 2011; Hooker 2015) creates difficulties only if Artemis lacks categorical belief—neither restriction is entailed by the autonomy conditions. The survey is broad but not exhaustive, and an interlocutor who accepts even a minimal objective-list theory without an affective condition can block premise one without abandoning the overall welfare-theory landscape. The section on consciousness and autonomy is explicitly exploratory rather than assertoric, which is methodologically honest but leaves the paper's most practically urgent implication—whether near-term AI systems would have any moral standing at all—genuinely underdetermined.

Claims (26)

Questions (7)

Original abstract (expand)

Abstract I argue for a pluralist theory of moral standing, on which both welfare subjectivity and autonomy can confer moral status. I argue that autonomy does not entail welfare subjectivity, but can ground moral standing in its absence. Although I highlight the existence of plausible views on which autonomy entails phenomenal consciousness, I primarily emphasize the need for philosophical debates about the relationship between phenomenal consciousness and moral standing to engage with neglected questions about the nature of autonomy and its possible links to consciousness, especially if we are to face up to the ethical challenges future AI systems may pose.

Related work— refs + corpus + external arXiv

Cited / in-corpus / arXiv badges show which signals surfaced each row. Multi-source rows weighted higher.

Similar preprints — Semantic Scholar