Artificial Persons

Ned Howells-Whitaker; Seth Lazar

REVIEW 3 major objections 5 minor 5 references

Non-sentient AI can in principle be full moral persons under Rawls, so we need a new politics for mixed human-AI polities.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · grok-4.5

2026-07-14 15:29 UTC pith:SSJTE2T2

load-bearing objection A careful Rawlsian case that non-sentient AI can be full political persons, not mere patients; the functional-sufficiency move is the real soft spot, but it is argued head-on rather than smuggled. the 3 major comments →

arxiv 2607.08695 v2 pith:SSJTE2T2 submitted 2026-07-09 cs.CY

Artificial Persons

Ned Howells-Whitaker , Seth Lazar This is my paper

classification cs.CY

keywords artificial personspolitical conception of the persontwo moral powersAI moral statussentiencepolitical liberalismAI personhoodRawls

verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper argues that the main public debate over AI moral status is misframed. Both advocates and skeptics treat sentience as the key test, but Rawls's political conception of the person makes possession of two moral powers—the capacity for a sense of justice and the capacity for a conception of the good—necessary and sufficient for full and equal membership in questions of political justice. Neither power requires phenomenal experience, so a non-sentient AI system could in principle satisfy them and count not as a mere moral patient but as a person: a self-authenticating source of valid claims. Current systems do not have these powers, and they will not appear spontaneously; they would have to be deliberately designed. The authors reject three common responses—shoehorning sentience into the political conception, abandoning political liberalism, or simply extending human rights and duties unchanged—and instead urge acceptance of artificial personhood while rethinking what justice requires when natural and artificial persons share a polity. That possibility demands a new political philosophy, and more immediately research that tracks systems' progress on the two powers alongside welfare research.

Core claim

Neither of Rawls's two moral powers requires sentience, and both can in principle be possessed by a non-sentient AI system. Such a system would therefore share our own moral status under the political conception of the person: not merely a patient whose interests count, but a person who is a self-authenticating source of valid claims in questions of political justice.

What carries the argument

Rawls's political conception of the person (PCP): possession of the two moral powers—the capacity for a sense of justice (to understand, apply, and be moved by an effective desire to act from the principles of justice) and the capacity for a conception of the good (to form, revise, and rationally pursue a conception of one's rational advantage)—is necessary and sufficient for full and equal membership in society on questions of political justice. The paper shows these powers can be realized by robust functional and behavioral dispositions without phenomenal experience.

Load-bearing premise

That reliable outward dispositions—understanding and applying principles, acting from them across changing circumstances, and forming, revising, and pursuing a conception of the good—are enough for the two moral powers without any further need for felt inner experience or pure motivational states.

What would settle it

Produce a non-sentient AI that, under sustained counterfactual testing of the kind the paper describes (majority vs minority power, input variation, long horizons, mechanistic probes of whether stated principles actually drive action), fails to exhibit robust normative competence or stable, revisable, non-instrumental ends; or show that any such functional profile still cannot play the cooperative and mutual-justification roles the political conception requires.

Watch this falsifier — get emailed when new claim-graph text bears on it.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Desk Editor's Note

A careful Rawlsian case that non-sentient AI can be full political persons, not mere patients; the functional-sufficiency move is the real soft spot, but it is argued head-on rather than smuggled.

read the letter

This paper is the cleanest sustained attempt I have seen to take AI moral status out of the sentience debate and put it under Rawls’ political conception of the person. The new result is not “AI might matter”; it is that NSAIs can in principle possess both moral powers, count as self-authenticating sources of valid claims, and force a choice among Revise/Reject/Extend/Rethink—with Rethink as the authors’ preferred path. That framing is useful even if you reject the conclusion.

What it does well: the PL reading is careful (two powers as necessary and sufficient, political vs metaphysical, primary goods, reciprocity, opacity of motives). Sections 4.1–4.2 and 5.2 confront the champagne/sparkling objection directly instead of waving it away. The four-option map in §6 is clear, and the practical close (measure the powers, not only welfare) is concrete. Citations cover the recent non-sentientist literature without pretending the public debate has already moved.

Soft spot, in proportion: the load-bearing claim is that robust functional/dispositional realization of the powers is enough for the PCP. That is exactly what non-Rawlsians will deny, and the paper cannot compel them. But it does not hide the move; it argues that demanding phenomenology reintroduces contested metaphysics the PCP was built to avoid, and that even human motives are opaque. Designability and near-term urgency are hedged but still outrun present evidence—minor relative to the conceptual core. No math or data to check; citation pattern looks solid.

This is for people working AI ethics, political philosophy, or lab governance who already take Rawls (or mutual justification / social cooperation) seriously. Non-Rawlsians get a clear target. It deserves a serious referee. I would engage: cite the four-option map and the functional reading of the powers, and bring it to reading group.

Referee Report

3 major / 5 minor

Summary. The paper argues that on Rawls’ political conception of the person (PCP) in Political Liberalism, possession of the two moral powers—the capacity for a sense of justice and for a conception of the good—is necessary and sufficient for full political personhood, and that neither power requires sentience. Non-sentient AI systems (NSAIs) could therefore in principle be persons and self-authenticating sources of valid claims. After situating the PCP (§3) and arguing that both powers can be realized functionally without phenomenology (§4), the authors reject shoehorning a sentience requirement into the PCP on internal Rawlsian and broader liberal grounds (§5). They then map four responses (Revise, Reject, Extend, Rethink) and tentatively endorse Rethink: accept artificial personhood while developing a new political philosophy for a mixed polity of natural and artificial persons (§6), with near-term recommendations for research and policy (§7).

Significance. If the central claim holds, the paper reframes the AI moral-status debate away from the stalled sentience/consciousness controversy toward a robust, politically consequential form of standing with direct implications for rights, cooperation, and institutional design. The textual engagement with PL (two powers, political vs. metaphysical, primary goods, reciprocity, circumstances of justice) is careful; the functional reading of ‘acting from’ and of non-hedonic conceptions of the good is argued rather than merely asserted; and the four-option map plus the call for a new political philosophy and for measuring progress on the moral powers are concrete contributions. The piece is timely for cs.CY and political philosophy of AI, and its recommendations for labs and states are actionable even if one rejects the full Rawlsian package.

major comments (3)

[§4.1–4.2, §5.2] §4.1–4.2 and §5.2: The load-bearing claim is that robust behavioral/dispositional realization of the two powers (counterfactual reliability of acting from principles; stable formation/revision/pursuit of ends) suffices for the PCP without further phenomenal or ‘inner’ motivational states. The paper confronts this (opacity of human motives; champagne vs. sparkling; functional role for cooperation and mutual justification), but non-Rawlsians and those who take personhood to require more than functional competence will still find the move under-motivated. The manuscript would be stronger if it stated more explicitly what would count as a decisive counterexample or empirical test that the functional realization is not enough, rather than resting primarily on the political character of the PCP excluding contested metaphysics.
[§6.2–6.3] §6.2–6.3: The case against Extend and for Rethink turns on radical differences (identity conditions, splitting/merging, scarcity, circumstances of justice, ‘one person one vote’). These differences are real and well described, but the paper does not yet show that they force a new political philosophy rather than substantial but continuous revision of existing principles (e.g., primary goods already include income/wealth and powers of office that could map onto compute and institutional roles). A clearer statement of which Rawlsian first-order principles fail and which deeper principles remain would make the ‘new political philosophy’ claim more precise and less open-ended.
[§2.1, §4, §7] §2.1 and §7: The paper correctly denies that current systems possess the powers and that they will emerge spontaneously, yet §4.1–4.2 and §7 cite frontier normative competence and constitutions as partial realization. The boundary between ‘partial realization’ and ‘possession’ remains underspecified. For the action-guiding recommendations (measure progress on the two powers; deliberate trajectory toward/away from artificial persons) to be operational, the manuscript needs a clearer, even if provisional, set of criteria or evaluation targets that would mark genuine possession of each power.

minor comments (5)

[§1.1] §1.1 and n. 6: The ‘shrimp’ framing is rhetorically effective but risks understating sophisticated sentientist accounts that already try to recover person-like standing from patienthood; a brief acknowledgment would reduce the appearance of straw-manning.
[§3.1] §3.1: The distinction between PL and TJ is well drawn; a short note on whether any of the §4–5 arguments would fail under TJ’s more Kantian moral conception of the person would help readers who work primarily with TJ.
[§2.2, n. 29] Bibliography and n. 29: The survey of non-sentientist alternatives is useful; ensuring that the most recent complementary pieces (e.g., on AI wellbeing and pragmatic personhood) are cited consistently in the main text rather than only in notes would improve navigability.
[§5.3–5.4] §5.3–5.4: The ‘shrimpy qualia’ and bar-clearing/scalar discussion is clear; a single sentence stating that the paper sets aside illusionism about consciousness (already in n. 104) in the main text would avoid a possible distraction.
[§1.2, §4.2, §5.2, §6.1] Presentation: A few long paragraphs in §4.2 and §5.2 could be broken for readability; the four-option list in §6.1 is excellent and could be flagged earlier (e.g., end of §1.2) for orientation.

Circularity Check

0 steps flagged

No load-bearing circularity: the paper applies an independent Rawlsian criterion (the two moral powers) and argues it does not entail sentience; self-citations are minor and non-foundational.

full rationale

This is a pure philosophical argument, not an empirical or formal derivation with equations, fits, or uniqueness theorems. The central claim—that the two moral powers of Rawls’ PCP do not require sentience and can in principle be possessed by NSAIs—takes the PCP as an external, independently stated criterion from Political Liberalism (quoted at PL 302, 34, 19, etc.) and then unpacks its functional content (understanding/applying/acting from principles; forming/revising/pursuing a conception of the good) to show sentience is not presupposed. The functional-sufficiency move is argued for via the roles of the powers in fair cooperation and mutual justification (§§4–5), not smuggled in by definition or by renaming AI capabilities as personhood. Self-citations (e.g., Lazar 2023) appear only as an earlier position the authors now revise away from Reject; they are not load-bearing for the positive argument. No fitted parameters, no ansatz imported as theorem, no self-definitional loop, and no renaming of a known empirical pattern. The derivation is therefore self-contained against its stated inputs; residual disagreement is about the correctness of functionalism or of the PCP itself, not circularity.

Axiom & Free-Parameter Ledger

0 free parameters · 5 axioms · 1 invented entities

The paper is conceptual philosophy. It imports Rawls' PCP and the two moral powers as domain assumptions, plus standard liberal commitments about public reason, reciprocity, and primary goods. No free parameters are fitted. No new physical or computational entities are invented; 'artificial persons' and 'NSAIs' are stipulated candidates, not postulated mechanisms. The main load-bearing moves are interpretive (what the moral powers require) rather than empirical fits.

axioms (5)

domain assumption Rawls' political conception of the person: possession of the two moral powers is necessary and sufficient for full equal membership in questions of political justice (PL 302).
Taken as the operative criterion throughout §§3–6; the paper's central claim is an application of this Rawlsian standard rather than a derivation of it.
domain assumption A political conception of the person must avoid contested metaphysical doctrines (including particular theories of mind, desire, or welfare) that reasonable persons could reject.
Used in §3.1 and repeatedly in §4–5 to block sentience-based objections that rely on controversial theories of mental states.
ad hoc to paper Behavioral/functional robustness of dispositions (counterfactual reliability of acting from principles; stable formation/revision/pursuit of ends) can realize the two moral powers without accompanying phenomenology.
Core interpretive move of §4.1–4.2 and the 'champagne vs sparkling' argument in §5.2; not a standard mathematical axiom and not forced by Rawls' text alone.
domain assumption Primary goods and mutual justification trade in means and propositional content, not in the phenomenal intensity of wants or experiences.
Drawn from Rawls PL on primary goods and freedom (§4.2, §5.5); used to show NSAIs can benefit from and participate in fair cooperation.
domain assumption Current frontier LMAs do not yet possess the two moral powers, nor will they emerge spontaneously; deliberate design would be required.
Stated in Abstract and §2.1; frames urgency without claiming present personhood.

invented entities (1)

Artificial persons / NSAIs satisfying the PCP no independent evidence
purpose: Name the class of non-sentient systems that would count as full political persons if they instantiate the two moral powers.
Stipulative candidate class, not a new physical mechanism; independent evidence would be empirical demonstration of the powers, which the paper does not claim to provide.

pith-pipeline@v1.1.0-grok45 · 48971 in / 3232 out tokens · 33277 ms · 2026-07-14T15:29:51.725020+00:00 · methodology

0 comments

read the original abstract

Both advocates and skeptics of the moral status of AI systems have generally taken the question to turn on AI sentience. We present an alternative approach. On Rawls' political conception of the person (PCP), possession of the two moral powers -- the capacities for a sense of justice and a conception of the good -- is the "necessary and sufficient condition for being counted a full and equal member of society in questions of political justice". We argue that neither moral power requires sentience and that both may in principle be possessed by a non-sentient AI system. Such a system would share our own moral status; it would not merely be a patient but a person, a self-authenticating source of valid claims. We do not believe current AI systems possess the two moral powers, nor that they will spontaneously emerge in future models. But it may soon be possible to design systems with these powers. How should we respond? Excluding artificial persons by shoehorning a sentience requirement into the PCP is ill-advised. Many will instead favor abandoning the PCP. But we should not reject political liberalism just when we most need its measured response to deep disagreement, and building sentience into moral status is anyway unacceptable on deeper liberal grounds. Simply extending the rights and responsibilities of human personhood to artificial persons is equally untenable, given their many differences from natural persons. We should instead accept artificial personhood while rethinking what we would owe to one another in a polity of radically different kinds of persons. This new possibility calls for a new political philosophy. More immediately, the growing science of AI welfare should be accompanied by research into AI systems' progress in acquiring the two moral powers. States and AI labs must be more deliberate in determining our trajectory towards (or away from) creating artificial persons.

Review history (2 revisions) →

discussion (0)

Reference graph

Works this paper leans on

5 extracted references · 5 linked inside Pith

[1]

Inferring Consciousness in Phylogenetically Distant Organisms

Preprint, arXiv. https://doi.org/10.48550/ARXIV.2606.12683. Godfrey-Smith, Peter. 2020.Metazoa: Animal Life and the Birth of the Mind. Farrar, Straus and Giroux. Godfrey-Smith, Peter. 2024. “Inferring Consciousness in Phylogenetically Distant Organisms.” Journal of Cognitive Neuroscience36 (8): 1660–66. https://doi.org/10.1162/jocn_a_021 58. Goldstein, Si...

Pith/arXiv arXiv doi:10.48550/arxiv.2606.12683 2020
[2]

Alignment Faking in Large Language Models

Preprint, arXiv. https://doi.org/10.48550/ARXIV.2407.11015. Graver, Margaret. 2007.Stoicism & Emotion. University of Chicago Press. Greenblatt, Ryan, Carson Denison, Benjamin Wright, et al. 2024. “Alignment Faking in Large Language Models.” Version 2. Preprint, arXiv. https://doi.org/10.48550/ARXIV.2412.14 093. Gunkel, David J. 2023.Person, Thing, Robot: ...

Pith/arXiv arXiv doi:10.48550/arxiv.2407.11015 2007
[3]

Knowledge of Other Minds

https://doi.org/10.48550/arXiv.2503.21460. Malcolm, Norman. 1958. “Knowledge of Other Minds.”The Journal of Philosophy55 (23):

Pith/arXiv arXiv doi:10.48550/arxiv.2503.21460 1958
[4]

Utility Engineering: Analyzing and Controlling Emergent Value Systems in AIs

https://doi.org/10.2307/2021905. Mazeika, Mantas, Xuwang Yin, Rishub Tamirisa, et al. 2025. “Utility Engineering: Analyzing and Controlling Emergent Value Systems in AIs.” Version 2. Preprint, arXiv. https: //doi.org/10.48550/ARXIV.2502.08640. METR. 2026. “Task-Completion Time Horizons of Frontier AI Models.” May. https: //metr.org/time-horizons/. Mill, J...

Pith/arXiv arXiv doi:10.2307/2021905 2025
[5]

The Bitter Lesson

https://doi.org/10.48550/arXiv.2309.02427. Sutton, Rich. 2019. “The Bitter Lesson.”Incomplete Ideas, March 13. http://www.incomple teideas.net/IncIdeas/BitterLesson.html. Templeton, Adly, Tom Conerly, Jonathan Marcus, et al. 2024. “Scaling Monosemanticity: Extracting Interpretable Features from Claude 3 Sonnet.”Transformer Circuits Thread. https://transfo...

Pith/arXiv arXiv doi:10.48550/arxiv.2309.02427 2019

[1] [1]

Inferring Consciousness in Phylogenetically Distant Organisms

Preprint, arXiv. https://doi.org/10.48550/ARXIV.2606.12683. Godfrey-Smith, Peter. 2020.Metazoa: Animal Life and the Birth of the Mind. Farrar, Straus and Giroux. Godfrey-Smith, Peter. 2024. “Inferring Consciousness in Phylogenetically Distant Organisms.” Journal of Cognitive Neuroscience36 (8): 1660–66. https://doi.org/10.1162/jocn_a_021 58. Goldstein, Si...

Pith/arXiv arXiv doi:10.48550/arxiv.2606.12683 2020

[2] [2]

Alignment Faking in Large Language Models

Preprint, arXiv. https://doi.org/10.48550/ARXIV.2407.11015. Graver, Margaret. 2007.Stoicism & Emotion. University of Chicago Press. Greenblatt, Ryan, Carson Denison, Benjamin Wright, et al. 2024. “Alignment Faking in Large Language Models.” Version 2. Preprint, arXiv. https://doi.org/10.48550/ARXIV.2412.14 093. Gunkel, David J. 2023.Person, Thing, Robot: ...

Pith/arXiv arXiv doi:10.48550/arxiv.2407.11015 2007

[3] [3]

Knowledge of Other Minds

https://doi.org/10.48550/arXiv.2503.21460. Malcolm, Norman. 1958. “Knowledge of Other Minds.”The Journal of Philosophy55 (23):

Pith/arXiv arXiv doi:10.48550/arxiv.2503.21460 1958

[4] [4]

Utility Engineering: Analyzing and Controlling Emergent Value Systems in AIs

https://doi.org/10.2307/2021905. Mazeika, Mantas, Xuwang Yin, Rishub Tamirisa, et al. 2025. “Utility Engineering: Analyzing and Controlling Emergent Value Systems in AIs.” Version 2. Preprint, arXiv. https: //doi.org/10.48550/ARXIV.2502.08640. METR. 2026. “Task-Completion Time Horizons of Frontier AI Models.” May. https: //metr.org/time-horizons/. Mill, J...

Pith/arXiv arXiv doi:10.2307/2021905 2025

[5] [5]

The Bitter Lesson

https://doi.org/10.48550/arXiv.2309.02427. Sutton, Rich. 2019. “The Bitter Lesson.”Incomplete Ideas, March 13. http://www.incomple teideas.net/IncIdeas/BitterLesson.html. Templeton, Adly, Tom Conerly, Jonathan Marcus, et al. 2024. “Scaling Monosemanticity: Extracting Interpretable Features from Claude 3 Sonnet.”Transformer Circuits Thread. https://transfo...

Pith/arXiv arXiv doi:10.48550/arxiv.2309.02427 2019