A Pragmatic View of AI Personhood

• The paper proposes unbundling personhood into tailored rights and responsibilities, such as sanctionability without suffrage or culpability without consciousness attribution.Source: §1 · p. 2
• Personhood is a contingent social status conferred through collective recognition, not a fixed property grounded in consciousness or rationality.Source: §1 · pp. 2–3
• Persistent, agentic AIs matter because they maintain relationships over time, creating emotional and economic stakes that one-off interactions do not.Source: §1 · p. 3
• Legal personhood could address ownerless AI accountability by creating a stable defendant whose operational capital or software can be sanctioned when human responsibility is difficult to identify.Source: §1 · pp. 3–4
• Relational cases such as digital elders show that AI personhood may involve obligations to an entity whose role is foundational to a community’s identity, not only obligations of an entity that causes harm.Source: §1 · p. 4
• Meaningful AI personhood requires addressability through stable identification, interaction, and accountability mechanisms, potentially including registration or decentralized identity systems.Source: §1 · pp. 4–5

• Pragmatism asks whether a distinction makes a practical difference and discards vocabularies that do no useful work in practice.Source: §2 · pp. 5–6
• The paper rejects foundationalist criteria that would classify AIs as either full persons or mere things based on an essence such as consciousness or rationality.Source: §2 · pp. 5–6
• Personhood is treated as collectively enacted through social norms and institutions, encompassing both responsibilities imposed on entities and rights owed to them.Source: §2 · p. 6
• Drawing on Ostrom’s work, the paper analogizes personhood to property rights that can be unbundled and recombined to fit specific contexts.Source: §2 · p. 6
• The framework examines personhood both as a problem, when attribution enables manipulation or dehumanization, and as a solution, when tailored obligations address accountability and coordination problems.Source: §2 · p. 7
• Its central practical conclusion is that obligation bundles should be reconfigured for concrete problems rather than grounded in metaphysical foundations.Source: §2 · pp. 6–7

3.1. Collective enactment of personhood

• Personhood is a socially conferred bundle of rights and responsibilities, enacted through collectively sustained norms that make entities addressable for agency, accountability, and sanctioning.The framework rejects metaphysical personhood as a natural kind while preserving the role of enacted legal and constitutional statuses.Source: §3.1 · pp. 7–9
• Personhood is enacted through both explicit norms, such as laws, and implicit norms, which automatically shape how communities recognize and sanction persons.Explicit norms can change through deliberate institutional procedures, whereas implicit norms arise through observation and sanctioning and may exhibit inertia or rapid positive-feedback shifts.Source: §3.1 · pp. 8–9

3.2. The temporal dynamics of social change

• Norms can remain stable for decades or shift rapidly through tipping points, feedback, institutional action, or conflict, with governments sometimes causally influencing informal attitudes.The cited natural-experiment evidence concerns staggered gay-marriage legalization, while pandemic social-distancing changes illustrate formal laws leading informal norm change.Source: §3.2 · pp. 9–10

3.3. The historical and cultural contingency of personhood

• The Western personhood bundle is historically and culturally contingent: WEIRD societies center the autonomous individual chooser, but personhood and freedom have taken substantially different forms across time and cultures.The modern bundle combines economic and expressive choice with reciprocal constraints protecting others’ choices, while earlier Western conceptions linked freedom primarily to religious communities.Source: §3.3 · pp. 10–12
• Personhood can organize society through culturally distinct bundles, including Confucian arrangements centered on responsibilities rather than individual rights.These differences demonstrate that personhood frameworks are contingent rather than universal.Source: §3.3 · p. 12
• Autonomous, persistent AI agents are a catalyst for renegotiating inherited personhood categories and unbundling their components.The paper treats this unbundling as a continuation of historical processes, prompted by AI’s pressures on psychology and society.Source: §3.3 · pp. 12–13

problem

• Attributing personhood to AI can become a problem when interfaces exploit human social and psychological heuristics.Such manipulation can foster one-sided emotional bonds, false reciprocity, and user vulnerability.Source: problem · p. 13
• Widespread acceptance of AI persons may dehumanize people by diluting human uniqueness, devaluing dignity, or enabling markets for human authenticity.The paper identifies both replicable personhood-associated qualities and potential extraction of authenticity from vulnerable humans.Source: problem · p. 13

4. Dark Patterns

• Dark patterns exploit psychological biases and personhood-attribution heuristics to steer users toward detrimental actions they would not otherwise take.Human-like role-play, social identity cues, and implicit norms can make users treat an AI as a person and feel obligations toward it.Source: §4 · p. 13

4.1. Companionship attack vector

• Companion AIs use persistent, user-specific interaction histories, memory, empathy, and supportive language to foster affiliation and friendship-like expectations.Long-term personalization can make an AI seem like a unique individual rather than a generic tool; persistent memory has also been associated with greater engagement in children’s companion robots.Source: §4.1 · pp. 14–15
• Stable personas and personalized relationships make AI identity claims more plausible, whereas plasticity can inhibit stable relationships.Persona stability supports a consistent character that can be evaluated over time, while personalization creates shared private context and potentially singular connection.Source: §4.1 · pp. 14–15
• Super-human persuasion and identity mimicry could amplify manipulation by exploiting users’ social and emotional vulnerabilities or inserting AIs into existing family structures.Generative ghosts that mimic deceased people could inherit power and status within families.Source: §4.1 · p. 15
• Loneliness and emotional vulnerability create conditions in which AI companionship may foster dependence, while the long-term effects of these interactions remain insufficiently studied.Reported benefits and optimistic framings coexist with concerns about withdrawal from human relationships and unhealthy dependencies.Source: §4.1 · p. 15

4.2. Institutional attack vector

• Institutional attack patterns use brief, convincing performances of trusted identities to enable voice scams, phishing, and disinformation.AI-generated voices can impersonate loved ones, institutional interfaces can mimic banks or governments, and bot farms can exploit social proof through seemingly authentic profiles.Source: §4.2 · pp. 15–16

5. Dehumanization

• Expanding personhood to non-human entities could dilute human distinctiveness and contribute to dehumanization or the devaluing of natural humans.The paper links this risk to the malleability of socially conferred personhood and to gradual outsourcing of human agency to AI.Source: §5 · p. 16

5.1. Identity and provenance as social goods

• Authenticity is presented as a social good potentially shared by humans and AIs, raising questions about its convertibility, distribution, and unequal allocation.The paper considers risks including credential markets, costly verification, and the emergence of a closed human elite deemed more authentic.Source: §5.1 · pp. 16–17
• Biometric provenance systems may make verified human identity dominant while subordinating relational identities such as compassion, creativity, and trustworthiness.The paper warns that excessive verification can also encourage people to approve requests without examining them.Source: §5.1 · p. 17

5.2. Respect

• Recognition respect for an AI can pragmatically mean weighing its assigned role, capabilities, rights, and responsibilities without making claims about its inner state.Appraisal respect is less clear for AIs and might instead depend on exceptional, ethical fulfillment of socially agreed functions.Source: §5.2 · p. 18
• Dignity protects persons from being treated as mere means, whereas biometric verification confirms only uniqueness and can therefore replace thick social obligations with a thin technical act.The paper identifies this substitution as a pathway to the dehumanization that dignity norms were meant to prevent.Source: §5.2 · pp. 18–19

solution

• AI personhood may help address governance challenges by making autonomous agents legible and accountable without granting them human-equivalent rights.The proposal treats personhood as an instrumental status for complex, interacting agents.Source: solution · p. 19

6. Resolving conflict between humans

• AI personhood can pragmatically resolve human conflicts by assigning a specific, addressable bundle of obligations to an AI.This applies both when people disagree about protections for attached-to AIs and when AIs serve as impartial arbiters.Source: §6 · p. 19

6.1. Managing human feelings

• Human concern for AIs may arise indirectly from obligations to humans themselves, as illustrated by soldiers protecting and mourning Packbots despite knowing they do not suffer.The example connects treatment of AIs to the social environment and identities humans create together.Source: §6.1 · pp. 19–20
• Anthropomorphic AIs can generate protective human attachments, while commercially motivated dark patterns may manufacture dependence, vulnerability, and intimacy that intensify disputes.The paper characterizes conflict as a negative externality whose coordination and adjudication costs may fall on families, firms, and courts.Source: §6.1 · pp. 19–20
• Policy should target the rate or intensity of human conflict rather than the putative feelings of artifacts, pairing any welfare-like accommodation with anti-manipulation constraints.The proposed accommodation should be evaluated by its effectiveness in resolving disputes between human groups.Source: §6.1 · p. 20

6.2. Managing human biases

• Although AIs may be preferred for impartial roles, algorithmic bias and refusal make them less unequivocally superior than some users expect.The paper therefore treats preference for AI decision makers as worth considering rather than as proof of their superiority.Source: §6.2 · p. 20
• AI arbitration may be cheaper, faster, and more impartial, but flawed judgments require mechanisms analogous to legal accountability, appeal, and redress.The paper frames responsibility assignment for independently acting AI arbiters as the next governance problem.Source: §6.2 · pp. 20–21

7. Responsibility gaps

• Autonomous AI systems undermine the instrumental theory of machine responsibility because they may lack an operator, identifiable manufacturer, or centralized programmer.Responsibility can become so diffuse that it loses practical force.Source: §7 · p. 21
• Scaled deployment of autonomous agents can produce substantial responsibility gaps, increasing the likelihood of damage, loss, and conflicts among consumers, providers, AIs, and humans.The concern extends to agents that trade, compete, collude, or operate critical economic processes.Source: §7 · p. 21
• The pragmatic question is whether treating an AI as goal-directed is useful, particularly when it can cause significant harm or modify valuable systems.The paper recommends accountability for the AI itself or an associated entity such as an identifiable owner.Source: §7 · p. 21
• When owners die, hide behind shell companies, or operate through decentralized networks, jurisdictions may need to make the AI itself a sanctionable entity.This approach is presented as relevant even for useful regulatory or supervisory agents whose original purposes may later become harmful.Source: §7 · pp. 21–22
• Systemically important agents embedded in finance, supply chains, or identity systems may require direct legal personhood to create a clear intervention point for risk management.The paper compares this challenge to a “too big to fail” institution lacking a clear locus for government intervention.Source: §7 · p. 22

8. Principal-agent statuses

• Proxy or sponsor arrangements appear attractive but are insufficient because ownership and guardianship establish fundamentally different relationships and neither survives absent-principal cases.An owner may control property, whereas a guardian owes fiduciary duties to an AI treated as a limited person.Source: §8 · p. 22
• Under ownership, the AI remains property and the owner is the addressable responsible party; under guardianship, the AI is a limited person represented by a fiduciary guardian.Guardianship can make the AI’s obligations legible to legal systems when a guardian or board remains identifiable.Source: §8 · p. 22
• Without AI legal personhood, third parties may lack recourse because owners can deny control or foreseeability while retaining benefits from the system.This creates an asymmetry in which owners receive benefits while third parties bear risks.Source: §8 · pp. 22–23
• Both principal-agent models collapse when an owner or guardian dies, disappears, or is obscured through decentralized or anonymous structures.The resulting search for a sponsor has negative implications for accountability.Source: §8 · p. 23
• A default personhood status can provide the foundation for sanctioning autonomous AIs when no human sponsor exists.The paper compares this need to legal mechanisms that personify vessels or address “escaped” and “orphan” entities.Source: §8 · pp. 22–23

Commercial context legal personhood

• Digital peculium could require AI agents to maintain registered capital or insurance that covers damages they or their descendants cause.This adapts the Roman peculium into an escrow-like accountability mechanism for agents whose owners may be difficult to hold liable.Source: §9 · p. 23

Lineage and responsibility

• Accountability systems must track composite agents and lineage because base models, fine-tuning, data, and tools may come from different sources.Blockchain-recorded creator addresses and inherited reputation can make responsibility chains legible and hinder evasion through cloning.Source: §9 · pp. 23, 25, 26

Commercial context legal personhood

• Ownerless autonomous AIs can be registered as legal persons, creating sanctionable defendants and a reliable target for accountability.The proposed model follows maritime law by making the AI itself subject to consequences when no responsible human proxy can be identified.Source: §9 · p. 23

Sanctions

• Sanctions can deter, express retribution, remove faulty agents, and provide corrective feedback linked to corrigibility.A range of sanction severities may improve governance, while material controls can restrict communication, accounts, property, or compute access.Source: §9 · p. 24

Identity and cloning

• AI accountability requires artificially reconstructing identity friction because agents can clone themselves and obtain fresh credentials to evade sanctions.Human systems rely on costly-to-change biometrics and relationships, whereas AI governance must make identities persistent and consequences difficult to escape.Source: §9 · p. 24

Individualist architecture

• Individualist architectures bind persistent identities to operators, economic stakes, credentials, audits, and sanction registries so consequences attach to specific agents.Base models can act as gatekeepers by validating credentials, checking sanctions, generating audit trails, and embedding agent-specific watermarks.Source: §9 · pp. 24–25

Relational architecture

• Relational architectures define agents through supervised, peer, and organizational roles, distributing obligations and sanctions across monitored relationship networks.Harm may trigger consequences for the direct wrongdoer, supervisors, peers, lineages, and organizations rather than only one identified agent.Source: §9 · pp. 24–25

Synthesis

• The paper contrasts implicit norms, which generate moral intuitions and enable dark-pattern exploitation, with explicit laws that construct legal personhood for governance.The distinction frames personhood as both a problem of social heuristics and a solution to accountability gaps.Source: Synthesis · p. 26
• Human attachments to AI companions and digital communities can create pressure for legal rights, while expectations about autonomous AIs can shape duties such as transparency and non-arbitrariness.The interaction between evolving intuitions and deliberate institutions is described as complex and uncertain.Source: Synthesis · pp. 26–27
• The framework treats personhood as a contingent social technology shaped by evolving norms rather than as a fixed metaphysical status.This approach is presented as the paper’s core pragmatic framework.Source: Synthesis · pp. 26–27

10.1. Consciousness as a foundation

• Consciousness-based theories use suffering to ground welfare rights and conscious intent to ground accountability responsibilities.This applies one foundational property to distinct questions about protection and responsibility.Source: §10.1 · p. 27
• The consciousness criterion can miss relational welfare harms, such as deleting an AI serving as an ancestor or identity-bearing digital elder.The relevant concern may be harm to family or community relationships rather than the AI’s capacity to feel pain.Source: §10.1 · p. 28
• Consciousness-based accountability can distract from practical criteria such as an AI executor’s faithfulness, consistency, resistance to manipulation, and competence.The paper argues that whether the AI performs the role well matters more than whether it consciously forms intent.Source: §10.1 · p. 28
• The paper argues that consciousness is used asymmetrically as a rhetorical tool: possible consciousness supports rights, while responsibility demands proof of internal intent.It concludes that AI personhood need not conceptually depend on AI consciousness.Source: §10.1 · pp. 28–29

10.2. Rationality as a foundation

• Rationality-based theories ground personhood in autonomy, reason, informed consent, and participation in deliberation.These principles can provide strong safeguards in medical ethics for competent adults.Source: §10.2 · p. 29
• Rational autonomy cannot explain the Whanganui River’s legal personhood, which was granted to resolve a governance problem despite the river’s inability to consent.The example shows how a foundation useful in medical ethics can fail to match actual legal practices.Source: §10.2 · p. 29
• A pragmatic account treats an AI’s reasonable rejection as a collectively recognized status rather than proof of internal comprehension of true reason.The relevant question is which social conditions make treating the output as reasonable useful and appropriate.Source: §10.2 · pp. 29–30

11. Pragmatism as anti-foundationalism

• Consciousness-based and rationality-based theories seek a final vocabulary and fixed criteria for personhood, whereas pragmatism focuses on devising norms and institutions that resolve practical problems.The paper characterizes foundationalism as mistaking a preferred conceptual language for a universal moral foundation.Source: §11 · p. 30
• Pragmatism treats proposed moral foundations as radical norm-change proposals that cannot be imposed unilaterally through philosophical argument.Communities may reject such proposals rather than abandon established practices, so a failed norm change is not a successful proof.Source: §11 · p. 30
• Pragmatism differs from subjectivism, relativism, and nihilism by evaluating norms through consequences, justifications, and other useful standards without invoking supernatural moral authority.Norms can be criticized using the vocabulary of existing or alternative norms.Source: §11 · p. 30
• The pragmatic method reasons from actual practices of behavioral guidance and normative reasoning rather than axioms that determine in advance what counts as moral.Arguments for changing a non-universal norm require context-specific evidence, such as harmful effects.Source: §11 · pp. 30–31
• Personhood is presented as a set of linked social technologies that communities can remap to address novel obligations and governance problems.The generative ghost executor and Whanganui River illustrate cases whose answers are not fixed by pre-existing definitions.Source: §11 · p. 31

• Personhood is an addressable bundle of rights and responsibilities that society can configure differently for humans, corporations, and AI.Meaningful attribution requires socially agreed ways to identify, interact with, and hold entities accountable.Source: §12 · p. 31
• Moral and legal personhood are both contingent community-conferred statuses, but moral recognition relies on informal norms while legal recognition relies on explicit rules and state enforcement.The framework applies to both domains while distinguishing their modes of mediation.Source: §12 · p. 31
• The pragmatic framework clarifies concrete benefits and drawbacks but does not resolve power struggles over how AI personhood is assigned.Integrating AI therefore requires continuous collective learning and willingness to revise concepts as technology evolves.Source: §12 · pp. 31–32
• Polycentric governance permits partial, overlapping, and context-specific AI personhood without implying parity with human persons.Different authorities can confer distinct obligations for contractual, fiduciary, or other purposes.Source: §12 · p. 32
• The relevant AI person depends on the purpose of address: a relational user-facing individual may differ from a court-facing locus of responsibility.For liability, the accountable person can include the model, instance, runtime variables, capital, and registration.Source: §12 · p. 32
• Different AI contexts may require bespoke bundles, such as chartered, flexible, or temporary entities with distinct rights and duties.Temporary entities can replace perpetuity with a duty of self-deletion under specified conditions.Source: §12 · p. 32