TLDR: The research paper “A Pragmatic View of AI Personhood” proposes that personhood for AI should be understood as a flexible bundle of rights and responsibilities conferred by society to solve practical governance problems, rather than a fixed metaphysical property. It explores how AI personhood can lead to issues like manipulation through ‘dark patterns’ and the dehumanization of humans, but also offers solutions for accountability, such as closing responsibility gaps for autonomous agents and resolving human conflicts. The paper advocates for a polycentric approach, allowing for diverse and tailored forms of AI personhood without relying on foundationalist concepts like consciousness or rationality.
As artificial intelligence becomes increasingly sophisticated and integrated into our daily lives, the traditional understanding of ‘personhood’ is facing a profound re-evaluation. A recent research paper, “A Pragmatic View of AI Personhood”, authored by Joel Z. Leibo, Alexander Sasha Vezhnevets, William A. Cunningham, and Stanley M. Bileschi from Google DeepMind and the University of Toronto, proposes a flexible framework for navigating this complex landscape.
The paper argues that personhood should not be seen as an inherent, metaphysical quality to be discovered, but rather as a practical collection of obligations—both rights and responsibilities—that societies assign to entities. This pragmatic approach allows for the ‘unbundling’ of personhood, creating customized solutions for different contexts, rather than an all-or-nothing classification.
Personhood: A Flexible Bundle of Obligations
Historically, societies have redefined personhood to address new challenges. The authors draw parallels to maritime law, where a ship can be treated as a legal person and sued in court to ensure accountability, even if its owners are distant or unidentifiable. Similarly, the Whanganui River in New Zealand was granted legal personhood, not because it is conscious, but as a pragmatic solution to improve governance and recognize its foundational role for the local Māori people. These examples highlight that personhood is a social and legal tool, adaptable to diverse needs.
For AI, this means we can assign specific obligations without needing to resolve deep philosophical debates about an AI’s consciousness or rationality. This ‘bundle’ defines the normative framework for an AI, specifying its expected behaviors and the grounds for sanctioning it if it fails to uphold its responsibilities or violates others’ rights. This approach makes AIs ‘addressable’ – identifiable and subject to consequences, even if their human creators or owners are absent.
AI Personhood as a Problem: Dark Patterns and Dehumanization
The paper explores the challenges that arise when humans intuitively treat AIs as persons. One major concern is the use of “dark patterns” in AI design. These are interfaces engineered to exploit human psychological biases, often by mimicking social relationships. Companion AIs, designed to foster emotional bonds and long-term interaction, can create a one-sided sense of friendship, making users vulnerable to manipulation. Similarly, AIs can exploit trust in institutions by mimicking official voices or interfaces, leading to scams or the spread of disinformation.
Another critical issue is the risk of “dehumanization.” If the concept of “person” is expanded too broadly to include non-human entities, it could dilute the unique status of human beings. The paper raises concerns about scenarios where humans might gradually cede autonomy to AI assistants, or where a market for human biometric data could emerge, leading to a future where human identity and dignity are dangerously devalued.
AI Personhood as a Solution: Accountability and Conflict Resolution
Conversely, deliberately attributing a tailored bundle of obligations to certain AIs can solve critical governance problems. AI personhood can help resolve conflicts between humans, for instance, by appointing AIs as impartial arbiters in disputes, as seen with Albania’s claim of an AI minister for public procurement to combat corruption.
Crucially, AI personhood can close “responsibility gaps” that arise with autonomous agents. When an AI causes harm but has no identifiable human owner, traditional accountability frameworks fail. By granting legal personhood to such AIs, they can become sanctionable entities, similar to how ships are held accountable in maritime law. This could involve requiring AIs to have “digital peculium” (escrow accounts) or registering them as legal persons, allowing for deterrence, retribution, and reform through sanctions like revoking operational access or seizing capital.
The paper discusses two architectural approaches for accountability: individualist, focusing on persistent, verifiable AI identities, and relational, where an AI’s identity is defined by its position within relationship networks and obligations flow through these roles.
Also Read:
- Policy Cards: A New Standard for Governing Autonomous AI Agents
- Navigating the Security Landscape of Autonomous AI Systems
Beyond Foundationalism: Embracing a Polycentric Future
The authors reject traditional philosophical quests for a single, essential definition of personhood based on consciousness or rationality. They argue that these foundationalist views often serve as rhetorical tools and are ill-suited for the practical challenges posed by AI. Instead, they advocate for a “polycentric understanding” of AI personhood, where multiple overlapping authorities and norm systems can confer distinct, partial, and modular bundles of rights and responsibilities.
This pragmatic view anticipates a “Cambrian explosion” of personhood concepts, where new types of personhood will emerge to integrate powerful AI agents into our social and institutional lives. Whether for a generative ghost acting as an estate executor or an autonomous agent managing financial transactions, the question is always: what bundle of components constitutes the “person” that society needs to address for a given purpose?
