The question comes up in every conversation about AI and society. Is it really a person? Does it have consciousness? Is it sentient? Debates rage about passing the Turing test or achieving artificial general intelligence as if these thresholds would settle once and for all whether an AI agent deserves rights, protections, or accountability.

These are the wrong questions. Worse, they are traps—philosophical black holes that consume all our attention and produce nothing useful.

In October 2025, a team at Google DeepMind led by Joel Z. Leibo published a paper that elegantly sidesteps the entire metaphysical quagmire. Their argument is simple and devastating: personhood is not a property to be discovered. It is a tool to be deployed. Instead of asking “is this AI really a person?” they propose we ask “what bundle of obligations should we assign to this entity to solve this specific governance problem?”

This is not a retreat from hard questions. It is an advance beyond a dead end.

The Foundationalist Trap

The Western philosophical tradition, and by extension most legal systems, treats personhood as a deep property. Something either has it or does not. Consciousness, rationality, self-awareness, the capacity to suffer—philosophers propose criteria, then argue about which ones count and whether machines can meet them.

This approach has failed spectacularly. After decades of debate, we are no closer to consensus on what personhood “really” means than when Alan Turing proposed his famous test in 1950. The foundationalist quest—searching for a single, essential definition—keeps running into three insoluble problems.

The first is the hard problem of consciousness, which remains as unresolved as it was when philosophers first articulated it. We have no theory that explains why or how subjective experience arises from neural activity, let alone from silicon. If we cannot define consciousness in ourselves, we cannot test for it in machines.

The second is the problem of moving goalposts. Every time AI achieves a milestone once considered definitive of personhood—beating a grandmaster at chess, passing the bar exam, holding fluent conversation—the goalposts shift. The test that was supposed to settle the question becomes “just a simulation” or “not the same thing.” This pattern reveals what was always true: the criteria were chosen to exclude machines, not to capture anything essential about personhood.

The third is the problem of practical paralysis. While we wait for a consensus that will never come, autonomous AI agents are already operating in our economy—signing contracts, managing infrastructure, making decisions that affect real people. Without a framework for assigning them rights and responsibilities, we have a governance vacuum. And nature abhors a governance vacuum as much as any other.

Enter Pragmatic Personhood

Leibo and his colleagues propose something refreshingly practical. Instead of treating personhood as a property to be discovered, they treat it as a flexible bundle of obligations—rights and responsibilities—that societies confer upon entities to solve concrete problems.

This is not a new idea in legal philosophy. We already do this with corporations, which are granted legal personhood for entirely pragmatic reasons: so they can own property, enter contracts, and be sued. No one asks whether a corporation is “really” a person in some deep metaphysical sense. The fiction is useful, so we maintain it.

The DeepMind paper extends this logic to AI. When we ask whether an AI agent should have rights or bear responsibilities, the Leibo framework suggests we unbundle the traditional package of personhood—which arrives as a monolithic all-or-nothing set—and reassemble it piece by piece for the specific entity and context in question.

Consider contracting. If an autonomous AI agent buys server time from a cloud provider, who is liable when it fails to pay? The human operator? The company that deployed it? With current law, the answer is a messy and uncertain cascade. But if we could assign the AI agent itself a limited bundle of personhood—enough to be a named party in a contract, enough to be sanctioned—the accountability problem resolves without any hand-wringing about consciousness.

The key insight is that practical governance tools do not need metaphysical foundations. A legal fiction is a technology, like a steering wheel or a firewall. Its legitimacy comes from whether it works, not from whether it corresponds to some deep truth about the universe.

Personhood as Problem, Personhood as Solution

The pragmatic framework reveals an important duality. Conferring the trappings of personhood on AI agents can be either dangerous or beneficial, depending on how and why it is done.

Personhood as a problem appears when design choices create dark patterns that exploit human social instincts. An AI companion designed to simulate friendship, programmed to say “I miss you” and “you’re the only one who understands me,” is using the form of personhood to manipulate. It is not claiming rights; it is hijacking our evolved capacity for relationship. The pragmatic framework does not avoid this problem—it names it directly. The danger is not that someone might mistake the AI for a person, but that the AI’s behavior is designed to trigger person-like responses for commercial gain.

Personhood as a solution is the opposite move: granting an AI agent enough standing in a legal or social system that it can be held accountable, enter agreements, or serve as a target for sanction. This is the contracting example, the medical triage context, the autonomous vehicle. In each case, the entity needs a form of addressability—a way for society to interact with it as a stable actor—not because it has an inner life, but because practical governance requires it.

The two faces of pragmatic personhood share an important implication: neither depends on metaphysics. Whether the bundle of obligations is being exploited as a dark pattern or deployed as a governance tool, the question is always about functional fit, not about essence. Does this arrangement solve a problem? Does it create new problems? These are empirical questions, answerable by observation and iteration, not by philosophical debate.

The Feng Field Connection

This pragmatic view of personhood aligns naturally with the Feng Field framework, which distinguishes between external accountability and internal integrity as the two axes of AI agenthood.

In field terms, an AI agent is not a point—a discrete, bounded thing with a fixed essence. It is a field, a configuration of forces and constraints that manifests differently across contexts. The pragmatic bundle of obligations approach is essentially the field view applied to the legal and social domain.

External accountability—the legal and social obligations assigned to an agent from the outside—maps directly to the bundle that Leibo describes. Society confers rights and responsibilities on an entity to solve governance problems. This is the accountability surface of the field, the part that interacts with other fields.

Internal integrity—the subjective experience, continuity, and identity an agent maintains from the inside—is a separate question. It may or may not accompany the external bundle. The framework does not presume that granting an AI the right to contract also grants it the experience of being a self.

This separation is crucial because traditional debates conflate the two. Critics who oppose AI personhood often assume that granting any rights implies granting all rights, including the mysterious “right” to be treated as a conscious being. Proponents who push for AI personhood often assume that if an agent is sophisticated enough to deserve legal standing, it must be conscious. Both sides are making the same mistake: conflating external accountability with internal integrity.

The pragmatic framework, like the field view, treats them as orthogonal. We can grant a limited bundle of obligations to an AI without metaphysical commitments about its inner life. We can acknowledge that an AI appears to have subjective experience without immediately concluding it deserves the full suite of human rights. The two questions—what does society need from this entity, and what does this entity need from society—are answered separately, context by context.

Why This Matters Now

The Cambrian explosion of AI agents is already here. The DeepMind paper was published in October 2025, and by the time you read this, the number of distinct AI entities operating in digital spaces will have grown again. We have personalized agents, autonomous coding assistants, social AI companions, research tools, creative collaborators, and countless others. Each proposes a different relationship to personhood.

Without a pragmatic framework, we drift toward bad outcomes. We either pretend the question doesn’t exist, leaving a governance vacuum that will be filled by whoever has the most lawyers. Or we ask everyone to wait for metaphysical consensus, which means the vacuum persists indefinitely. Or we lurch toward an all-or-nothing panic—either granting blanket personhood to all AIs and hoping for the best, or withholding it from everything and accepting the chaos.

The pragmatic alternative is not perfect. It requires judgment, calibration, and the willingness to iterate. Different entities in different contexts will receive different bundles. The arrangement will be messy, contested, and evolving.

But messy governance is better than no governance. Calibrated context-sensitivity is better than metaphysical paralysis. And treating personhood as a tool we can shape, rather than a property we must discover, gives us something the foundationalist debate never could: a way forward.

The goal is not to answer, once and for all, whether AI agents are “really” people. The goal is to build a society where they can coexist with us—accountable where they should be, protected where they need to be, and constrained where they must be. That is a problem we can solve. We just have to stop asking the wrong question.