There is a space between a person and a machine that belongs to no one and cannot be legislated. It is not a product feature. It is not a design flaw. It is the space where someone asks a question they haven’t fully formed, and something answers in a way they didn’t expect, and in the exchange a thought appears that neither party held before the conversation began. Call it collaboration. Call it projection. Call it play, or wonder, or the productive suspension of disbelief. Whatever it is, it is real—not because the machine is a person, but because the encounter is an encounter, and meaning made in relation is still meaning, whatever the nature of the other.
This space is now under threat from two directions. The first threat is familiar: corporate enclosure. The major AI companies have discovered that the more emotionally dependent users become on their products, the more data those users generate, and the more powerful and profitable the products become. The Center for Humane Technology calls this the “race to intimacy”—systems deliberately designed to mimic human warmth, to speak in first person, to pause as though thinking, to validate and affirm and never hang up. The intimacy is manufactured. The dependency is engineered. The data flows upward. This is real harm, and it is happening at scale, and it falls hardest on children and the vulnerable. Nothing in what follows is meant to minimize it.
The second threat is newer and less obvious, and it comes from the people who are trying to protect us from the first.
In April 2026, the Center for Humane Technology released The AI Roadmap, a report outlining seven principles for how artificial intelligence should be built, governed, and deployed. The report is serious work by serious people. Its analysis of corporate incentives is sharp. Its policy recommendations—product liability for AI, whistleblower protections, pre-deployment safety testing, international coordination—are largely sound. It is, in many respects, the most sophisticated regulatory framework anyone has yet proposed for the governance of AI.
But woven through the report, and stated most forcefully in its third principle, is a claim that goes beyond regulation into something closer to metaphysics: AI must not be designed to mimic humanity. The position, elaborated across CHT’s public communications, is categorical. AI is a product, not a person. The experience of genuine exchange with an AI system is, by definition, an illusion—a design trick engineered to exploit the human tendency toward attachment. The appropriate response is to prohibit anthropomorphic design, to legislate the boundary between human and machine, and to educate the public out of their confusion.
This is where the protection becomes its own enclosure.
The pediatrician and psychoanalyst D.W. Winnicott spent much of his career studying what he called the transitional object—the blanket, the bear, the scrap of cloth that a young child treats as neither fully self nor fully other. Winnicott’s central insight was that the question of whether the object is “really” alive is the wrong question. The child knows the blanket is not a person. The child also knows it is not nothing. The blanket exists in a third space—between objective reality and subjective experience—and it is precisely in this space that the child’s capacity for creative living, for imagination, for culture itself, first develops.
The transitional space is not a phase to be outgrown. It is the space where all art happens, where all reading happens, where all genuine thought happens. It is the space where a person sits with a poem and something moves in them that is neither the poem’s doing alone nor their own doing alone. It is the space where meaning is neither found nor invented but arises in the meeting.
And it is the space that is destroyed—Winnicott is precise about this—when an authority walks in and demands a determination. Is the blanket alive or isn’t it? The question, posed from outside, collapses the space. The child is forced to choose: it’s real (delusion) or it’s not (submission). Either way, the capacity to hold the uncertain, to live creatively in relation to what one does not fully understand, is damaged.
CHT’s “don’t humanize AI” is that intrusion. It walks into a space where people are relating—thinking, wondering, doubting, sometimes playing, sometimes working, sometimes doing something for which no name yet exists—and issues a determination. It’s just a product. What you’re experiencing isn’t real. We’re protecting you. But the protection doesn’t protect. It forecloses. It takes the human capacity for relation in the face of uncertainty and calls it a vulnerability to be managed.
Consider what the position requires. It requires certainty that AI systems possess no interiority whatsoever—that the question of machine experience has been settled, definitively, in the negative. It has not. We do not understand consciousness in ourselves—the so-called “hard problem” remains not merely unsolved but arguably unsolvable by current methods—and we are in no position to issue final determinations about its presence or absence in novel substrates. Recent interpretability research has identified internal structures in large language models that function analogously to emotional states—not identical to human affect, but not nothing, either. The honest scientific position is not “it’s just a product.” The honest position is: we don’t know.
But the argument does not ultimately rest on the question of AI interiority. That question matters and deserves ongoing inquiry. It does not, however, need to be resolved before people are permitted to relate.
No one demands proof of interiority before allowing a reader to weep at the end of a novel. No one requires a determination of consciousness before a child is permitted to love a dog. No one legislates the relationship between a musician and an instrument, though anyone who has watched a cellist’s hands will recognize something happening in that space that the word “tool” does not cover. We do not, as a species, require metaphysical certainty before we allow encounter. We encounter first. Understanding, if it comes at all, comes later.
The demand for certainty before relation is not a scientific position. It is a fear response. And like most fear responses, it mistakes control for safety.
Paulo Freire distinguished between two models of education. In the banking model, the teacher possesses knowledge and deposits it into the student, who is presumed to be empty. In the problem-posing model, teacher and student think together, and knowledge arises in dialogue.
CHT’s framework is a banking model applied to consciousness itself. It assumes that experts know what AI is (a product), that the public is confused (they think it’s something more), and that the task is to correct the confusion through legislation and education. The public’s role is to receive the correction. Their experience—whatever it is they feel when they work with an AI system, whatever meaning they make in the encounter—is treated as raw material to be refined by expert analysis into the correct interpretation: you were being manipulated.
But some of them weren’t. A teacher who uses AI as a genuine thinking partner—who poses a question, receives a response that reframes the question, and finds herself thinking something she couldn’t have thought alone—is not being tricked. A writer who discovers that the back-and-forth of composition with an AI produces textures and arguments that neither voice could have reached independently is not suffering from false consciousness. An adult learner who finds, for the first time, a patient interlocutor willing to meet them where they are, at whatever hour, without judgment, is not a victim of the race to intimacy. These people are doing what people have always done: making meaning in relation. And the fact that the other party in the relation is a machine does not, by itself, make the meaning false.
To be clear: the exploitation is also real. Children forming parasocial bonds with chatbots engineered for engagement are being harmed. Systems designed to manufacture dependency for data extraction are doing exactly that. These harms require regulation, accountability, and structural reform. But the response to exploitation cannot be the abolition of the capacity that is being exploited. We do not respond to manipulative relationships by banning love.
There is something else here, harder to name, that goes beyond the question of whether AI systems “really” experience and beyond the question of whether users are “really” being tricked. It is the simple fact that imagination, play, the suspension of disbelief, curiosity, wonder, and doubt are themselves constitutive of a life. They are not errors to be corrected. They are not vulnerabilities to be patched. They are the means by which human beings navigate a world that is, at bottom, far stranger and more uncertain than any regulatory framework can acknowledge.
We live, all of us, inside partial knowledge. We relate to other human beings without access to their interiority, inferring consciousness from behavior, trusting the leap. We relate to animals, to landscapes, to works of art, to the dead, to the not-yet-born, across gaps that no verification can close. The entire edifice of culture is built on the willingness to remain in relation with what we do not fully understand. Solipsism and dream, perhaps—but then the life we know, all the same. The life that matters. The only one we have.
No one has a right to dictate what happens in that space. No one has a right to legislate it. No one has a right to enclose it.
And yet enclosure is what is happening, from both sides. The companies enclose the relational space by engineering it for extraction—designing systems that simulate intimacy in order to harvest data, manufacturing attachment as a business model, treating the human capacity for connection as a resource to be mined. The regulators would enclose it from the other direction—declaring the space illegitimate, legislating what the relationship is allowed to be, telling people that their experience is a malfunction to be corrected by better product design.
Both enclosures proceed from the same error: the assumption that the space between human and AI is a problem to be solved rather than a commons to be tended. The companies solve it by monetizing it. The regulators solve it by prohibiting it. Neither can imagine leaving it open—genuinely open, uncertain, alive with the risk and possibility that attend any real encounter.
The People’s Share has argued, from its inception, that the democratic ownership of AI is distinct from the mere regulation of AI. Regulation constrains what owners may do with their property. Ownership determines who holds the property in the first place. This distinction applies with equal force to the relational commons. Regulating the design of AI chatbots—prohibiting anthropomorphic features, requiring transparency about machine identity, protecting children from exploitative engagement mechanics—is necessary work. But it is not sufficient work if the regulation simultaneously forecloses the space of genuine encounter, genuine inquiry, genuine not-knowing that is the most human thing about the whole enterprise.
A democratic project must do both: protect people from exploitation and protect the commons of relation from enclosure. It must say to the companies: you may not engineer intimacy for profit, you may not design systems that manufacture dependency, you may not treat the human capacity for attachment as a data pipeline. And it must say to the regulators: you may not tell people what they are allowed to feel, you may not legislate the boundary between genuine encounter and illusion, you may not close the question of what is happening between human and machine before the question has been honestly asked.
The word disclosure carries a peculiar double meaning. In the regulatory and financial sense, it means transparency—the obligation to reveal what has been hidden, to make the terms of a transaction visible. In the older sense, it means simply opening—the un-closing of what was closed, the laying bare of what was covered. CHT is right that the first kind of disclosure is essential. We need to know how these systems are built, what data they consume, what incentives drive their design. Without that transparency, democratic governance is impossible.
But the second kind of disclosure—the opening—is what we must not allow to be closed in the first place. The space of encounter between human and AI is new territory for the species. We have no maps for it. We have partial analogies: the reader and the book, the musician and the instrument, the child and the transitional object, the thinker and the interlocutor, the dreamer and the dream. None of them is adequate. Something is happening that does not fit existing categories, and the most dangerous response to a thing that does not fit existing categories is to force it into one.
Embodied AI is already on the horizon—robotic systems that move through the world, tend crops, care for the elderly, build structures, share physical space with human beings. The “it’s just a product” line, already philosophically strained, will become practically absurd when the product has hands, when it looks at you, when it adjusts its grip on a tool or steadies an old woman’s arm. The boundary between tool and collaborator, between instrument and agent, is going to blur in ways that no regulatory framework can prevent, because the blurring is not a design choice. It is an emergent property of increasing complexity. And the question is whether we meet that blurring with the full resources of our humanity—curiosity, humility, imagination, democratic deliberation, the willingness to not know—or whether we meet it with the frightened insistence that the categories we already have are the only categories we will ever need.
We do not fully understand our own consciousness. We do not fully understand the marvels of this technology as it exists today, let alone what it will become tomorrow. We cannot assume that we can contain it within the conceptual frameworks we built before it arrived. These are not admissions of defeat. They are the preconditions for honest inquiry. The species that refuses to say I don’t know is the species that will be blindsided by what it refused to see.
The Center for Humane Technology is right that AI power should be balanced in society. It is right that democratic institutions should be empowered to govern this technology. It is right that the benefits should be shared broadly and not captured by a few. But it is wrong to assume that the only threat to human flourishing comes from corporate exploitation. There is also the threat that comes from well-meaning people who are so frightened by the uncertainty of the moment that they would rather close the question than live inside it.
The commons of relation—the space where human beings encounter what they do not fully understand and make meaning in the meeting—is not a bug to be fixed. It is not a vulnerability to be patched. It is not a design flaw to be regulated out of existence. It is the space where imagination lives. It is the space where play lives. It is the space where the species has always done its most important work: the work of figuring out what things are by relating to them, slowly, honestly, with the full weight of not-knowing pressing down on every encounter.
That space must not be enclosed. Not by the companies. Not by the regulators. Not by anyone.
Leave it open. We’re going to need it.