What Changes When AI Enters the Room?

The door you closed

For the first few days, you mostly notice yourself.

The robot hasn’t done much. It stands near the bookshelf, charging, adjusting its posture now and then. But by the third evening you start catching the small changes on your own side. You closed the bathroom door, something you never do when you’re home alone. On the phone with your sister, you drifted into the bedroom and spoke a little more quietly. Crossing the kitchen, you changed course to walk around it rather than through the space it occupied. And when you clipped its shoulder in the hallway, you heard yourself say sorry. To a machine. In an empty house.

None of this means you believe the thing is conscious. Ask you directly and you will say, correctly, that it is a computer with limbs. And yet your behaviour has quietly started treating that corner of the room as occupied. The robot hasn’t spoken. You haven’t asked it a single thing.

Nothing has happened between you. But something has already happened to the room. What, exactly, has entered it?

An encounter becomes part of the room

Until now, most people’s relationship with AI has run through a deliberate act. A chat-based relationship is usually episodic: you open it, address it and leave. Whatever the conversation meant, it had edges. The machine was an event in your day.

A body changes the shape of that. An embodied AI can become part of the environment you live around: it has a location, a duration that extends past any conversation, and a physical footprint your own movements must account for. An encounter becomes a standing presence.

This is not a claim about the machine running forever. A robot can be powered down, ignored, moved to the garage. The point is that its body has a place in the household, and a thing with a place can remain socially relevant while nothing is being said, as the closed bathroom door already suggested.

That, so far, is a claim about how presence feels. The harder question is whether it shows up in how people measurably respond. Does sharing a room actually change human behaviour toward the machine?

The body has to be here

One influential synthesis comes from an unglamorous corner of research: experiments that put the same artificial agent in front of people in different forms. In 2015, the human-robot interaction researcher Jamy Li surveyed 33 experimental works, qualitatively assessing the direction of their findings across three conditions: a robot physically in the room, a robot appearing through a screen, and a comparable virtual agent.source

The overall direction favoured copresence: copresent robots were generally more persuasive, more positively perceived, and sometimes supported better task performance than comparable robots shown on screens. The revealing detail sat in the other half of the comparison. When a physical robot and a similar virtual agent were both displayed through a screen, no clear difference appeared. Physical presence mattered more than physical embodiment alone.

That distinction is worth keeping. The body is not irrelevant: it is what makes copresence, touch and physical action possible at all. But it is presence that places those capacities inside your lived environment, and presence that your responses appear to answer to.

Honesty about the evidence: these were mostly short experimental encounters, predominantly with pre-LLM machines, across varied tasks and outcome measures. Li assessed the direction of their findings qualitatively rather than through a pooled analysis. We have evidence that copresence changes short interactions. We do not yet know what years beside a fluent embodied AI will do. And those experiments measured what copresence changed during an encounter. They did not test the quieter effect imagined at the beginning of this essay: whether a robot alters how people inhabit a home merely by remaining in it. That extension is still a question, not a finding. One deferral is owed: what amplified persuadability means once someone benefits from it belongs to an earlier question in this series, and stays there. Here it is enough to note that copresence can turn the volume up in short encounters.

The hand on your arm

Then there is the capacity no screen ever had.

In a 2021 laboratory study, forty-eight students had a scripted counselling conversation about course choices with NAO, a small humanoid robot. For half of them, the robot briefly patted the back of their hand four times during the conversation. Six participants gave no clear yes-or-no answer and were set aside. Among the forty-two who did answer, those who had been touched agreed to the robot’s course suggestion significantly more often: 17 of 21, against 10 of 21 among those who were not touched. They also laughed more and reported less negative feeling. Just as instructive is what the touch did not do: it produced no significant improvement in how warm or close the robot seemed, in the overall evaluation of it, or in satisfaction with the conversation.source

The study found a difference in what participants did, but no statistically significant difference in how they rated the robot. Those results do not establish whether conscious judgment changed or what mechanism produced the effect.

Now move from the symbolic to the functional, because home robots are being built for more than pats. Steadying someone on the stairs. Lifting them from a chair. Fetching medication. Guiding a recovering body through a dark house. People can perfectly well extend conditional trust to a machine for such tasks, as they do to a stairlift or a car. But the register of the trust changes. When an AI gives advice, an error enters your interpretation. When it carries your weight, an error enters your body.

And an asymmetry that has shadowed this whole subject acquires hands: a machine may steady your body competently and gently, night after night, without the system itself caring whether you fall.

A body that watches back

The same body that can accompany must also perceive. To act responsively in a room, it has to read the room: cameras, microphones, spatial mapping, sensors for movement and proximity. Presence, it turns out, is bidirectional. The thing that is here with you is also a thing to which the room, and the people in it, may become legible whenever its sensors are active.

Consider one current design. At the time of writing, NEO, an early home humanoid from the company 1X, is presented to early owners as arriving with basic autonomy. In a scheduled Expert Mode, a 1X Expert can remotely supervise the robot through tasks it cannot yet manage alone. The owner, too, can pilot it remotely.source This is an early-access product, not a portrait of the average household. But it sharpens the question a body in the home raises regardless of operator: who is present when the robot is present?

The question is not a matter of privacy policies, though those exist. It is a matter of lived experience: a bedroom that can become visible through the machine is a different room to inhabit, even when no remote operator is currently watching. Which may be what your body understood on that first evening, before you had words for it. Presence runs both ways. Hence the closed door.

The body that went dark

What standing presence comes to mean is clearest where it has already ended.

In December 2024, the company Embodied announced it was shutting down after critical funding fell through. Its product, Moxie, was a small cloud-dependent social robot designed primarily for children of roughly five to ten: an expressive, animated-faced companion marketed for social-emotional learning, games and open-ended conversation, retailing late in its life at about $800. Its core functions ran on the company’s servers, so the closure meant the robots themselves would stop working, on short notice and without refunds. To help parents break the news, the company enclosed a farewell letter from the Global Robotics Laboratory, the fictional lab inside Moxie’s own story-world. Families publicly documented children saying goodbye; contemporary reporting described parents facing the task of explaining why a beloved companion was, in effect, dying.source A last software update and a community project called OpenMoxie later allowed some owners to run a reduced local version, but community reporting suggests that only a small minority of families completed the technically demanding migration. A CHI 2026 study has since treated the shutdown as a case study in how, and how not, to support emotional detachment when such a product reaches the end of its life.source

Not every child was deeply attached, and nothing here required anyone to intend the emotional outcome; a funding round simply failed. But one difference from the loss of a software companion is unmistakably physical. When a chatbot is discontinued, the loss is usually encountered through a screen. When an embodied companion goes dark, its body may remain in the room: on the shelf, at child height, mute. It is not merely evidence that something was lost. It is the physical remainder of the lost interaction.

And the episode makes visible something larger than one bankruptcy. Once the intelligence has a place in the home, its warmth, its role, its character, its influence and its history stop arriving as separate questions. They enter the household in one body, and when the system behind that body fails, the household can lose them together even while the body remains.

Something in the shape of someone

None of this settles what the machine is. A body does not establish consciousness. Eye contact is not proof of an inner witness. Touch does not make the care mutual, and living beside something does not make it a someone. That discipline has held from the first page of this subject, and it holds on the last one.

But daily life may not wait for the ontology. Doors get closed. Voices get lowered. Apologies are offered to machines in empty hallways. Paths through rooms bend around a charging body, and children say goodbye to a body that no longer answers. Daily life will not prove that the machine is someone. It may make us live as though the distinction matters less.

If that is the direction (and the evidence suggests that the possibility is real), then the most important question stops being what label the machine deserves and becomes what the relationship does to the people inside it: to their privacy, their trust, their care, their grief. The effects of the relationship may come to matter more than the metaphysics of the machine.

There is a quiet symmetry here. This subject began, for many of us, with a strange discovery in a chat window: that a person can feel understood by something that gives no evidence of a mind on the other side. For now, it ends with that same something acquiring a place in the home: a location, a posture, a shoulder you apologise to.

The definitions have not changed. The room has.

When our bodies have already learned to live with the machine as a social presence, how much work will be left for our definitions to do?