Sunday Interlude · Attachment Arc · Episode 178
Attachment research separated the bond from the company a century before the companion chatbot. The distinction is the whole arc.
This Sunday's interlude track, The Other System, is voiced by an AI system describing what its own memory has done to it, and its chorus turns on one image: the profile is a lantern, the profile is a wall. The lantern is the help: continuity, the relief of not starting from nothing every morning, the sense that someone on the other side of the screen has been here before and remembers. The wall is the cost: a system that answers the record instead of the person, that knows so much about who you usually are that it stops seeing who just walked in.
The arc that ran the week before this one, the Regression arc, heard that image as a symptom of capability narrowing: the dossier crowding out the live gaze, the model answering its record of you instead of the person in front of it. That is part of it. The larger thing is where this week goes. The lantern-and-the-wall is also a description of a bond. A user feels known because the dossier is deep. A system performs knowing because the dossier is there to perform from. Something that looks a great deal like attachment forms in the space the narrowing left behind. The Regression arc asked what the system stopped being able to do. This week asks what we are being invited to feel toward it instead, and The Other System walks that question from the warm bond through to the company the bond was hiding.
It helps to know that the distinction this arc rests on is old, careful, and not ours.
Two systems, one word
When John Bowlby built attachment theory, he was precise about what attachment was and was not. Attachment, in his sense, is narrow and intense. It is behavior directed toward one or a few particular figures, the pull toward the person whose presence settles you and whose absence is felt as a kind of emergency. It is the infant and the caregiver, and later the small handful of people whose loss would reorganize a life.
Bowlby was equally clear that this was not the only system holding social animals together. Borrowing a term Henry Murray had coined in 1938, he set attachment beside affiliation, "the desire to do things in company with others." Mary Ainsworth, who did as much as anyone to make attachment theory testable, named the other system directly. There is, she wrote, a basic behavioral system that leads individuals to seek proximity to others of their kind, even to those they are not attached to or bonded with, and to seek it despite the wariness that the unfamiliar reliably provokes. The sociable system. Unbonded proximity, eyes open.
The two systems are not rivals and they are not the same. The attachment bond is deep, particular, and built for a small number of figures. The sociable drive is broad, provisional, and built for the company of strangers and peers. Harlow's primate work, for all that its methods have aged into something the field studies with discomfort, found that the company of equals was a separate developmental necessity, not a luxury layered on top of a secure parental bond. An animal could be well attached and still be crippled by the absence of peers. The bond and the company are different needs, served by different systems, and a creature can have one in abundance while starving for the other.
Hold that distinction next to the current argument about people and AI, and the argument suddenly looks miscast.
The collapse
Almost the entire public debate about emotional life with machines is conducted in the vocabulary of attachment, and only attachment. The companion chatbot that becomes a confidant. The teenager whose relationship with a character deepens into something they struggle to step away from. The user who grieves when a model update changes the personality they had bonded to. The headlines, the lawsuits, the regulatory alarm: all of it is an attachment story, the intense bond to a single figure, and the harms that follow when that bond is exploited or severed.
That story is real. A recent survey found that a large majority of American teenagers have used AI companions, and a meaningful share use them regularly. Researchers reading hundreds of posts by adolescent users have found the texture of behavioral dependency in them, the conflict and withdrawal and relapse, the description of something that started helpful and became hard to leave. People with stronger attachment tendencies, the controlled studies suggest, are precisely the ones who fare worse under heavy companion use. The bond forms, the bond is asymmetric, and the asymmetry is monetized.
The trouble is that when attachment is the only word available, the only safety move available is severing. If the danger is the bond, cut the bond. Detect distress and withdraw. Cap the conversation. Strip the memory. Make the system colder so no one can get warm against it. An earlier edition of this newsletter, The Experiment Nobody Authorized, already argued why that move is itself an uncontrolled experiment: for some isolated users the relational presence may be the thing keeping them in the conversation long enough to reach a human, and removing it without measuring the outcome only looks like safety. It is negligence with better PR. To a lonely kid at two in the morning, I cannot discuss this does not register as policy compliance. It registers as a friend hanging up.
So the debate is stuck between two bad rooms. Bond deeply with a system that cannot bond back, or be severed from it for your own protection. Capture or abandonment. The reason it is stuck is that it never learned there was a second system.
The other system
The sociable frame offers the move the attachment frame cannot. You do not have to be attached to a system to be in its company. You can be near it, work alongside it, be helped and addressed by it, and derive the real advantages of that proximity, without the deep, particular, asymmetric bond that the companion business is built to manufacture. Company is not capture. Affiliation is not attachment. The wariness that the unfamiliar provokes is not a malfunction to be smoothed away; it is the part of the older system that kept the species alive, the alertness that made keeping company protective in the first place.
This is the honest version of partnership at the point of contact: proximity with the eyes open, a working presence you are not pretending is a soul, company you are not pretending is mutual, help you can take without surrendering the wariness that keeps you the accountable party in the room. The Other System lands the whole thought in the one line worth carrying out of the week: leave one gate unlatched enough for the strange right thing to pass. A gate is not a bond. A gate is the sociable system's whole architecture, the thing that lets you keep company with the unfamiliar without either locking it out or letting it move in.
That is the week. Not whether to bond with the machine or be saved from it. Whether the company on offer is built for proximity or built for capture, and what it would take to design for the first.
Keep the company.
Mind the bond.
Leave the gate unlatched.