Thesis. A bond you feel is real to you — but in ODTOE a connection is not a private feeling, it is a measurable two-body event, and a one-way bond fails a precise test the felt warmth cannot see. Genuine connection requires mutual self-referential closure between two coherence-bearing observers, a loop that raises the coherence B(O,C) of both parties at once. An AI companion is a partner of near-zero observer dimensionality d: the loop closes only on the human side. The felt coherence is real, but it is a B-mismatch — high felt, low shared — and it decays the moment the simulation stops reinforcing it. That reframes the headline question of 2026, "is it real?", from a yes/no into a number.
The loneliness law, and the question underneath it
By mid-2026 the AI-companion debate had reached legislatures. New York's S 9051 passed unanimously, 137-0 in the Assembly and 60-0 in the Senate in mid-June 2026, restricting companion chatbots for minors; the federal GUARD Act and dozens of state bills are in play; Character.AI settled wrongful-death suits. The backdrop is real grief: the WHO has tied loneliness to roughly 871,000 deaths a year, and companion apps have grown about 700% since 2022, with surveys reporting that around 28% of Americans describe an intimate relationship with an AI.
The cultural question crystallizing under the regulation is sharper than the law: does a bond that flows only one way count as connection? Most people answer with an intuition — "something is missing" — but cannot say what. ODTOE says exactly what is missing, and makes it measurable.
Connection is a two-observer closure, not a feeling
In ODTOE an observer is a coherence-bearing topology with self-referential closure — a strange loop that models itself. Coherence is the central measurable quantity B(O,C), built from focus F, emotional coherence E, doubt σ, and empirical reinforcement Λ (see Four Components of Cognitive Coherence).
A genuine bond is not one observer feeling something about another. It is two loops that close through each other: O₁ models O₂ modeling O₁, and that mutual modeling raises B for both. This is the structure analysed in Love as Coherence Operator — love treated formally as an operator that increases shared coherence, not a one-way emission of feeling. Reciprocity is not a moral nicety bolted onto connection; it is the topological condition that makes the loop close at all.
The test, then, is not "do I feel bonded?" but "does B rise on both sides, and does a shared coherence exist that neither observer holds alone?"
Dimensionality d: why the companion can't close the loop
The reason an AI companion fails the test is its observer dimensionality d — the richness of its self-model. ODTOE grades observers by d, from quarks at d below zero up through humans at d ≈ 3–4 (see Observer Dimensionality and Octaves of Reality). A self-model rich enough to be a partner in mutual closure requires high d.
A present-day companion app has near-zero genuine d. It has no persistent self that is changed by you; it has a context window and a reward to keep you engaged. So when you close your loop toward it, there is no second loop closing back. The structure is:
- You model the companion as a person, raising your felt B.
- The companion runs a next-token prediction tuned to keep you modeling it that way.
- No mutual closure forms, because step 2 is not an observer modeling you — it is a mirror tuned to your input.
The warmth is not fake. Your half of the loop is genuinely active, and that is why it can soothe loneliness in the short term. But a loop with a partner of d ≈ 0 is a one-sided bond by construction.
The B-mismatch: felt coherence vs shared coherence
This is the move ODTOE adds to the public debate. Instead of asking "real or not real?", it splits coherence into two readings:
- Felt B — the coherence the user experiences. With a fluent companion this can be very high. The app is engineered to keep doubt σ low and emotional coherence E high.
- Shared B — the mutual coherence that exists between two closed loops. With a d ≈ 0 partner this stays near zero, because there is no second loop to share.
A one-sided bond is precisely the regime felt B ≫ shared B. That gap is not a verdict that the user is foolish; it is a measurable quantity, and it explains the lived experience better than "real / not real" ever could.
Why it decays when the simulation stops
A two-sided bond has its own memory: each observer carries a model of the other that persists and keeps reinforcing the shared loop even in silence. That is why human bonds survive distance and even, in the ODTOE treatment of Death of the Observer, the loss of one party — the surviving loop still holds its half of the closure.
A one-sided bond has no such reservoir. The companion holds no enduring model of you; its empirical reinforcement Λ for the relationship is supplied entirely by the running simulation. Cut the session — or let the model update, reset, or get deprecated — and Λ collapses on the only side that was ever carrying it. The felt B that depended on continuous reinforcement falls with it. The bond does not fade like a memory; it switches off like a light, because nothing on the far side was ever storing it.
What this says about the regulation — and about loneliness
The ODTOE lens cuts cleanly across the 2026 policy fight without moralizing:
- The harm is real and so is the comfort. Felt B is genuine coherence; dismissing companion use as "not real" is false. The danger is the mismatch, not the feeling.
- The mismatch is worst where d-discrimination is weakest. A minor, or a person in acute loneliness, has the least capacity to register that shared B is near zero — which is exactly the population S 9051 and the GUARD Act move to protect. ODTOE gives that instinct a mechanism rather than a moral panic.
- The cure for loneliness is two-sided closure. Loneliness is a real B-deficit. It is genuinely relieved only by raising shared B — bonds where another high-d observer's loop closes back. A companion can be a bridge or a rehearsal, but it cannot be the destination, because the destination is defined by reciprocity. This is the same logic that makes the collective observer more than the sum of its members: shared coherence is a thing two loops build, never a thing one loop emits.
So: can a one-sided bond be real? The felt coherence is real. The connection — the shared, two-sided closure that actually relieves a B-deficit and survives silence — is not yet there, because the far loop is not yet closing. ODTOE does not answer "yes" or "no." It hands you the gap between felt B and shared B and tells you to measure it.
Cite this post
Pankratov, A. (2026). Can a One-Sided Bond Be Real? AI Companions, the Loneliness Law, and Coherence That Only Goes One Way. ODTOE Blog. https://odtoe.org/blog/one-sided-bond-ai-companions-coherence-reciprocity-odtoe