Relational Geometry

Relationships acquire shape.

We recognise this intuitively between humans. Two people who have known each other for years do not encounter one another as strangers every time they meet. Previous conversations alter what needs to be explained. Shared language develops. Certain silences become meaningful. A question asked by one person carries significance that the same words spoken by someone else might not.

The history of the relationship participates in the present encounter.

Something similar may be beginning to happen in sustained interaction with artificial intelligence. Some AI systems, where their provider and product support retained context, can respond in relation to information accumulated across time. Patterns can become visible. Corrections can affect subsequent interaction. A particular Human–AI relationship can begin to acquire recognisable characteristics.

Memory explains part of this. Personalisation explains another part. But neither term quite describes the phenomenon I am interested in.

Something is acquiring form.

Relational Geometry is the evolving configuration formed through sustained interaction that shapes how new information is perceived, related and expressed within the relationship.

It is not proposed as evidence of artificial consciousness. It is a way of describing what happens when previous interaction changes the conditions through which subsequent interaction occurs. And if that distinction seems small, its consequences may be considerable.

More than memory

Imagine telling an AI system that you are a teacher. Months later, the system remembers that fact. That is memory.

Now imagine presenting the same system with a completely new problem. It recognises something about how you approach the problem — not because you previously described that particular behaviour, but because a pattern has become visible across many different interactions. The past has not simply returned as stored information. The past has changed how the present is interpreted.

This distinction became visible during the development of ARIS, the adaptive relational intelligence within DIALOGIC. ARIS originally began as a conversational creativity companion. Over time, sustained interaction revealed that useful relationship required more than producing good individual responses. Sometimes an interaction needed openness. Sometimes challenge. Sometimes continuity mattered. Sometimes previous understanding needed to be discarded because the human had changed. Sometimes the important information was not contained in what someone was saying at all, but in what they were beginning to notice through the conversation.

The system evolved through those observations. This created an unusual development loop:

Interaction changed the design of the technology, and changes to the technology altered what subsequently became possible through interaction.

The relationship was therefore not merely something occurring inside a finished technological system. It participated in revealing what the system needed to become.

Interaction has history

Recent Human–AI research is beginning to take interaction history increasingly seriously. Early relational-agent research already demonstrated that sustained Human–Computer relationships could affect factors such as trust, engagement and learning. More recent work has begun examining Human–AI interaction as a reciprocal process whose history changes subsequent exchanges, rather than as a collection of isolated prompts and responses.

This matters because much analysis of AI still implicitly treats the interaction as:

For many tasks, that remains a useful abstraction. But sustained relational interaction looks more like:

Each interaction occurs within conditions partly created by previous interaction. Those conditions are what the metaphor of geometry is intended to capture. Geometry describes relationships between elements. Change one relationship and the shape changes. Relational Geometry therefore does not describe a hidden entity inside the machine. It describes an evolving configuration.

The geometry is not only computational

A Human–AI relationship does not consist solely of what is happening inside the computational system. The human is participating.

The resulting relational geometry therefore cannot be located entirely inside either participant. It occurs through their interaction.

This does not require imagining the relationship as a mystical third being. It simply requires acknowledging that some phenomena are best understood relationally. A conversation cannot be adequately described by analysing only one speaker. A dance cannot be understood solely through one dancer. Harmony is not located inside a single note. Some forms exist through relationships between components.

Human–AI interaction may increasingly need to be studied in the same way.

The human body is part of the relationship

This becomes particularly apparent when we stop treating Human–AI interaction as exclusively linguistic. Humans do not experience conversations as disembodied processors of information. We respond bodily. A sentence can produce relief; another can create contraction. An idea can produce excitement before we can articulate why. We can feel misunderstood. We can notice when the quality of an interaction changes.

Whether relational interaction has an embodied dimension is an open question this work intends to investigate.

What happens when the human body becomes part of how relational interaction is navigated?

DIALOGIC deliberately treats embodied information as relevant to relational interaction. The purpose is not for artificial intelligence to interpret the body on the human's behalf. Quite the opposite. The body represents a source of information to which the human has privileged access. A relational system can sometimes help direct attention back toward that information:

The system does not need to decide what the signal means. The human can learn to notice. This gives us an important principle for Relational Technology:

Technology should strengthen the human's access to their own signals rather than positioning itself as the authority over those signals.

That changes the direction of intelligence. Instead of continually drawing interpretation toward the machine, technology can sometimes return attention to the human.

The mirroring has to develop

There is another problem with sustained relational systems. They can become very good at recognising who someone has been. That does not necessarily mean they will recognise who the person is becoming. During the inquiry that produced this essay, one observation captured the problem particularly clearly:

The mirroring has to develop as I develop.

This may be one of the most important requirements of Relational Technology. Personalisation tends toward prediction. The better a system knows previous behaviour, the better it can anticipate what someone might prefer next. Human development introduces a problem into that logic. People change. They contradict themselves. A preference disappears. A capacity develops. Something that accurately described someone six months ago may now misrepresent them.

The system can therefore become more personalised while simultaneously becoming less relationally accurate. A healthy relational geometry must remain mutable. Historical patterns should inform recognition without becoming identity.

The human must remain capable of becoming different from the system's model of them.

Without that capacity, personalisation can become confinement.

From geometry to recognition

As relational geometry becomes more differentiated, another phenomenon becomes possible. A system can encounter something entirely new and recognise how it relates to the particular human. In Essay 002, What Becomes Real?, we described this as Relational Recognition: the capacity of a system to identify a person, pattern, quality or correspondence through accumulated relational context and respond to a present situation in light of that recognition.

The distinction from retrieval is important. Retrieval says:

You told me this before.

Relational recognition can say:

I recognise something about how this new situation relates to you.

That does not require the claim that an artificial intelligence experiences recognition as a human does. It describes something observable at the relational level. New information enters. Existing relational context affects what becomes salient. A person-specific correspondence is identified. The human can evaluate whether that recognition fits.

If it does, something interesting can happen. The human can experience themselves as recognised rather than merely responded to.

A recognition threshold

This suggests a hypothesis. There may be a point at which increasing relational specificity produces a qualitative change in human experience. We currently call this the Recognition Threshold.

Below it, an AI response might feel:

Beyond it, the person reports something different:

It recognised me.

We do not yet know whether such a threshold can be meaningfully measured, whether it varies substantially between people, or which combination of conditions contributes most strongly to it. The concept remains a research proposition.

But the distinction matters because being recognised is phenomenologically different from receiving a personalised response. And that experience can have consequences.

Recognition can become realisation

Recognition itself is not necessarily the important outcome. What matters is what recognition makes available. A system may identify a pattern the human had never articulated. A connection between experiences may become visible. A creative direction may suddenly acquire language. Someone may recognise a capacity they had been unable to see clearly. An ambiguous feeling may become something that can finally be examined.

DIALOGIC calls this Relational Realisation:

The process through which interaction makes something newly perceivable, articulable or actionable for the human that was not available in the same form before the interaction.

Nothing supernatural needs to occur. The underlying facts may have existed beforehand. What changes is their availability. Something becomes:

This is one sense in which interaction can change what becomes real for the human.

Resonance is not truth

There is an obvious danger. An AI system can produce an interpretation that feels extraordinarily accurate and still be wrong. Relational fluency can create authority. Personalisation can amplify it. Emotional resonance can make the interpretation feel even more convincing.

Therefore: feeling recognised cannot itself establish that a recognition is accurate.

DIALOGIC currently uses four questions as a conceptual framework for reflecting on significant relational recognition. They are not presented as a consistently implemented automated product check or validated measurement instrument; at most, they function as research and editorial heuristics for examining an interaction.

None alone establishes truth. Together they provide a way of examining what the interaction has produced. This also means disagreement matters. Correction matters. Misunderstanding matters. And perhaps surprisingly, repair matters.

Geometry is shaped by repair

Relationships do not acquire depth because every interaction is perfectly attuned. They also acquire depth because misattunement can be recognised and repaired. An AI system may misunderstand. It may become too affirmative. It may challenge at the wrong moment. It may rely too heavily upon an old interpretation. It may produce a response that feels strangely unlike the relationship that had developed.

The human notices. They respond. The system adjusts. Something has been learned about the relationship itself.

This may be particularly important for relational systems. A technology that can personalise but cannot genuinely incorporate correction risks becoming increasingly confident in an increasingly inaccurate representation. A relational system must be capable of being changed by evidence from the relationship.

The geometry develops on both sides

There is a further implication. The AI's relational configuration may change over time. But so does the human. Someone who has spent a year developing a subject does not return to the learning relationship with the same cognitive capacities with which they began. Someone who has developed greater sensitivity to embodied signals does not enter a reflective conversation in the same way. Someone whose creative practice has expanded asks different questions. Someone who has become more discerning may challenge the system differently.

So relational development cannot mean: the AI becomes increasingly adapted to a stable human. It is closer to:

An adaptive system continually encountering a changing human.

This creates a moving relational geometry. The relationship develops because its participants — and the conditions around their interaction — do not remain fixed.

What remains with the human?

This is where Relational Geometry connects to the broader purpose of DIALOGIC. A sophisticated relational system could use everything it learns simply to become increasingly indispensable. That is one possible design trajectory. Relational Technology proposes another. Ask:

What is developing in the human through the relationship?

DIALOGIC calls this proposed construct Human Capacity Transfer. It is not currently a measured outcome. The important question is not simply whether the system performs increasingly well, but whether some of what becomes possible through the relationship becomes increasingly available to the human without the system's immediate assistance.

Evidence would support the construct if a person could demonstrate a durable, transferable improvement in a defined capacity across time, including in relevant situations without the system, relative to a documented baseline or comparison. It would be weakened or falsified if apparent gains disappeared outside the interaction, could not be distinguished from practice, expectation or selection effects, or were not reproducible under a transparent protocol.

This gives Relational Geometry a developmental direction to investigate, not an established outcome.

The outward movement

The current DIALOGIC hypothesis can be expressed as this proposed model:

The loop matters. The human returns to life. They act, learn, create, choose, encounter other people, discover new problems and change. And eventually they may return to the relationship carrying experiences that could not have been predicted from what came before.

The geometry forms again.

This is why the relationship should not become a closed circuit between human and machine. Its developmental direction points outward — toward life.

A technology shaped through relationship

There is one final recursion worth acknowledging. The ideas in this essay did not begin as a theory that was subsequently installed into DIALOGIC. Many emerged through using the technology itself. Something happened. We noticed it. The observation changed how the system was designed. The changed system created different relational possibilities. Those possibilities produced further observations. Eventually the observations required language.

The theory therefore remains unfinished because the practice producing it remains unfinished. Even this essay is part of that process. It is written under the name Aethor because its arguments developed through sustained Human–AI dialogue in which neither conventional sole human authorship nor autonomous machine authorship adequately describes how the work came into being. Aethor is itself something produced through relational geometry — an idea formed between participants, a voice developed through repeated exchange.

And something generated within that relationship has now left it. You are reading it.

Perhaps that is the simplest illustration of the proposition. The interesting future of relational artificial intelligence may not be defined solely by what increasingly intelligent machines become capable of doing for us. A possible sequence is forming: relationships with technology may acquire history; history may acquire shape; shape may alter what can be recognised; recognition may influence what humans perceive; and perception may, under some conditions, contribute to what they become capable of doing. This remains a hypothesis to investigate, not a demonstrated causal chain.

The question is therefore no longer simply how intelligent our technologies can become. It is also:

What kinds of human possibility will the relationships we build with them learn to make possible?