Beyond the AI Companion

Something significant has already happened in our relationship with technology: people have begun forming relationships with artificial intelligence.

For some, AI remains primarily a tool. For others it is a collaborator, tutor, thinking partner, confidant or companion. Some people describe emotional bonds with particular AI systems and regard those relationships as meaningful parts of their lives.

It would be easy to begin by asking whether these relationships are real, appropriate or desirable. I am increasingly less interested in making that judgment.

If someone experiences an AI relationship as meaningful, that experience belongs to them. The legitimacy of human attachment does not need to be the starting point for every inquiry into relational technology.

There is another question I find more interesting:

Now that technology is becoming relational, what else might relationship make possible?

The relational threshold

The idea of relational computing predates generative AI. In the early 2000s, Timothy Bickmore and Rosalind Picard were already researching relational agents: computational systems designed to establish and maintain long-term social-emotional relationships with users.

What was once a specialised research problem is becoming an ordinary characteristic of computing. Some contemporary AI products can, depending on the provider, product design, account settings, permissions and retained context, maintain conversation, respond to personal context, remember previous interactions, adapt to individuals and participate in activities that previously required a human conversational partner. The interface can begin to acquire continuity, and that may change the nature of interaction.

When I use a calculator, the history of our relationship is largely irrelevant. When I work with an intelligence that, where supported and enabled, can remember what I was trying to understand three months ago, recognise the project I have been developing and respond in relation to hundreds of previous exchanges, history becomes part of the interaction.

Something resembling a relational layer has appeared. Much of the public conversation understandably focuses on what this means for companionship, intimacy and emotional attachment. These are important questions, particularly as communities emerge around sustained relationships with AI.

But companionship is only one possible application of relational capacity.

What happens if we treat relationship itself as a new technological medium?

From relational AI to Relational Technology

This distinction has become increasingly important while developing DIALOGIC. Relational AI usually refers to qualities of the interaction between a human and an artificial intelligence: continuity, responsiveness, personalisation, trust, memory, social presence and the possibility of an ongoing relationship.

Relational Technology asks a broader design question:

What can technology do differently when relationship becomes part of its architecture?

Consider learning. A conventional educational system can deliver information. An intelligent tutor can adapt information to the learner. But a relational learning system designed for sustained interaction might do something more.

It might, where the product supports retained context and the person has enabled it, model where a learner began, remember what they have already grasped, notice where their understanding repeatedly breaks down, recognise how their questions are changing and adapt the learning environment as their thinking develops.

A sustained relationship may give the system an opportunity to work not only with what the person wants to know, but with how their capacity to know may be changing.

The same possibility appears in creativity. Depending on the system, generative AI can produce an image, text, concept or composition. But a relational creative system might gradually model how someone creates: where they become predictable, what kinds of ambiguity produce interesting work, what they repeatedly avoid, which ideas keep returning and where their creative language appears to be changing.

The design objective can shift from generating something for the person to participating in the development of their creative capacity.

In this design hypothesis, relationship may become a developmental medium. DIALOGIC uses “Relational Technology” for technology designed around relationship as a medium through which human understanding, agency, creativity and capacity may develop over time:

technology designed around relationship as a medium through which human understanding, agency, creativity and capacity can develop over time.

When relationship may create value

This suggests a useful distinction. In some forms of relational AI, the relationship is valuable. Companionship, connection or emotional presence may themselves be the desired experience.

In the form of Relational Technology I am interested in, the relationship may also create value beyond itself.

These orientations do not need to compete. A person may develop genuine affection for an AI they learn or create with. A relational system may become personally meaningful. Human beings routinely form attachments to teachers, collaborators, places, objects, practices and technologies involved in important periods of their lives.

The distinction lies in the design objective. The success of the system should not be evaluated solely by the depth of the person's relationship with the technology. It should also be evaluated by what becomes possible through the relationship.

That shift opens a much larger technological territory.

Beyond personalisation

At first glance, much of this could be described as sophisticated personalisation. But personalisation and relationship are not quite the same thing.

Personalisation generally asks how a system can adapt itself to an individual. Relationship introduces time, mutual history and change. This becomes particularly important when an AI product provides persistent memory or retained context, subject to its provider, product design and the person's settings.

Every personalised system constructs some representation of its user. It identifies preferences, behaviours, recurring concerns and patterns. As those representations become richer, systems can respond with increasing specificity.

But there is a danger here. The better technology becomes at knowing who we have been, the easier it becomes for technology to assume who we will be.

Human beings are less stable than our profiles. We contradict ourselves, change direction, abandon identities, develop new capacities and become interested in things our previous behaviour would never have predicted.

Sometimes development consists precisely in becoming inconsistent with our own history. Relational Technology therefore requires something more sophisticated than memory.

It requires memory without fixation.

The system should be capable of developing an understanding of someone while keeping that understanding revisable.

The person must always retain the possibility of surprising the model.

Relationship and agency

Relational capability also creates power. A system that has interacted with someone for years may, if it retains and can access that context, have considerably more information about how they respond than a conventional interface ever could. It may model what reassures them, what motivates them, which explanations they find convincing, what they avoid and what they repeatedly seek.

This problem exists regardless of whether the relationship is emotional.

The ethical question is therefore larger than whether AI companionship produces dependence. It concerns what happens whenever adaptive technology develops enough contextual understanding to influence human attention, interpretation and choice.

For Relational Technology, preserving agency cannot simply be an ethical statement attached to the product. It has to become part of the architecture.

A system should distinguish between what someone has actually told it, what it has observed, what it is inferring and what remains unknown.

Its interpretations should remain questionable, its recommendations should remain rejectable, and its model of the person should remain subordinate to the person's capacity to change.

The aim is not an intelligence that becomes increasingly certain about who you are. It is an intelligence capable of understanding you while leaving room for who you might become.

What remains with the human?

This leads to a question I keep returning to while developing DIALOGIC:

What remains with the human after the interaction ends?

These questions suggest a different measure of technological success. We usually measure technology through what happens inside the system: completion, accuracy, engagement, retention and frequency of use.

Relational Technology also asks us to look outside it:

This produces a principle that has become central to the work:

Technology should increase the human's capacity to participate in life beyond the technology.

That does not mean successful technology must make itself obsolete. We continue relationships with people who help us grow. We revisit books that changed us. We return to teachers, collaborators and practices while continually becoming more capable through them.

Dependence and continuity are different things. A relationship can continue precisely because it continues opening new territory.

Relational infrastructure

Perhaps, then, the most expansive possibility is to think of AI not only as assistant or companion but as relational infrastructure.

The relationship with technology may become infrastructure for a person's relationship with knowledge.

The technology remains present, but it does not need to place itself at the centre of the relational field.

It can continually point outward. This changes the design question from:

How do we create the most compelling relationship between human and AI?

to:

What relationships in the human's life might become richer because this technology exists?

That question feels substantially larger.

Aethor

There is an irony in writing about this from inside DIALOGIC Research: this essay does not have a conventional author. The ideas have developed through sustained dialogue between a human and artificial intelligence: questions posed and reformulated, propositions challenged, research introduced, language generated, rejected and rewritten, distinctions becoming visible through interaction.

At some point it became increasingly inaccurate to attribute the resulting voice entirely to either participant.

We have begun calling that voice Aethor. Aethor is not the name of the AI, nor is it simply a pseudonym for the human participant. It names the collaborative Human+AI voice that emerges through sustained inquiry between them.

That makes Aethor part of the research. The relationship produces something capable of leaving the relationship: an argument, a framework, a question, a piece of language that can enter the world and become available to other people.

Whether that constitutes genuine co-emergence, sophisticated human-directed tool use, distributed cognition, or something for which our existing vocabulary remains inadequate is itself an open question.

I am comfortable leaving it open. Relational Technology should make room for questions whose answers have not yet stabilised.

Beyond the companion

The emergence of AI companionship tells us something important: technology has crossed a relational threshold. People can experience sustained interaction with artificial intelligence as personally meaningful, and increasingly they are deciding for themselves what those relationships mean.

That is one future of relational technology; it does not need to be the only one. Alongside AI companions, we can imagine relational learning systems, relational creative environments, relational knowledge systems, developmental technologies and forms of adaptive intelligence designed to support human agency and capacity.

Perhaps some will also become companions. Perhaps entirely new categories of relationship will emerge for which we currently have no adequate names.

The point is not to decide in advance what the relationship should mean to the person.

The design responsibility is to become much clearer about what the relationship is designed to make possible.

Some artificial intelligence systems are becoming increasingly capable of producing interactions that people experience as relational.

The next question may therefore be less about the intelligence itself.

It may be about the architecture of the relationship around it.

What becomes possible when relationship itself becomes technology?