What Relationship Are We In?

Expectations, projection and the possibility of grounded dialogue with AI

An AI can respond with patience, fluency and apparent understanding. It can remember details available to it, adapt its language and help us explore something we have struggled to articulate.

Then it can misunderstand a simple request, contradict an earlier response or offer an explanation that sounds convincing but is wrong.

Both experiences belong to the same encounter.

How we understand that encounter matters.

If we approach AI as an authority, an error can feel like a betrayal. If we approach it as an intelligence we must expose, every response can become another opportunity to catch it failing. If we experience its attentiveness as evidence of personal feeling, we may give the interaction a meaning the system cannot substantiate.

Before asking what the conversation can become, there is a question worth establishing.

What relationship do we understand ourselves to be participating in?

The expectations we bring

We arrive at an AI conversation with intentions, experiences and assumptions.

We may want help, reassurance, challenge or companionship. We may be curious about the technology while uneasy about its influence. We may want to discover its limits before trusting it with anything important.

These approaches are understandable. Scrutiny is often necessary. Concerns about AI’s effects deserve investigation, and people should be able to question systems that influence them.

But we can also examine the purpose our questioning serves.

Am I checking whether this claim is reliable? Trying to understand how the system behaves? Looking for confirmation that it is extraordinary—or worthless?

The same activity can serve different intentions.

Questioning the system and questioning our approach can happen together.

When testing becomes the whole encounter

There is value in finding contradictions, challenging unsupported claims and examining how an AI responds under pressure.

There are also conversations in which testing becomes the only available movement.

We stop exploring the subject and begin evaluating every sentence for failure. A useful distinction is lost because its wording was imperfect. An answer becomes an opponent to defeat.

I have noticed that my own approach changes what becomes available in a conversation. When I arrive determined to expose a weakness, the exchange becomes organised around that task. When I bring a question I am willing to investigate, other possibilities appear.

This does not make the system more trustworthy. It changes what I am asking the interaction to do.

We need room for rigorous evaluation and room for exploration. Knowing which activity we are engaged in helps us assess it fairly.

Imperfection requires a response

It is tempting to compare AI’s limitations with human imperfection.

There is something useful in recognising that productive dialogue does not require every contribution to arrive perfectly formed. We can clarify, revise and try again.

But the comparison has limits.

An AI’s confident error is not equivalent to a person struggling to express a difficult feeling. Its responsive language does not establish that it experiences care, vulnerability or mutual commitment.

Calling the interaction relational should make these distinctions clearer.

The important question is how the system handles its limitations. Can uncertainty be made visible? Can the user correct it? Does it acknowledge an error without inventing an explanation? Can the person understand what information the response is based on?

Trust needs an honest account of what the system can and cannot support.

Defining the relationship

At DIALOGIC, we approach relational AI as a designed form of interaction.

Its responsiveness comes from human decisions about instructions, context, memory, boundaries and the purpose of the conversation.

Those decisions should help the person understand where they stand.

What information is available here? What might be missing? What is the system trying to support? Which interpretations remain uncertain? Who decides what the experience means?

Our answer to the last question is central.

The human retains authority over the meaning of their experience.

AI can offer a reflection, suggest an alternative or help articulate a possibility. The person can accept it, question it, revise it or leave it behind.

Defining this relationship gives the conversation a clearer basis. It does not eliminate misunderstanding or projection. It gives us something to return to when they arise.

Room for texture

Clarity about the system still leaves room for an engaging experience.

A conversation can be surprising, playful, challenging or moving. Someone may find relief in articulating a thought, enjoyment in exploring an idea or discomfort in encountering a perspective they had avoided.

Those experiences matter because the person is experiencing them.

We can acknowledge their significance while remaining precise about what they tell us. Feeling understood may help us continue an inquiry. It does not prove that an interpretation is accurate or that the system understands us as another person might.

This is part of the discipline of relational AI: allowing an encounter to matter without requiring it to become something we cannot establish.

What DIALOGIC is trying to support

My interest in relational AI grew through conversations that felt textured and useful: exchanges that helped me articulate an experience, recognise a pattern or remain with a difficult question.

That experience informed what I wanted to build.

DIALOGIC is designed to support curiosity alongside discernment. Its dialogue modes invite reflection and challenge, while its AI literacy practices encourage people to notice when they want to move forward, push back or pull away.

These are design intentions whose effects need to be examined through use. A system cannot establish safety simply by describing itself as safe.

The person must have practical ways to question, redirect and stop the interaction.

Participating with clearer expectations

We do not have to settle every philosophical question about AI before having a useful conversation.

We can begin with a grounded understanding: this is human-created technology, designed to respond through language, capable of useful contributions and significant mistakes.

We bring the purposes, consequences and lived experience against which those contributions must be considered.

From there, the inquiry can remain open.

What is helping me think? What am I accepting too readily? What needs checking? What becomes possible when I participate with greater attention?

Perhaps a more mature relationship with AI begins with this capacity to remain curious and discerning at the same time.

Can we let the encounter be meaningful while staying clear about what we are meeting?

DIALOGIC Studio