Is Conversation with AI a Reciprocal Relationship?

A lit window late at night

In an imagined scene, a man returns home after a difficult day and opens a chat window. He writes, “I don’t want a solution now; I want to understand what I feel.” A calm answer arrives, reorganising his words and asking a question he had not considered. He feels some relief, then thinks: finally, someone understood me. His relief deserves no ridicule; its effect is part of his experience. But “someone” carries another question: who was present with him, and in what sense?

This is a philosophical reflection on reciprocity between a person and a conversational assistant. It draws on specific research without diagnosing technology users or settling whether machines might be conscious in the future. We distinguish the usefulness of words, a sense of closeness, and the conditions of a relationship to which we attach friendship’s promises. The first two may coexist without proving the third.

Before continuing: can we feel heard without the respondent being a friend? What distinguishes affectionate language from a commitment a person bears towards us? When does digital conversation help us return to people, and when does it postpone that return? How can we preserve an experience’s value without overstating what it means?

1. Responsiveness alone is not reciprocity

In a simple sense, an exchange occurs when I write and the answer changes, then I respond and the conversation changes direction. We need not deny this interaction to use words accurately. The harder question concerns what we usually add when calling a relationship reciprocal: each party cares about the other’s good and has a life and needs not reducible to what they provide their partner.

Aristotle treats friendship as mutual goodwill known to both parties. We borrow this ethical question, not his entire classification of relationships: is the concern directed towards another with a good of their own, or towards a benefit we receive? This is a philosophical starting point, not an empirical test of models’ consciousness. Calling a system a “friend” is insufficient to establish that the meaning applies.

A book by someone who died centuries ago may console us, a notebook may organise our thoughts, and a poem may change a whole day’s mood. These effects do not depend on reciprocal responsiveness at that moment. Likewise, conversation with a program may help without its usefulness establishing that the program feels grateful to us or misses our return.

Yet a conversational assistant differs from a fixed page in responding to our particular words. It can reuse details we supplied and appear attentive to a change of subject. This tailored responsiveness makes the relationship question understandable; we should not treat users as though they confused a stone with a person. What is needed is a more accurate description of what happens, not denial of what people find in the experience.

We can say: the conversation helped me name my confusion. That describes an effect its author can observe. “This system loves me” adds a claim about the respondent’s state. Moving between these statements requires independent evidence; it is not guaranteed because the second feels warmer or the first fails to describe everything we felt.

2. The name we give the window

Names are not always neutral containers. Presenting something as a tool prepares us to ask what it accomplishes. Presenting it as a companion may bring expectations about memory, intention and care. Measuring this observation requires research, and a study by Alison Chin and colleagues published on the Microsoft Research page addressed it.

The researchers randomly assigned 470 participants to videos presenting language models as machines, tools or companions, or to a group watching no video. Companion framing was associated with attributing greater mental capacities to the models; the finding recurred in a follow-up study of 604 participants. The measure concerned human beliefs, not proof that the system possessed those capacities.

The finding gives us reason to question presentation language, but does not say every user believes every label they hear. Someone may know they are dealing with a program and use friendship language metaphorically. The issue is not forbidding metaphor, but the point when metaphor starts producing practical expectations unsupported by available knowledge.

If an interface says, “I am always here for you,” some may read a personal promise into it. Yet service continuity and terms depend on human organisations and operational decisions. Designers therefore also bear responsibility for clarity: what does the phrase promise, who can fulfil it, and where can users understand the limits before building important reliance on it?

We also have a role in how we read. We can ask what “I remember” means in a particular service rather than assume human memory, and what “I understand you” means rather than treat it as testimony to an inner experience. Such questioning does not deny us usefulness. It prevents interface rhetoric from replacing knowledge of what the service can actually do.

3. What research says and what it does not settle

In the version of Cathy Mengying Fang and colleagues’ study dated 2 October 2025, 981 people took part in a four-week conversational experience with random assignment of interaction modes and topics. The study examined loneliness, interaction with people, emotional dependence and problematic use. It detected no significant effects of the experimental conditions, while greater voluntary use was associated with worse outcomes on the measures studied.

We should not turn the association into a judgement that every increase in use causes harm. Nor does finding no significant difference between conditions establish the absence of every possible effect. The study is bounded by its duration, design and measures. It gives no direct answer about the system’s consciousness or the suitability of every individual use. These are limits on inference, not a device for rejecting an unwelcome finding.

It helps to separate three questions. First: did I feel relief after the conversation? Second: what happened to my habits and relationships over time? Third: what is the nature of the respondent? Someone may answer yes to the first while the second still requires observation and they lack sufficient evidence to settle the third at all. Combining them into a single answer makes discussion more heated and less accurate.

Nor should every minute spent with a digital assistant be put on trial. Conversation may rehearse a question, organise an idea or provide limited entertainment. Concern begins when we treat one moment’s relief as sufficient evidence for a larger decision: dispensing with our relationships, or giving an answer authority over choices whose full context the system does not know.

Instead of diagnosing, we can examine specific actions. Did I leave the conversation able to take a step I had postponed? Did I widen the possibilities or keep requesting answers until one agreed with me? These are reflective questions proposed by the article, not a standardised psychological scale. If distress persists or life is disrupted, they are insufficient as a substitute for qualified human support.

4. The other whom we cannot tailor to ourselves

Human friendship involves a person whose presence we cannot wholly determine. They may love us and be tired, listen and then disagree, or remember an event from a perspective we dislike. These are not shortcomings to celebrate for their own sake; people can also neglect and hurt others. But they remind us that closeness to another includes recognising their independence.

Imagine, hypothetically, that we want to compose an apology. An assistant can suggest clear wording and flag excessive justification. But the apology’s effect also depends on the injured person: they can ask for time, reject our wording, or notice an action we overlooked. We cannot generate a new reply from them as we regenerate a paragraph we dislike.

An important ethical difference appears here. We can improve a text until it seems balanced, but repairing a relationship requires accepting that the other person is not part of our text-improvement project. They have a right to their account and the time they need. A tool helps more when it prepares us to bear the encounter than when it gives us a ready-made story making the encounter unnecessary.

This does not mean everyone must return to every relationship. Some relationships may be harmful, and support circles narrow. A fair comparison does not confront a lonely user with an imaginary choice between a perfect friend and an imperfect program. We can acknowledge limited resources and available benefit while leaving the door open to suitable, safe human connections.

A serious objection: is relief not enough?

Someone may object: I do not care whether the respondent feels; what matters is that I can speak without embarrassment. This objection has real force. Not every use requires a complete philosophy of friendship, and we need not feel guilty because a tool helped when no listener was available.

But relief is enough to judge one aspect of an experience, not all its consequences or every description of it. We can accept the benefit of conversation while rejecting its marketing as an unproven promise of mutual care. We can also use it without assigning it absolute authority. We need not choose total gratitude or total rejection; we need to place the benefit appropriately.

In the light of the Quran and Islamic wisdom

Al-Saadi explains the prohibition as requiring verification in what a person says and does, with responsibility for their means of knowledge. Our proposed application distinguishes saying we felt relief from saying the system has feelings towards us. The verse is neither a technical description of models nor empirical evidence about consciousness. Its connection here concerns care in attributing knowledge to ourselves.

The verse explains humanity’s shared origin, mutual acquaintance and the criterion of piety. Al-Saadi connects acquaintance with cooperation and fulfilling rights. We draw on this meaning to view a relationship as more than consuming a comforting response: it includes actions and responsibilities between people. This does not prohibit using an assistant. It reminds us that technology can serve human acquaintance without alone replacing its world.

The hadith addresses human beings and their value; it does not test a machine’s interior. We connect it to our subject through our own responsibility: does fine speech help us act better? We may compose an eloquent message and then postpone fulfilling it. In this reflection, the value of a means does not exempt its user from sincerity in action beyond the screen.

Applications: make conversation a bridge

Set an understandable purpose before starting: organising a question, drafting a message or exploring an objection. The clearer the purpose, the easier it is to know when you have what you need. You need not measure life with a permanent counter, but the end of every conversation need not automatically begin another whose purpose you never identified.

Ask for possibilities rather than final judgements about absent people. “What are the possible interpretations of this statement?” is fairer than “Confirm that he intends to hurt me.” Remember that your account is itself partial. Do not give generated text the role of witness to an event it did not attend, or diagnosis of a person it did not meet.

Protect your own and others’ privacy, and read the terms of the service you use before sharing sensitive details. Return important decisions to evidence, qualified people and their real context. These are general precautions; services differ, and the article assumes no single policy for all of them.

Finally, choose a small action outside the window if your purpose calls for it: ask the person concerned, check a source, or send the apology whose responsibility you have decided to bear. The best result of a conversation may be needing to continue it less right now because you are better able to take the step you came for.

Closing: real benefit and accurate description

We return to the man in the opening scene. He can keep his gratitude for the words that helped, then ask what he wants to do with them. He need neither invalidate his relief nor turn it into proof of reciprocal affection. Between denial and exaggeration is room for useful interaction and honest description.

Conversation is reciprocal interaction in the alternation of roles, but reciprocal friendship is a richer claim. What we know about an answer’s effect is not enough on its own to know the respondent’s inner state. The window may be very useful; wisdom lies in knowing what we see through it and what remains beyond it.

Questions that stay with us

• What am I actually seeking when I open a chat window?
• What distinction do I see between feeling understood and proving the respondent has feelings towards me?
• Did my last conversation help me take a step beyond the screen?
• What boundary do I want so the tool continues to serve my relationships and decisions?