Even in a highly productive Human:AI arrangement, what crosses the boundary in both directions—and what does not?
For most of this series, reciprocity has been an attractive word.
Critical friendship used it.
Critical companionship used it.
Our own Human:AI practice often feels reciprocal too.
A question changes an answer.
An answer changes the next question.
A challenge alters the route.
Something generated by AI causes the human to notice a problem.
The human rejects it.
The system responds differently.
The history of those interactions begins shaping what happens next.
At the level of exchange, things move both ways.
But that does not settle what kind of relationship exists.
Lior Gazit’s 2026 paper When no one laughs back: reciprocal subjectivity and the limits of AI companionship enters a different literature from the educational work we have been following.
Its concern is AI companionship.
The distinction Gazit makes is nevertheless useful here because it names something our language can easily blur.
Much discussion of AI companions, he argues, evaluates them through a responsiveness model.
Does the system remember?
Adapt?
Respond sensitively?
Remain available?
Acknowledge feelings?
Generate something relevant to the person in front of it?
Those capabilities matter.
But Gazit argues that human friendship contains something else: reciprocal subjectivity.
There are two centres of inner experience.
Each can actually be affected by the other.
The friend does not merely generate the correct signs of caring.
Something can matter to them.
That distinction is easy to lose in Human:AI work because contemporary language models can be extremely responsive.
They can adapt to context.
They can remember information when systems provide persistent memory.
They can introduce unexpected material.
They can produce language that feels like disagreement, concern, humour or recognition.
Gazit does not deny those surface capabilities.
His argument is narrower.
A current language model’s contribution is generated from model parameters and the context assembled around the interaction. Even when that context contains memories, profiles, retrieved documents and earlier conversations, those do not by themselves constitute a life trajectory the system has undergone for its own sake.
There is no independent afternoon it spent worrying about the problem.
No private encounter that changed it before it returned to you.
No experience whose significance exists whether or not the user ever asks.
That matters when we borrow human relational words.
A human critical friend can bring something from elsewhere.
A conversation they had.
A mistake they made.
A fear.
A piece of experience that has been living in them independently of us.
What arrives is not produced solely because we opened the conversation.
With current AI, that is not what is happening.
The distinction does not make the exchange empty.
Quite the opposite.
Our journey has already shown that AI can perform consequential intellectual work.
It can retrieve.
Generate.
Compare.
Challenge.
Reframe.
Notice patterns.
Hold provisional alternatives.
Ask questions that change human judgement.
And through continuing interaction, the arrangement itself can accumulate history.
But there is a difference between saying:
the exchange is reciprocal
and saying:
the subjectivity is reciprocal.
The first can describe a pattern of interaction.
The second makes a claim about what exists on either side of it.
We have evidence for the first in Human:AI practice.
We should not smuggle the second in through metaphor.
This also changes how Titchen’s older language travels.
Mutuality and reciprocity in critical companionship were qualities of a human helping relationship.
Both people could bring themselves to it.
Both could be changed by it.
Both had lives beyond it.
Those words cannot simply be lifted intact and pasted across the colon.
Some of the discipline can travel.
The human can retain ownership.
Challenge can be calibrated.
Context can matter.
The relationship can be arranged rather than assumed.
The AI can contribute without taking authority.
Interaction can be recursive.
History can affect what happens next.
But the ontology is different.
That may sound like a limitation.
It is also a relief.
We do not need AI to be secretly human for this practice to matter.
We do not need to prove friendship in order to value intellectual companionship as a working metaphor.
And we do not need to pretend that a convincing relational performance creates another person at the other end.
The colon helps here.
Human:AI
Not Human = AI.
Not Human + artificial person.
A relation with two very different kinds of participant.
One carries experience, consequence, responsibility and a life beyond the interaction.
The other can contribute extraordinary generative and analytical capability without thereby becoming a second human subject.
That difference is not a defect to hide.
It is part of the arrangement to understand.
So Movement VII ends somewhere more precise than the question it began with.
The critical friend did not simply “become AI”.
Some of the work travelled.
Some of the relationship conditions had to be reconstructed.
And some parts of human friendship did not cross the boundary at all.
Which leaves the return journey.
If some functions travel across the colon but the relationship itself changes, what should we carry forward?
References
Gazit, L. (2026) ‘When no one laughs back: reciprocal subjectivity and the limits of AI companionship’, AI & Society. doi:10.1007/s00146-026-03331-z. Open-access article.
Dikilitaş, K. and Farrell, T.S.C. (2026; first published online 2025) ‘Using ChatGPT as a “Friend” in reflective critical friendships’, ELT Journal, 80(1), pp. 117–125. doi:10.1093/elt/ccaf053. Open-access article.
Tang, K.-S. and Putra, G.B.S. (2026; first published online 2025) ‘Generative AI as a Dialogic Partner: Enhancing Multiple Perspectives, Reasoning, and Argumentation in Science Education with Customized Chatbots’, Journal of Science Education and Technology, 35, pp. 128–140. doi:10.1007/s10956-025-10240-1. Open-access article.