If the AI is only one part of the arrangement, what happens when human interlocutors remain in it too?
Something important disappears when Human:AI is framed only as a replacement question.
Can AI replace the teacher?
Can AI replace the mentor?
Can AI replace the critical friend?
Each version imagines one chair becoming empty and asks whether a machine can sit in it.
Some of the emerging research looks different.
It adds another chair.
Şakire Erbay-Çetinkaya studied thirty-nine pre-service English teachers in Türkiye using a triadic professional-support framework.
The three sources of reflection were:
peer
ChatGPT
faculty advisor
The interesting unit was therefore not one learner in conversation with one AI.
It was a small ecology of reflection.
Different interlocutors could do different work.
That possibility had already appeared in the study by Arefian, Çomoğlu and Dikilitaş.
Their EFL teachers reflected individually with ChatGPT and collaboratively inside a community of practice.
The AI exchange could come before the human one.
Teachers could use ChatGPT to articulate ideas, explore possibilities and arrive at the community conversation better prepared to contribute.
The sequence matters.
AI was not necessarily the destination of reflection.
It could be part of the preparation for another relationship.
This gives complement a more concrete meaning.
A human interlocutor may bring knowledge of the person, the institution, the classroom and the history of earlier decisions.
AI may be available immediately and able to generate several framings quickly.
A peer may recognise what feels normal in the local context.
A faculty advisor may bring disciplinary or pedagogical expertise.
The learner may use all of them differently.
There is no reason to assume that one participant must perform every function.
That sounds attractive.
It also creates a new problem.
Every additional seat changes the room.
Teraif’s classroom study gives us a useful warning.
Students using generative AI as a critical friend / Socratic opponent became more autonomous in some respects, but the classroom also became quieter, with reduced student-student and student-teacher interaction.
Adding one form of dialogue altered another.
That means a hybrid arrangement cannot be judged merely by asking whether the AI conversation was helpful.
We also have to ask what happened around it.
What human conversation disappeared?
What became possible?
Who was approached first?
Which disagreement was taken to a person and which to a machine?
Did AI help somebody arrive at the human conversation with more to say?
Or did its availability make the human conversation easier to avoid?
The third seat can extend a room.
It can also crowd one.
This takes us back, unexpectedly, to the older critical-friendship literature.
That literature kept asking who owned the problem, who authorised the relationship, what trust made possible, and whether support developed the learner’s independence.
AI does not remove those questions.
It changes their arrangement.
A teacher can now ask a machine before asking a colleague.
A student can test an argument privately before exposing it publicly.
A practitioner can generate a countercase without waiting for another person to be available.
Those are meaningful changes.
But the human ecology still matters.
Perhaps more than before.
Because the machine is so available, somebody has to decide when not to use it.
When the problem needs another person’s lived knowledge.
When confidentiality changes what can be entered into a system.
When disagreement needs consequences.
When the person affected by the decision needs a voice.
When a relationship should not be simulated because the real relationship is itself part of the work.
The most promising picture emerging from these studies is therefore not AI replacing the critical friend.
It is a changing arrangement of who or what is invited into the thinking, when, and for what purpose.
That is a different design problem.
And it leaves one final relationship question unresolved.
The AI can respond to us.
We can respond to it.
The conversation can move.
We can learn.
But what kind of reciprocity is actually occurring?
Even in a highly productive Human:AI arrangement, what crosses the boundary in both directions—and what does not?
References
Erbay-Çetinkaya, Ş. (2025) ‘A triadic support framework for reflective teacher identity: Peer, AI, and faculty collaboration within a community of practice’, Teaching and Teacher Education, 168, 105239. doi:10.1016/j.tate.2025.105239. Publisher record.
Arefian, M.H., Çomoğlu, I. and Dikilitaş, K. (2026; first published online 2024) ‘Understanding EFL teachers’ experiences of ChatGPT-driven collaborative reflective practice through a community of practice lens’, Innovation in Language Learning and Teaching, 20(2), pp. 318–333. doi:10.1080/17501229.2024.2412769. Open-access article.
Teraif, Z. (2025) ‘Can AI be friends? An innovative approach to integrating Generative AI as a critical friend in Bahrain Polytechnic’s English foundation programme’, Innovation in Language Learning and Teaching, advance online publication. doi:10.1080/17501229.2025.2551124. Open-access article.