What if challenge, dialogue and alternative perspective can be designed without calling the relationship friendship?
That question leads out of the critical-friendship literature altogether.
Kok-Sing Tang and Gde Buana Sandila Putra begin somewhere else.
Their concern is dialogic learning in science education.
In Generative AI as a Dialogic Partner, they designed a customized chatbot called Dialogic Science Teacher.
Its job was not to deliver the correct answer as efficiently as possible.
It was built to keep the conversation moving.
The chatbot used reverse prompting and Socratic questioning. Students were asked to choose positions, explain their reasoning, encounter alternative perspectives and respond.
Twenty-one students aged sixteen and seventeen used the system while revising science topics in two Indonesian high schools.
The researchers then examined the chatlogs.
Among the characteristics they identified were:
perspective-taking
reasoning
arguing
creative thinking
That list feels familiar by now.
Not because Tang and Putra are secretly writing about critical friendship.
They are not.
The route is different.
But their design reaches some of the same practical territory: another presence in the conversation pushes the learner to articulate, examine and sometimes revise a position.
That matters because it prevents this series becoming trapped by its own vocabulary.
A useful Human:AI thinking relationship does not have to descend from critical friendship.
Other traditions have been asking how dialogue, opposition and multiple perspectives can make thinking more visible.
One especially useful move in Tang and Putra’s paper concerns voice.
They do not treat the chatbot as another human mind entering the classroom.
Drawing on the idea of heteroglossia, they describe GenAI outputs as drawing together multiple human voices embedded in training data and shaped again through the user’s prompt.
The AI “voice” is therefore not presented as an independent human-like voice.
It is a generated mixture.
That does not prevent dialogue.
It changes what dialogue means.
The learner still has something to respond to.
The generated answer may contain a counter-position the learner had not considered.
The interaction can still make reasoning explicit.
But the epistemic authority does not have to belong to the chatbot.
Tang and Putra explicitly argue for GenAI as dialogic partner rather than definitive knowledge provider.
That distinction may be more important than the label attached to the system.
Zainab Teraif arrives at a related idea from classroom practice.
In a Bahrain Polytechnic English foundation programme, seventeen students used generative AI during class activities over four weeks.
Teraif describes instructing learners to use the AI as a Socratic opponent: a dialogue partner that challenges them to justify thinking, locate gaps and refine arguments.
She also calls this a form of critical friendship.
Again the labels overlap.
But the classroom evidence introduces a useful warning.
The AI conversations were not simply added on top of an unchanged classroom.
Teraif reports quieter classrooms and reduced student-student and student-teacher interaction.
Students also appeared to adapt over time and become more autonomous in using the tools.
Both can be true.
An AI conversation can create opportunities for individual challenge while changing the amount and form of human conversation around it.
That is not a minor implementation detail.
It changes the ecology of learning.
A dialogue with AI can make thinking more active while making the room between people quieter.
The point is not that one is necessarily good and the other bad.
It is that introducing a new interlocutor changes more than the interaction visible on the screen.
This is where the second river meets the first.
Critical friendship asked what another person can do to support and challenge learning without taking it over.
Dialogic AI asks how a system can be configured to provoke reasoning rather than merely answer.
They are not the same tradition.
But both arrive at a problem of how help is arranged.
And that makes the next question difficult to avoid.
The Dialogic Science Teacher behaved differently because it had been designed differently.
Teraif’s students were explicitly told to use AI as a Socratic opponent.
The relationship did not simply appear because an AI model was present.
If the behaviour changes when the role, prompts and design change, where exactly does the “critical friend” reside?
References
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.
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.