Invisible Inequality in Peer–AI–Peer Interaction: Rethinking Inclusion in Digital EFL Classrooms
Drawing on Sociocultural Theory, this qualitative study reveals that Peer-AI-Peer interactions in digital EFL classrooms generate invisible epistemic inequalities where expert peers dominate AI-mediated knowledge construction, suggesting that true inclusion requires shifting focus from mere technology access to equitable engagement and the strategic role of expert scaffolding in sustaining collaborative learning.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
In classrooms around the world, a quiet revolution is taking place. Teachers are handing students powerful new tools: artificial intelligence. These digital assistants can brainstorm ideas, fix grammar, and organize thoughts, promising to level the playing field for every learner. The hope is that by giving everyone access to the same smart machine, education becomes fairer. But this optimism often overlooks a subtle truth about how people learn together. Learning is rarely a solo act; it happens in the space between people, where a more experienced student helps a less experienced one. This dynamic, known as peer scaffolding, relies on trust and conversation. When a third party—a computer program—enters this conversation, it changes the rules. The question researchers are now asking is not just whether students can use the tool, but who actually controls the conversation when the tool is present. Does the machine help everyone equally, or does it quietly shift power to the student who already knows the most?
A team of researchers in Vietnam set out to answer this by watching real students write essays with the help of an AI chatbot. They focused on pairs of students: one who was strong in writing, and one who was still learning. They called this setup a "Peer–AI–Peer" interaction. The researchers wanted to see what happened when these two students worked together, asked the AI for help, and then decided what to do with the answers. They recorded the students' conversations, watched their faces for signs of hesitation or confusion, and saved the text of their chats with the computer. They looked at the entire writing process, from the moment students first thought of an idea to the final moment they polished their sentences. The goal was to find "invisible inequality"—differences in how students participated that are hard to see but have a real impact on learning.
What the researchers found was that the presence of the AI did not create a fair triangle of interaction. Instead, it created a lopsided relationship. In almost every group, the stronger student, or the "expert," took control of the device. This student typed the questions, read the AI's answers first, and then decided which parts to share with their partner. The weaker student, or the "novice," often ended up watching or waiting. Even though both students had access to the same technology, the expert became the gatekeeper. They acted as a filter, interpreting the machine's complex suggestions and translating them for their partner. This meant the novice student rarely spoke directly to the AI, and when they did, their input was often overruled or ignored. The inequality was not about who had the tool; it was about who had the authority to decide what the tool meant.
This imbalance grew stronger as the writing task progressed. In the beginning, when students were just brainstorming ideas, the expert used the AI to generate a list of possibilities and then presented them to the novice. The novice could agree or suggest a small change, but the expert held the power to choose the final direction. As the students moved into the middle of writing the essay, the expert used the AI to fix sentences and improve the flow of the text. Again, the expert decided which corrections to keep. By the time the students were finishing their work and checking for errors, the pattern was most extreme. The expert asked the AI for final advice, evaluated the suggestions, and made the last decisions on what to keep or change. The novice student was often reduced to a passive observer, nodding along or simply typing what the expert told them to type.
The study revealed that the most significant problem was not a lack of access, but a lack of "epistemic authority." This is a fancy way of saying that one student lost the right to be seen as a source of knowledge. Even when the novice student had a good idea or a valid concern, the expert's control over the AI made the novice's voice seem less important. The machine's output was treated as the final word, but only if the expert validated it first. If the expert ignored a suggestion from the novice, the novice rarely pushed back. The researchers noted that this happened because the novice trusted the expert, and the expert trusted the machine, creating a chain of authority that left the novice behind. The inequality was invisible because everyone was technically "participating," but the quality of that participation was vastly different.
Interestingly, the researchers found that this unequal dynamic did not always stop learning from happening. In some cases, the expert student seemed to recognize that their partner needed help and used the AI to provide extra support. They would explain the machine's answers in simpler terms or ask the novice to try a specific part of the task. This suggests that the inequality, while unfair in terms of power, could sometimes be used to keep the group moving forward. However, the researchers warned that this is a fragile balance. If the expert is too controlling, the novice stops thinking for themselves and just follows orders. The study suggests that simply giving students a computer does not guarantee a fair classroom. True inclusion requires more than just access to technology; it requires careful attention to who is speaking, who is listening, and who is making the final decisions.
The researchers concluded that for AI to be a truly helpful tool in education, teachers need to change how they organize group work. They suggested that students should take turns controlling the device, so that the novice gets a chance to ask the questions and interpret the answers. Tasks should be designed to force discussion before and after the AI is used, ensuring that the machine supports the conversation rather than replacing it. The study did not claim that AI is bad or that it should be banned. Instead, it showed that without careful guidance, the technology can accidentally reinforce the very gaps it was meant to close. The real challenge is not in the code of the machine, but in the human habits of the classroom. If we want AI to help everyone learn, we must ensure that the power to use it is shared, not hoarded by the student who already knows the most.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.