AI Can Translate Everything in Virtual Exchange, but Should It?

Virtual Exchange has quietly become one of the most influential strategies for internationalizing higher education. By connecting students across continents through online collaboration, it promises intercultural competence without the financial, environmental, and logistical costs of physical mobility. Yet a new perspective article by Judit Háhn of the University of Jyväskylä, published in Frontiers in Computer Science, argues that this celebrated inclusiveness hides an uncomfortable truth: most Virtual Exchange projects run on English as a lingua franca, and that monolingual habit may be reproducing precisely the hierarchies the method claims to dissolve. As generative artificial intelligence and real-time translation tools flood into these classrooms, Háhn contends that the technology could either democratize participation or quietly sterilize the very friction through which intercultural learning actually happens.

The core problem Háhn identifies is structural rather than technical. Universities frequently promote Virtual Exchange as inherently inclusive, but participation is shaped by unequal access to linguistic resources. Drawing on the logic of neoliberal internationalization, many initiatives prioritize individual competence development and employability outcomes while overlooking how language constrains meaningful engagement. Within these settings, proficiency in English functions as symbolic capital, in the sense theorized by sociologist Pierre Bourdieu: it positions some participants as more legitimate interlocutors than others. Differences in fluency ripple outward into students’ emotions, confidence, and willingness to contribute, which means that a pedagogy designed to open doors can end up locking some of them.

Háhn’s own empirical research with Virtual Exchange students illustrates how this plays out in practice. Students with weaker English skills consistently reported anxiety, hesitation, and reduced agency during collaboration. Many internalized deficit narratives, describing themselves as inferior in relation to imagined fluent peers, a dynamic Háhn calls the linguistically deficient self. Others described feeling othered by more fluent classmates. Tellingly, the more fluent participants sometimes adopted paternalistic roles, expressing concern for weaker speakers while implicitly reinforcing hierarchies of competence. The insight is that linguistic inequality is not simply a matter of proficiency levels; it is relationally and discursively constructed through how participants perceive, evaluate, and position themselves and others in relation to language.

To frame an alternative, the paper assembles several theoretical strands. Critical Virtual Exchange repositions the practice as socially accountable work grounded in decolonial awareness, asking students to examine how technology, language, and power intersect to shape participation. Relational learning shifts the goal away from extracting skills from the Other and toward engaging with participants’ histories, positionalities, and situated knowledges, prioritizing reciprocity and mutuality over employability. Translanguaging theory treats learners’ full linguistic repertoires as legitimate resources for meaning-making rather than obstacles to a shared code. Together these perspectives contrast sharply with mainstream, instrumentalist designs in which interaction can become superficial as participants avoid confrontation to maintain what Háhn calls a false sense of harmony.

Into this landscape arrive AI tools: real-time translation, automated captioning, AI-assisted paraphrasing, and generative text support. Háhn uses the term AI-supported multilingualism to describe their use within Virtual Exchange to support interaction among participants with different language backgrounds. The potential affordances are concrete. Real-time translation can support participation in live discussions; AI-assisted paraphrasing can help students refine ideas before speaking; translation tools can let students speak and write in their own languages while the system facilitates mutual understanding; generative models can scaffold multimodal, multilingual projects and presentations. Crucially, these affordances remain pedagogical possibilities rather than established practices, and Háhn stresses that further empirical research is needed to evaluate how they shape participation.

There is early evidence that the vision is plausible. One recent study of a generative-AI-supported Virtual Exchange examined students who used AI image generation and translation tools to co-construct visual representations of cultural themes, with the technology conceptualized as a mediational tool supporting translanguaging. In that configuration, AI becomes one resource among many in a broader semiotic repertoire. Háhn suggests AI can strengthen strategies that Virtual Exchange participants already deploy, such as images, audio, and code-switching, by adding scaffolding that redistributes communicative responsibility and shifts the emphasis from linguistic accuracy to meaning-making.

But the paper’s sharpest argument concerns what AI cannot do: transform underlying language ideologies or power relations simply by mediating them. Because AI systems are trained on uneven datasets, they often reproduce dominant language hierarchies, privileging English in particular. Háhn warns against technological solutionism, the assumption that communicative inequalities can be resolved primarily through technical mediation, and against epistemic and discursive homogenization, the tendency of AI-generated outputs to normalize standardized academic discourse and reduce diversity of rhetorical and linguistic expression. Add ethical concerns about data privacy, surveillance, and algorithmic bias, and the picture becomes one of a technology that expands access while potentially reinforcing the norms it appears to bypass.

The conceptual heart of the article is what Háhn calls the paradox of communicative friction. Virtual Exchange depends, at least in part, on moments of misunderstanding, repair, and negotiation across linguistic and cultural difference; that friction is a constitutive element of relational learning. Yet AI systems are explicitly designed to minimize friction through translation, prediction, and linguistic smoothing. Recent discussions cited in the paper suggest AI-mediated communication may reduce the need for the linguistic negotiation and interactional repair work that traditionally underpin intercultural learning, meaning AI does not simply enhance Virtual Exchange but potentially reconfigures its epistemic and pedagogical foundations. Háhn is careful to note that valuing friction does not mean preserving linguistic barriers or treating difficulty as inherently desirable; the real question is which barriers should be removed and which forms of interactional work remain educationally valuable.

The ethical dimension deepens the warning. Drawing on the philosophy of Emmanuel Levinas, Háhn conceptualizes the Virtual Exchange encounter as an ethical event in which the self is constituted through responsibility to the other, an obligation that precedes knowledge, judgment, or linguistic competence. If AI is used to create seamless, frictionless interaction, the encounter risks being sanitized, its unpredictability and vulnerability stripped away. Ethical interculturality, in this view, requires learning to respond to alterity without imposing normative, native-speakerist frameworks, and the goal is not to eliminate the difficulty of understanding the Other but to remain responsible for the Other even, and especially, in the face of misunderstanding. Technological solutionism threatens to reduce the other to a data point or a translated output.

The paper closes with a constructive program rather than a rejection of AI. Háhn proposes three guiding principles for AI-informed Virtual Exchange design: mediated communication should be treated as an object of reflection, with AI tools explicitly acknowledged, examined, and problematized within activities; relational pedagogy should take priority over communicative efficiency, privileging dialogue, negotiation, and responsiveness over task completion; and multilingual epistemologies should be recognized, with learners’ full linguistic repertoires treated as legitimate and productive resources for meaning-making. In practice, tasks might incorporate AI-assisted multilingual dialogue while requiring students to reflect on how translation shapes meaning and participation. The upshot is a reframing of Virtual Exchange itself: not a site of frictionless communication, but a relational and ethically accountable educational practice in which AI serves as a scaffold for multilingual reflection rather than a substitute for engagement, and in which the future of the field depends on resisting the assumption that communicative friction is a problem to be eliminated rather than the very medium of learning.

Subject of Research: AI-supported multilingualism and linguistic justice in virtual exchange

Article Title: From monolingual virtual exchange to AI-supported multilingualism: a critical perspective

Article References: Háhn, J. (2026). From monolingual virtual exchange to AI-supported multilingualism: a critical perspective. Frontiers in Computer Science, 8, Article 1898906. https://doi.org/10.3389/fcomp.2026.1898906

Image Credits: AI Generated

DOI: 10.3389/fcomp.2026.1898906

Keywords: virtual exchange, generative AI, multilingualism, translanguaging, linguistic justice, higher education, intercultural learning, machine translation, critical pedagogy, language ideology, English as lingua franca, educational technology