University students in four Latin American countries describe generative artificial intelligence (GenAI) in strikingly dual terms — as a silent tutor that relieves academic pressure and as a tool that blurs the line between legitimate help and cheating, according to a study published in Frontiers in Education.
The research, led by Elizabeth-Guadalupe Rojas-Estrada of Nebrija University in Madrid, used the Zaltman Metaphor Elicitation Technique (ZMET), a qualitative method in which participants select and discuss images to surface deep associations. Twenty students from Argentina, Chile, Mexico and Peru took part, each choosing seven images and then describing what those images meant for their use of GenAI in university work.
The study found that participants constructed meaning around GenAI through several recurring metaphorical patterns. One common framing was that of an “other” — a silent companion or tutor to which students attributed a kind of agency, even though the researchers note that AI does not possess human-like mental states. Another was that of a pragmatic cognitive extension, valued primarily for efficiency: speeding up information retrieval, summarising long texts, correcting language and helping overcome writer’s block.
These representations, the authors write, point to a reconfiguration of the relationship between students and knowledge production. Traditional boundaries between technological assistance, collaboration and authorship become blurred. On an emotional level, some accounts depicted GenAI as a buffer against academic overload, offering relief. Yet that relief coexisted with ethical vigilance and unease about “crossing a line” in academic honesty. Students described a process of negotiating intellectual responsibility and deciding what counts as legitimate use of generative systems in educational settings.
The study also highlighted perceived gaps in institutional guidance. Participants used metaphors of misalignment to describe how university policies and teaching practices have not kept pace with the rapid, student-led adoption of GenAI. Some accounts framed GenAI as filling a void left by insufficient academic support from institutions. The authors argue that these findings underscore the importance of pedagogical mediation — clearer approaches to teaching and assessment in a context where GenAI is already part of university life.
According to the study, students’ perceptions are not simply divided between acceptance and rejection. They vary with experience, confidence and the type of academic task. Previous research cited in the paper notes that social science students tend to show higher concern about AI than those in engineering or education, a difference the authors link more to a lack of technical understanding than to epistemological worries about the impoverishment of interpretation or argumentation.
The study’s authors caution that the sample of 20 students is small and drawn from a specific regional context, so the findings are not generalisable to all university students. The ZMET method is designed to surface deep, qualitative insights rather than statistically representative data. The images participants selected are treated as prompts that expand the discursive repertoire available in interviews, not as direct evidence of hidden mental states.
For university students worldwide, the study offers a reminder that the debate over GenAI in higher education is not just about detection software or plagiarism policies. It is about how students themselves navigate a new landscape where the boundaries of authorship, trust and intellectual responsibility are being renegotiated in real time, often without clear institutional signposts.