Overreliance on artificial intelligence in academic research: A study of non-native english-speaking students’ experiences in research proposal writing at a South African university

Abstract

While Generative Artificial Intelligence (AI) tools can improve student learning in Higher Education, over-reliance can be detrimental. This paper argues that using AI in academic writing without reflection hinders the development of the cognitive and problem-solving abilities necessary for authentic research. Direct student experiences provide evidence to support this argument. Fourth-year university students who use AI in their research proposals are examined as a case study to better understand how this detrimental reliance affects their learning and critical thinking abilities. Thematic analysis was performed on qualitative data collected from the module questionnaires with open-ended questions and students' reflective journals, and the findings were used to support the argument that overreliance creates a "performance-comprehension gap" that frustrates authentic learning and knowledge building. Although there are certain advantages to using AI, such as improved grammar and polishing, using AI as a learning support can leave students able to produce a written research product but unable to defend or explain it. Although there are certain benefits to using AI, such as improving grammar and polishing, using AI as a learning assistant results in students producing an academic research product but being unable to defend or explain it. Excessive dependence on AI actively inhibits student thought. It is argued that Higher Education faculty and administration have a responsibility to address the consequences of such overreliance, and that modifications in the pedagogical approach are required to protect students' ability to work independently. This study adds to the literature by giving concrete, qualitative evidence of the actual cognitive hazards associated with AI overreliance, shifting the focus from hypothetical concern to demonstrable pedagogical consequences. It reinforces theoretical comprehension by clearly framing the issue within a constructivist learning paradigm, proposing AI overuse as an example of a scaffold that replaces rather than promotes knowledge building.