AI in higher education: GenAI overuse may deskill future graduates

Generative artificial intelligence, or genAI, is forcing universities into a high-stakes rethink of how they teach students, test learning and prepare graduates for the labour market, as staff weigh the technology’s educational benefits against risks to academic integrity and core skills, according to a new study published in AI & Society.

The study, titled “Pedagogical conundrum, assessment anxiety, and employability dilemma: a qualitative exploration of staff perspectives on the great GenAI quandary in higher education,” is based on interviews and focus groups with staff at a Russell Group university in England The research finds that GenAI has created a complex institutional challenge marked by a pedagogical conundrum, assessment anxiety and an employability dilemma.

GenAI opens new learning pathways but raises ethical concerns

The study finds that university staff do not view GenAI as a technology that can be simply accepted or rejected. Instead, they see it as a disruptive force already reshaping the daily practices of higher education. Tools such as ChatGPT, Grammarly, QuillBot and AI-powered translation systems are being used to support writing, summarise texts, explain difficult concepts, generate feedback, translate material and assist with routine academic work.

Staff recognised clear pedagogical opportunities. GenAI can make learning more accessible by simplifying complex ideas, helping students engage with unfamiliar material and supporting those who need help with academic writing. In fields such as medicine, environmental sciences and other applied disciplines, staff saw potential for AI tools to help students understand technical concepts and build confidence in handling dense subject matter.

The most productive use of GenAI, according to the study, comes when students are required to critically evaluate AI outputs rather than treat them as finished answers. This can include checking whether generated material is accurate, identifying missing context, reviewing sources, spotting bias and comparing AI responses with their own understanding. Used this way, GenAI can help students develop stronger critical thinking and improve their ability to judge information.

The technology also offers wider access to knowledge across language barriers. AI-powered translation can help students and educators work with material that would otherwise remain inaccessible. In some disciplines, this could expand the range of perspectives students encounter and support more inclusive learning.

However, the same technology also introduces serious risks. Staff warned that GenAI can produce information without enough context, accuracy or transparency. In high-stakes disciplines, unreliable AI-generated material could weaken foundational knowledge. In creative and language-based subjects, staff raised concerns that AI-assisted writing could promote standardised styles, reduce originality and narrow students’ creative development.

The study highlights concerns about hidden bias in AI outputs. Since GenAI tools are trained on large datasets that may contain racial, gender, social and cultural bias, their use in education can reproduce problematic assumptions. This creates a risk that students may absorb distorted or incomplete knowledge while believing it to be neutral or authoritative.

Staff also pointed to the environmental and financial costs of large-scale AI systems. These concerns add another layer to the debate over whether GenAI should be embedded widely into university teaching without clearer ethical guidance. The study argues that GenAI is not merely a study aid. It is becoming a force that shapes how students learn, think, write and understand knowledge.

Assessment anxiety grows as AI challenges academic integrity

Assessment emerges as the study’s most pressing concern. Staff feared that students could use GenAI in ways that evade detection, submitting work shaped heavily by AI rather than their own knowledge, effort or ability. This is crucial because university degrees depend on trust in assessment. When institutions award qualifications, they certify that students have met required academic standards. If GenAI allows students to pass assignments without doing the expected learning, the credibility of grades and degrees is placed under pressure.

The study finds that assessment anxiety is not limited to fears about cheating. It reflects a wider concern that GenAI may weaken the connection between assessment outcomes and actual student ability. Staff worried that students could bypass learning, produce acceptable coursework and still receive credit for skills they have not developed.

Existing detection systems offer limited reassurance. Traditional plagiarism tools were designed to identify copied text, not AI-generated responses. AI detection tools also raise concerns because they can wrongly flag human writing or fail to detect machine-generated material. This makes them unreliable as a basis for academic misconduct decisions.

Consequently, staff often rely on indirect signs, including fabricated references, unusual writing style, inconsistent tone, unexpected fluency or a mismatch between a student’s known ability and submitted work. These signs may raise suspicion, but they rarely provide firm evidence. The study shows that this leaves staff feeling responsible for protecting academic standards while lacking dependable tools to enforce them.

The findings point to a growing need for assessment redesign. Staff discussed more authentic and varied assessment formats, including oral examinations, presentations, viva-style questioning, staged assignments, in-person exams and tasks linked to real-world scenarios. These approaches could make it harder for students to outsource learning to GenAI while also providing a fuller picture of student ability.

However, the study warns that assessment reform cannot be based only on blocking AI use. As GenAI tools improve, assessment methods designed around today’s risks may quickly become outdated. A narrow focus on detection and prevention could also damage trust between students and staff.

The research suggests that universities need to rethink assessment more deeply. The goal should be to protect academic integrity while recognising that GenAI will remain part of education and work. That means designing assessments that test understanding, judgment, application and communication, rather than relying too heavily on tasks that AI can easily complete.

Graduate employability becomes a new GenAI dilemma

The third major issue is employability. Staff acknowledged that GenAI skills are becoming important in the labour market. As employers adopt AI tools for writing, coding, analysis, research, customer support, data handling and decision-making, graduates will need to know how to use these systems effectively and responsibly. This creates pressure on universities to prepare students for AI-enabled workplaces.

In some applied fields, staff were more willing to consider curriculum changes because GenAI is already affecting professional practice. For example, if AI tools can generate computer code or handle routine data tasks, students may need more training in how to specify problems, evaluate outputs, correct errors and make higher-level decisions.

The study shows that GenAI could change what counts as graduate competence. Students may need to develop not only subject knowledge but also the ability to judge AI-generated outputs, identify risks, use prompts responsibly, recognise bias and decide when human expertise is required. However, this creates a dilemma. The same tools that may improve employability can also weaken the development of core skills if students rely on them too heavily. Staff raised concerns that overuse of GenAI could reduce students’ ability to write clearly, manage time, communicate effectively, think independently and build deep subject knowledge.

These concerns are particularly strong in relation to language proficiency and creative work. AI-powered translation and writing support can help students participate more fully in academic life. But if such tools are used to replace rather than support learning, they may hide gaps in skills that degrees are expected to certify.

The study argues that universities must move beyond basic AI literacy. Teaching students how to use AI efficiently is not enough. What is needed is broader GenAI education that covers ethical use, critical evaluation, accuracy checks, bias awareness, disciplinary expectations and responsible professional practice.

The education must also be discipline-sensitive. GenAI will not affect medicine, engineering, social sciences, humanities and creative subjects in the same way. Some fields may use AI heavily for routine tasks, while others may face greater risks to creativity, language development or professional judgment. A single institutional policy is unlikely to capture these differences.