AI and Academic Integrity

AI and academic integrity is the set of ethical, policy and pedagogical questions raised when students use generative AI tools to produce coursework. It challenges traditional definitions of plagiarism and authorship, since a model can draft essays, solve problem sets and write code on demand, and it has pushed educational institutions toward new rules on disclosure, detection and assessment design.^[1]

Background

Concerns about machine-written text predate large language models — plagiarism detectors have long flagged essay mills and paraphrasing tools — but the public release of ChatGPT in November 2022 moved the issue from edge case to mainstream. Within months, universities worldwide issued emergency bans, then reversed them in favour of guidelines permitting disclosed use, recognising that prohibition was unenforceable and that professional practice was itself changing.^[1]

Two strands of debate emerged. The first is detection: commercial tools such as Turnitin's AI writing indicator estimate the probability that text was machine-generated. Their reliability is contested. A 2023 Stanford study of seven detectors found an average false-positive rate of 61.22% on TOEFL essays written by non-native English speakers without any AI assistance — meaning the technology most likely to wrongly accuse precisely the students least equipped to defend themselves.^[2]

The second is assessment: if take-home written work can no longer be trusted as evidence of learning, institutions have moved toward oral examinations and vivas, supervised in-class writing, process portfolios that show drafts, and oral defence of AI-assisted work. Related questions concern copyright and attribution — whether AI assistance must be cited — and research integrity, where fabricated references and AI-generated peer review have entered the scholarly record.^[3]

Key Concepts

Institutional responses cluster into three models: prohibition, still used in some primary and secondary schools and in high-stakes examinations; conditional permitted use, where students may use AI but must declare how, for which tasks and with which tool; and AI-integrated assessment, where using AI well is itself assessed. Most accreditation-adjacent bodies have converged on disclosure plus literacy: teaching students what the tools get wrong — hallucinated citations, confident errors — is treated as more durable than policing.^[3] Note also that AI detectors are not evidence on their own: vendors themselves advise against adverse action based solely on a detector score.^[1]

Applications and Impact

Beyond essays, the integrity question reaches code assignments, laboratory reports, business school case analyses and admissions essays. Institutions are responding by investing in authentic assessment, staff training and student-facing AI literacy modules, while publishers and research funders introduce generative-AI disclosure policies. The longer-term shift being debated is from assessing finished products to assessing learning processes.^[3]

>Key Takeaways

• Generative AI broke the old plagiarism model: text can be machine-made on demand, so disclosure and assessment design replaced blanket bans in most institutions.
• AI detectors are unreliable and biased — a 2023 Stanford study measured a 61.22% false-positive rate on non-native English writing.
• Institutions are shifting toward oral defence, supervised writing and process portfolios as evidence of learning.
• Malaysia's response runs through institutional guidelines, MQA programme standards and AI literacy modules, not a national prohibition.
• Detector output alone should never be the basis for a misconduct finding.

See Also

• AI Literacy
• AI and Copyright
• AI in Malaysian Education
• AI Watermarking
• Hallucination

References