Chief Justice of India Surya Kant warned law schools against finding students guilty of misconduct solely because AI-detection software flagged their work. Speaking on 21 November at the Commonwealth Legal Education Association’s First Annual International Conference on Legal Education in Hyderabad, he said, “No student should therefore be held guilty of misconduct on the word of a machine alone,” LiveLaw reported.
Key points
- Kant said an AI detector’s result alone should not establish student misconduct.
- He cited a 2023 study in which detectors wrongly flagged more than three in five essays by non-native English speakers.
- He urged law schools to set AI-use rules before assessments and test whether students can explain their work.
Patterns study and 91 TOEFL essays
Kant cited the 2023 study “GPT Detectors Are Biased Against Non-Native English Writers”, published in Patterns. He said seven widely used detectors incorrectly classified more than three in every five essays by non-native English speakers as machine-generated. The study tested 91 TOEFL essays and 88 essays by students in the United States, Lawbeat reported.
Kant warned that a student writing in English as a second or third language could be accused of cheating on the strength of a score they could not meaningfully challenge. He cautioned institutions against treating detection software as proof that a student had submitted machine-generated work.
Singapore’s approach and law school assessments
Kant said an outright ban on AI tools would be “neither practicable nor wise”. He cited the Supreme Court of Singapore’s approach, under which court users may use generative AI but remain responsible for everything they file. For law schools, he favoured rules given before an assessment begins, specifying which uses are allowed, which require disclosure and which would surrender the reasoning the exercise is meant to test.
Kant also proposed supplementing written assignments with oral explanations, a defence of the legal authorities used or exercises in identifying errors in machine-generated answers, Bar and Bench reported.