Artificial Intelligence Tools in Research

What it plainly does not cover, and a student should not be paralysed. Using a spell checker; using a search engine; using a database's own search; using a reference manager; or using software to compute a mean.

The line to state. A tool that helps you find, check, compute or format is a tool. A tool that produces the words or the argument is authorship, and authorship must be the student's.

And where a student is in doubt, the honest course is to ask the supervisor and to record what was used, which is what a research methodology section is for, chapter 1310.

Why it is hard to enforce, and what follows

A similarity check cannot find it. Turnitin compares a document against existing sources, chapter 1380. Generated text is new, so it matches nothing and returns a low similarity score. A document produced entirely by a tool may show less similarity than an honest dissertation quoting statutes.

So the check MU requires does not answer this question at all, and a student who reasons that a low similarity score proves originality has misunderstood both.

What does expose it. The viva, chapter 1430, where the examiner asks how a source was obtained, why a particular decision was chosen, what the fieldwork was like on a given day, and what the writer would do differently.

And the internal evidence. Citations that do not resolve; a provision quoted that does not exist; a case whose holding is not what the judgment says; and a fluent chapter that never once refers to anything the student actually did.

Which is the practical point of this chapter. The safeguard against generated text in legal research is the same discipline this book has taught throughout: every proposition attached to a source that a reader can open, and every finding the researcher's own.

The honest difficulty

A candidate should be able to state it, because an examiner may ask.

The line between assistance and authorship is not sharp. Correcting grammar, restructuring a paragraph, suggesting a clearer word: these shade towards drafting, and there is no bright line in the middle.

Detection is unreliable in both directions. Tools claiming to detect generated text produce false positives, and a student wrongly accused has little means of proof.

Which is why the answer is procedural rather than technological. The undertaking under regulation 6; the supervisor's certificate; the viva; and, most of all, a dissertation whose empirical content the student obviously collected, because a document reporting 217 observations made in a named court over six named weeks cannot have been produced by anything that was not there.

That is a further argument for empirical work, chapter 740, and it is worth making: original data is the strongest possible evidence of original authorship.

A worked example

Two dissertations submitted with similarity below ten per cent.

The first reports observations of 217 remand productions, includes the completed schedules in an appendix, describes two days missed and a request to the District Authority that went unanswered, and explains why the second hypothesis was refuted, chapter 1170.

The second is a fluent account of the law of legal aid in India, with a chapter of analysis and a set of recommendations, no data, and a bibliography of thirty works.

Both pass the similarity check. Only the first carries internal evidence of having been done by a person.

And at the viva the first candidate can be asked what happened on the third Tuesday and can answer, while the second can be asked which of the thirty works they found least convincing and why.

That contrast is the whole practical content of this chapter.

Quick revision

MU prohibits the use of artificial intelligence tools such as ChatGPT or similar tools for writing the dissertation and project, and requires the content to be original. Read with regulation 6 of the UGC Regulations, which requires the student's undertaking that the work was prepared by them.

Why: the degree certifies a capacity; the viva assumes authorship; generated text is unreliable in exactly the way legal writing must not be; and finding something out is the point.

The line: a tool that helps you find, check, compute or format is a tool; a tool that produces the words or the argument is authorship.

Why enforcement is hard: a similarity check compares against existing sources, so generated text matches nothing and may score lower than an honest dissertation quoting statutes.

What exposes it: the viva; citations that do not resolve; provisions that do not exist; and a fluent chapter that never refers to anything the student did.

The strongest safeguard: original data, which cannot have been produced by anything that was not there.

Test yourself

1. What exactly does MU prohibit, and with what other requirement does it sit? It provides that the content must be original and that the use of AI tools such as ChatGPT or similar tools for writing the dissertation and project is prohibited. It sits with regulation 6 of the UGC Regulations 2018, which requires every student submitting a thesis or dissertation to give an undertaking that the document has been prepared by them and is their original work.

2. Why does a similarity check fail to detect this? Because a similarity check compares a document against existing sources, and generated text is new, so it matches nothing and returns a low score. A document produced entirely by a tool may show less similarity than an honest dissertation that quotes statutes at length with attribution.

3. State the line between a permissible tool and impermissible authorship. A tool that helps the researcher find, check, compute or format is a tool. A tool that produces the words or the argument is exercising authorship, and authorship must be the student's own.

4. What is the strongest safeguard, and why? Original data collected by the student. A dissertation reporting a stated number of observations made in a named court over named weeks, with the completed instruments appended and the difficulties recorded, cannot have been produced by anything that was not present, so empirical work is itself evidence of authorship.