Words including “imperfect”, “scrutinize” and “apparent” are too advanced for a Year 12 student to use, according to artificial intelligence, and are considered evidence of plagiarism.

Here are other words AI program ChatGPT thought were too good for one high-achieving Cambridge High School pupil, leading the school to fail her assessment.

“Vibrant counterparts; non-pastel location; better perceive; effectively conveying the intensity of; utilizing, emphasizing and realization [for their American spellings]; detrimental; induce; eventual; employed; and online presence.”

Despite being an excellence-level student, the girl’s teacher flagged her work as potentially plagiarised and fed it through ChatGPT.

ChatGPT, released in November last year, is an artificial intelligence chatbot that generates text.

“The AI checker suggested parts of her work were written by AI,” Cambridge High School assistant principal Joanna Bartch told the girl’s mother.

“[The excerpts from her work] indicate to us that AI software has been used by [student] in preparing her NCEA Internal Assessment.”

But the girl was adamant the work was her own, and her mother described the use of AI by teachers to detect plagiarism as “Russian roulette” of infallibility.

Last week, the Herald reported the girl’s story, along with that of another student of Pukekohe High School. Countless concerned parents and educators have since shared similar examples.

“I am sure that you will agree that the words ... are not suspicious or remarkable words. My daughter has an excellent and well-developed vocabulary. She is mature, articulate and a prolific reader,” her mother said.

The phrases were from the girl’s NCEA Level 2 English assessment: “Analyse significant aspects of an extract from a film.”

The girl at the centre of the major cheating allegation chose to analyse an episode of Black Mirror - a British television show about “people grappling with the manipulative effects of cutting-edge technology”, according to a description online.

Experts have said the availability of AI tools has made “the opportunity for cheating and all manner of misconduct ... seemingly endless”.

Educational technologist Francis Valintine told the Herald following our first report about the allegations last week that it would be dangerous if education authorities began accusing people of using AI to cheat when it wasn’t always accurate.

“There is no 100 per cent certainty on any tool whatsoever and nobody is claiming to have that magic tool,” Valintine said.

The girl’s mother told the Herald: “Obviously [the school] were realising they had to deal with this artificial intelligence, which undoubtedly some children will be using - my daughter knows people that have.

“But what bothers me is that ChatGPT is so unreliable and yet it’s the screening method [for work], so, you know, undoubtedly they’re going to call kids out incorrectly,” she said.

“How is she going to feel every time she hands something in now that your robot is going to accuse her of being artificially intelligent?”

NZQA advised schools about the use of artificial intelligence in February - noting the arrival of ChatGPT was ringing alarm bells in the education sector.

“At worst, the chatbot can produce quality essays, reports, etc on any topic, which might escape detection by regular plagiarism checks and be passed off as the student’s work,” NZQA said.

NZQA lists three programmes schools could use to detect plagiarism - AI Writing Check, AICheatCheck and Turnitin - which have already been used in schools and universities.

But NZQA also said it could be “nigh on impossible” to stop students from using AI tech.

NZQA deputy chief executive of assessment Jann Marshall said they don’t tell schools exactly how they should check whether work was authentic.

“While AI detection tools are developing quickly and can be useful, they are not infallible,” Marshall said.

Cambridge High School principal Greg Thornton said the school’s procedures were in line with official guidance.

“We review each student’s work using teacher judgment. Only the scripts that are not considered to potentially be a student’s own work are further investigated,” he said.

Thornton said teacher judgment involved looking at students’ previous work and their “overall performance” in class.

He wouldn’t comment on the specific case involving the student but did outline the school’s process for dealing with potential plagiarism.

“It is not appropriate to discuss [the mother’s] concerns at this time as I am currently engaging with her in accordance with our concerns and complaints procedure,” Thornton said.

The girl’s mother doubted how accurate any teacher’s judgement of her daughter’s work was: “The professional judgement was unreliable as her English teacher is a first-year teacher [with] 11 weeks’ experience.

“No doubt this is the first assessment [the teacher] has marked,” she said.

“[The head of English] also does not think the words are likely for a Year 12 student. However, given [she] then confirmed she has never met or had a conversation with my daughter, or referred to any of her previous work, I don’t consider [her] to hold any relevant judgement.

“Also ... ChatGPT is now widely known and recognised as unreliable, dangerous, inconsistent, full of falsehoods. And that is exactly what I have been told from within your own teaching staff,” she said.

When the Herald asked ChatGPT - directly - if it could detect whether work was plagiarised, the computer programme offered this response: “I can certainly help detect plagiarism in a piece of writing.”

“However,” it continued, “it’s important to note that I don’t have access to every possible source of information, so my analysis might not be exhaustive. Additionally, while I can flag passages that match other sources, it’s ultimately up to a human evaluator to determine whether or not the similarities constitute plagiarism.”

Principal investigator of the Research on Academic Integrity in New Zealand (Rainz) Project and University of Auckland associate professor Jason Stephens has researched cheating for two decades.

Stephens said the problem was artificial intelligence tools had “arrived overnight”.

“This isn’t just a technological development, it’s a technological revolution,” the said.

“In education, this means rethinking not only how we teach and assess learning, but what we are teaching and assessing for.”