ChatGPT was tipped to cause widespread cheating. Here’s what … – ABC News

As teachers met before the start of the 2023 school year, there was one big topic of conversation: ChatGPT.

Education departments banned students from accessing the artificial intelligence (AI) writing tool, which could produce essays, complete homework, and cheat on take-home assignments.

Some experts said schools would be swamped by a wave of cheating.

And then? Well, school continued.

The expected wave never broke. Or if it did, it was difficult to detect.

Loading…

Many teachers said they strongly suspected students were cheating, but it was hard to tell for sure. 

Meanwhile, some schools went in the opposite direction, embracing the new AI tools. Principals said students needed to learn how to use a technology that would probably define their futures.

But there was one perspective missing in all this: that of the students themselves.

As the school year drew to a close, we spoke to Year 11 and 12 students about how they actually ended up using generative AI tools.

Overall, these are stories from a historic moment, and an insight into the future of education. This is the first year where high school students could easily access high quality AI writing tools.

Here’s what they had to say.

‘The chatbot could smash it out in seconds’

Some were initially curious, but also cautious.

Some used it once then stopped. Others kept going.

Some got caught, but many didn’t.

For Eric, who asked to remain anonymous, the arrival of ChatGPT in the summer of 2022–23 was a mixed blessing.

To stop students using AI to cheat on take-home assignments, his school switched to more in-class assessments.

But Eric, who has ADHD, struggles to concentrate in exams.

“That sort of crippled my chances at doing great in school,” he says.

But the new AI writing tool had its uses. In term one, he experimented with cheating, using ChatGPT to write a take-home geography assignment (though it didn’t count towards his HSC).

He wasn’t caught and got a decent grade.

Next, he gave ChatGPT his homework. 

“It’s just meaningless, monotonous work. And, you know, the chatbot could just smash it out in seconds.”

Eric estimates that, over the course of 2023, ChatGPT wrote most of his homework.

“So around 60 per cent of my homework was written by ChatGPT,” he says.

AI cheating creates divide among classmates

Students at other schools told similar stories.

Chrysoula, a Year 11 student in Sydney, initially used ChatGPT “very often to complete homework I deemed just tedious”.

A lot of her classmates were doing the same, she says.

“Any time someone read a good answer out in class discussions, there was someone leaning over whispering ‘Chat[GPT]?’

“Everyone doubted the authenticity of everyone’s answers.”

But as the year drew on, Chrysoula found her “critical and analytical thinking was slightly impaired”.

Worried about AI harming her learning abilities, Chrysoula blocked herself from accessing the ChatGPT website.

The students who are still cheating are easy to spot, she says.

They often waste time in class and leave coursework to complete at home, when they can access AI tools. Their essays overuse words that ChatGPT favours, like “profound”, or “metaphors for things like tapestries”.

“They even hand in entire assessments written entirely by AI.”

Phil, a Year 12 student at a different high school, also sees a divide that’s formed among his classmates, based on how much each student uses AI to do their coursework.

His school allows students to use ChatGPT for ideas and research but not to directly write assessments.

But many students still cheat on take-home assignments, Phil says. He’s worried they “aren’t learning anything” and their poor performance in the HSC exams will ultimately hurt his ATAR.

Harry, a Year 11 student at another school, uses ChatGPT for some homework, but not for assignments.

His reason? The AI’s answers aren’t good enough.

“When you want to get the top marks, I’d say that’s when you do want to use your own brain and you do want to critically think.”

How common is AI cheating in schools?

Many of the high school students we spoke with weren’t using AI to cheat, even though they could get away with it.

Whether this is representative of students in general is hard to say, as there’s very little reliable data on cheating rates.

But at least one survey suggests that rates of AI cheating among students may be lower than generally assumed.

In mid-2023, the Centre for Democracy and Technology asked US high school students how much they used generative AI tools to do their coursework, and then compared this with the estimates of teachers and parents.

Kristin Woelfel, a co-author of the report, says teachers and parents consistently overestimate how many students use AI to cheat.

“While 58 per cent of students report having used generative AI in the past year, only 19 per cent of that group reports using it to write and submit papers,” she says.

She says the survey data doesn’t support the inflammatory predictions about cheating made at the start of the year.

But Kane Murdoch, head of academic integrity at Macquarie University, says that even if the rate of AI cheating is low now, it’s likely to go up.

He believes students will gain confidence, and learn how to use AI to automate more of their coursework.

“It could be 2023 was the year they dipped their toe in the water, but 2024 and moving ahead you’ll get increasingly large body parts into the pool.

“And soon enough they’ll be in the deep end.”

Banning AI cheating hasn’t worked

Many of the students we spoke to say teachers have little power to stop them from using AI tools for homework or assignments.

The past year proved blocking access to AI tools, as well as detecting AI-written coursework, was ineffective.

Students described numerous ways of evading the blocks on accessing ChatGPT on school computers, or through a school’s Wi-Fi network.

They also told how they copied their ChatGPT-written answers into other AI tools, designed to confuse the schools’ AI-detection software.

“Nearly every AI detector I’ve come across is inaccurate,” Chrysoula says.

Mr Murdoch agrees.

He says educators were reluctant to rely on flawed plagiarism-detection tools to accuse a student of cheating.

“As an investigator [of cheating] I’m unwilling to accept the detectors word on it,” he says.

ChatGPT-maker OpenAI has warned that there is no reliable way for educators to work out if students are using ChatGPT.

Several major universities in the US have turned off AI detection software, citing concerns over accuracy and a fear of falsely accusing students of using AI to cheat.

Mr Murdoch says Australian universities are also wary of relying on detection software.

But they disagree over what to do next. Some believe that better detection is the answer, while others are pushing for a change to the way learning is assessed.

“Programmatic” assessment, such as interviews and practical exams, may be one answer.

“It would mean we don’t assess as much, and what we do assess we can actually hang our hat on,” Mr Murdoch says.

“It’s more like turning the ship in a very different direction rather than a slight course change.

“It’s a much more difficult thing to grapple with.”

Schools may go from banning bots to letting them teach

While the impact of AI on students has won most of the public attention, some education experts say the bigger story of 2023 may be how this technology changes the work of teaching.

Over the course of the year, schools have broadly gone from banning the technology to cautiously embracing it.

In February, state and federal education ministers slapped a year-long ban on student use of AI at state- and government-run schools.

Then, in October, the ministers agreed to lift the ban from term one next year.

The announcement brought public schools more in line with the approach of private and Catholic schools.

Federal education minister Jason Clare said at the time that “ChatGPT was not going away”.

“We’ve got to learn how to use it,” he said.

“Kids are using it right across the country … we’re playing catch-up.”

But some observers are worried this new embrace of AI will replace in-person, teacher-led instruction.

The education industry is now promoting an idea of “platformisation”, says Lucinda McKnight, a digital literacy expert at Deakin University.

“This is the idea that bots deliver education. The teacher shortage is solved by generative AI.”

Custom-built chatbots, personalised for each student and loaded with the relevant curriculum, would teach, encourage and assess. Teachers would be experts in teaching, rather than the subject being taught.

But robo-classrooms like this have their drawbacks, Dr McKnight says.

Interacting with (human) teachers plays an important part in a child’s emotional and social development.

Macquarie University’s Kane Murdoch says education institutions made a series of kneejerk reactions and false starts, such as the ChatGPT ban, in response to emergent AI technology in 2023.

Next year will see them make big changes, he predicts.

He’s concerned that AI cheating will ultimately devalue educational qualifications.

“We can’t expect this to go away. It is a game changer — it is existential,” he says.

“If the public lose faith in what we do, we lose.”

Student accounts suggest that, although not all want to cheat when they can, many are happy to automate tasks they see as rote learning rather than true tests of knowledge.

For Eric, who used ChatGPT to do most of his homework this year, the AI tool provided a shortcut through an already flawed system.

“I would have had such a worse time this year if ChatGPT wasn’t prominent,” he says.

“I think it would be very difficult for you to find a student that was sitting the HSC right now that hasn’t used it for something.”

Listen to the full story of how students used AI in 2023 and subscribe to RN Science Friction.