‘Crisis meetings’ as chatbot writes university essays

A top Australian university has failed the first student for using AI, sparking a warning that students will have to be made to hand-write assignments to prevent chatbot cheating.

A university has failed a student caught using a chatbot to cheat. Photo iStock

A top Australian university has failed the first student for using artificial intelligence to write an essay, an AI professor has revealed.

Computing engineer Toby Walsh, Scientia Professor of Artificial Intelligence at the University of NSW, said schools and universities will have to revert to handwritten essays and assignments in class to prevent chatbot cheating.

“Banning access to websites is totally useless because kids are smart enough to work around it – they can use a VPN (virtual private network), ‘’ he said on Sunday

“You’ve got to put them in a room with no (internet) access, with a pen and paper and no technology.

“We can’t give students take-home lessons anymore.’’

Professor Walsh said the release of OpenAI’s controversial ChatGPT – a powerful chatbot released in November – had triggered “crisis meetings’’ at his own university.

“We need to consider how we better assess things,’’ he said.

“I know of a student who failed their course because they cheated with it.’’

Professor Toby Walsh says teachers and lecturers might be suspicious and question students ‘if the writing is too good, or too middle of the road’.

Professor Walsh said the student had confessed to using the chatbot to write an essay, after a lecturer suspected the writing had been generated by AI.

He said ChatGPT could evade traditional plagiarism checks because “every time you run it, you get a different answer’’.

He said teachers and lecturers might be suspicious and question students “if the writing is too good, or too middle of the road’’.

“Especially if they’re not a native English speaker, and the quality of their writing jumps dramatically, you’d call them in,’’ he said.

NSW, Queensland and Tasmania have blocked the use of ChatGPT on school internet networks, to prevent students using it to churn out well-written and lengthy responses within minutes, without the need for research.

Professor Walsh – who wrote the book Machines Behaving Badly: the Morality of AI – said students must be taught to separate fact from fiction, and not simply swallow responses produced by AI.

He warned that students might easily mistake fiction for fact if they rely on chatbots to answer questions and write essays and assignments.

He said ChatGPT had “gone for quantity over quality’’ of information sourced from “all of Wikipedia, all of Reddit and a large chunk of social media’’.

“It’s not understanding what it’s saying,’’ he said.

“Not only does it make stuff up, but it sounds plausible.

“It’s fine if it’s producing advertising and business letters, but school kids could be learning things that are false.’’

Professor Walsh said AI would make truth “a much more fungible idea’’ and warned that responses could be manipulated by lobby groups or conspiracy theorists interacting with a chatbot en masse.

He said that in 2016 Microsoft created a Twitter bot called Tay, designed to be a chatty 19-year-old girl.

“It was (learning) using human feedback, and within 24 hours it turned into a racist, xenophobic, Nazi misogynist,’’ he said.

“They had to turn it off.’’

OpenAI’s website states that ChatGPT does not produce any information updated after 2021, and relies on a “thumbs up or down’’ response from users to know if an item is correct.

“This means that it is completely unaware of current events, trends, or anything that happened after its training,’’ it states.

“It will not be able to respond appropriately to questions or topics that require up-to-date knowledge or information.

“ChatGPT has no external capabilities and cannot complete lookups.

“This means that it cannot access the internet, search engines, databases, or any other sources of information outside of its own model.

“It cannot verify facts, provide references, or perform calculations or translations.

“It can only generate responses based on its own internal knowledge and logic.’’

Asked to write a biography of Australia’s prime minister, ChatGPT said it required a name, and then described Anthony Albanese as “leading the Labor Party in the opposition trying to regain power and form government in the next election’’.