One in 10 essays checked by Turnitin’s new detection software has been found to contain substantial amounts of writing likely to have been generated by artificial intelligence.

The firm has released figures from the first three months of the detector’s operation, during which time it was used on more than 65 million papers.

Of these, 2.1 million – 3.3 per cent – were flagged as having at least 80 per cent AI writing present. Nearly 6.7 million – 10.3 per cent – had more than 20 per cent AI writing present.

Campus resource: A simple hack to ChatGPT-proof assignments using Google Drive

Turnitin said the findings showed that “generative AI has made its way into classrooms, however, whether this is acceptable or not is determined by educators themselves”.

����Vlog

The edtech giant rushed out its new tool as fears grew over the likelihood that students were more likely to cheat on assessments thanks to the launch of more accessible, easy-to-use AI chatbots such as ChatGPT.

It claims to be able to detect AI writing 97 per cent of the time, with a “very low” false positive rate, although critics say this has not been independently verified.

����Vlog

Annie Chechitelli, chief product officer at Turnitin, said the AI detector was being “widely used”, with 98 per cent of the institutions it works with enabling the feature alongside other tools such as the firm’s anti-plagiarism software.

“Sharing usage and indication rates is one way that we can help improve understanding of the presence and use of generative AI in their teaching and learning practices,” Ms Chechitelli said.

“Given the urgency expressed by educators about these challenges and the public’s interest in AI text creation and AI text detection, we are committed to sharing these insights so that we can all begin to understand the trends that are currently shaping education.”

Patti West-Smith, Turnitin’s senior director of customer engagement, said she saw the purpose of the tool as helping educators and students start “meaningful conversations” about the “appropriate use of writing tools, proper citation and original thinking”.

����Vlog

Turnitin pointed out that having a very high proportion of likely AI writing in a document does not necessarily indicate misconduct because some institutions are instructing students to utilise the technology in their work.

Daniel Sokol, a barrister and expert in academic integrity who has flagged the dangers of falsely accusing students of using AI based on detection software, said most students using ChatGPT and other programs make some changes to minimise the risk of detection.

“The Turnitin figures suggest that more than one in 10 students have used AI and included large sections of AI-generated text in their assignments. Yet, have one in 10 students been accused of academic misconduct? I doubt it. I suspect many institutions are turning a blind eye to the problem as they would otherwise be overwhelmed with cases.”

Dr Sokol said universities that do want to permit AI-use should encourage students to be transparent, for example by including the question posed to the AI in an appendix.

����Vlog

Where AI is not permitted, the fact that an essay might be only 20 per cent written by a chatbot “is as academically unacceptable as an essay that is 20 per cent written by a private tutor”, Dr Sokol added.

“In both cases, the student has asked a third party to do a significant chunk of the work for them. Whether that third party is a human or an AI is irrelevant.”

����Vlog

tom.williams@timeshighereducation.com