Main page > AI Detection
Please note: The use of AI within teaching and learning is a rapidly developing area. This information reflects our current thinking, but please check back regularly as we will be updating the information on this page as things develop.
Several companies claim to have tools which can detect writing produced by AI. But are they reliable?
Let’s start with the things that we do know:
- The only AI detection service which we should be using is Turnitin’s. Students own their own work but give us permission to upload it to Turnitin as part of registering with the University. We don’t have permission to upload it to other services.
- We don’t know what would happen with work uploaded to other services. Would it become part of the AI model? End up in an essay mill? For that reason we should not use other AI tools with students’ work.
- Asking ChatGPT if a piece of text was produced by AI produces unreliable results.
Does Turnitin’s tool work reliably? The jury is out. Unlike their similarity report, which gives great detail on where work has been matched against, the AI report just gives a score and highlights sections which are alleged to have potentially been produced by AI. It is only intended as an indication that something might be worthy of further investigation rather than anything approaching definitive proof. We have seen plenty of examples of false positives (where work we know to be human produced is flagged as AI) and false negatives (where work we know to be AI is not flagged up) and we know that in mixed work it can struggle to separate the human and the AI produced section accurately.
July 2024 update
Turnitin have released an update which increases the word limit to 30,000 words, and introduces a new facility to indicate where AI generated content has been paraphrased or rewritten by tools like Grammarly or Quillbot.
At this time, we understand that this is not intended to detect where these writing tools have modified original work, but have not been able to confirm this.
Some researchers in the field of AI argue – both theoretically and empirically – that accurate AI detection is impossible. Turnitin claim that their tool has a 1% false positive rate.
Much like contract cheating, detecting AI writing through software is a huge challenge. As a University, we feel that designing assessments to minimise the risk of students being able to complete the task just by using AI is a better strategy than trying to detect where it has happened.
Turnitin’s AI report is useful to have, but it is just one tool available to us. We have some guidance on how to use it covering how to interpret the report and how it can be used within our Academic Misconduct policy.