ChatGTP and other AI
While AI text generation is nothing new, the recently released ChatGTP is pretty sophisticated in providing responses to a variety of prompts. In my department, we generated responses to prompts we use and put some results through the turnitin similarity detector. The AI response had almost no match (5%), and the content that did match was pretty generic phrasing related to the topic.
Since these AI platforms generate content in response to specific prompts, they content is dynamic -- sort of. Is there a way to account for this in the turnitin.com platform? What suggestions -- short of in class, paper-pencil responding -- do y'all have for combatting these new tools?
7 replies
-
Chris Begeman I would be interested in what you and others are seeing at your institutions in terms of AI-generated text. This is certainly a fast-moving area that we at Turnitin are exploring the impact of globally. This blog post from Eric Wang our Head of AI might be a good starting point.
-
Our experience (from a highly ranked UK university) is that the use of AI is increasing rapidly and Turnitin is entirely unable to detect it - ironically Turnitin often yields very low or even zero similarity scores to text that has been AI generated or modified, which is how we have seen this rising. Some online tools like GPTZero can spot this but they can't be deployed at scale at the moment and even if they could, I would fully expect AI bots to adapt over time to make detection difficult or impossible. Our approach - and I suspect many universities will follow suit - is to move away from online assessments altogether and go back to face to face, in-person handwritten exams. This of course means Turnitin's business model will collapse (the risk of plagiarism essentially goes away with in person exams) so if they want to survive they need to act very quickly on this. In my institution (where I'm chair of exams) I expect us to move away from online exams and coursework within the next academic year, as the availability of AI basically undermines all sense of academic integrity. I certainly hope Turnitin will come up with something more impressive than the above blog post, which seems to suggest they have no idea and no plan of how to deal with this.