AI detection has long been a feature of plagiarism checker tools, but software makers now dialling up on ChatGPT to reward original content.
OpenAI’s ChatGPT has sounded alarm bells in education circles about the threat of advanced AI writing tools to upend the authenticity of student essays. But these fears may be overblown.
Essay templates, study notes, and other prepared materials have long been available covering common assignment topics – material that has no doubt been food for ChatGPT, as it made its way through text scraped from the internet.
But academic thought is likely to be safe. And screening tools used by educators will adapt.
Turnitin, a provider of plagiarism checker software and other tools designed to encourage original work, plans to enhance its products in 2023 to be more watchful of ChatGPT misuse.
At the same time, academic guidelines will evolve to bring clarity on what is acceptable and what is not.
And, according to Turnitin CEO Chris Caren, the plagiarism checker firm has been researching and developing technology to recognize the signature of AI-assisted writing for the last two and a half years.
What’s more, there’s reportedly talk of AI writing tool developers, such as ChatGPT creator OpenAI, adding a digital watermark to AI-generated text.
According to New Scientist, which wrote about the development in December 2022, digital signatures “could help teachers and academics spot students who are using text generators such as OpenAI’s GPT to write essays for them.”
But before breathing a sigh of relief, it’s worth considering how quickly workarounds were found for digital rights management (DRM) applied to DVDs, CDs, PDFs, and other digital assets.
And digital watermarks embedded in AI-generated writing would soon become useless once bypasses emerged.
The use of a next-generation plagiarism checker could mitigate against scenarios where students hand in AI-generated writing as their own work.
In fact, educators may soon be able to get their hands on such tools and put the software to the test.
Turnitin is aiming to launch AI-created content detection software in 2023 and will be gathering data and user feedback from its customers based on trials of a prototype version.
Students copying from each other’s work is nothing new.
But detecting passages of AI-generated text inserted into an essay or other student assignment is likely to be a greater test for plagiarism checker software.
In fact, given that users can ask ChatGPT pretty much anything they like, the advanced AI chatbot could be prompted to generate output written in such a way as to evade a plagiarism checker.
But, as Eric Wang, VP of AI at Turnitin explained in an interview with The Register, machines – at the fundamental level – write in a very different way to humans.
Natural language processing is a statistical approach to generating text, and chatbots write by selecting the most probable, or likely, next word in the sequence.
Plus, with its software having been used by 15,000 institutions across 140 countries, Turnitin will be highly motivated not to let AI writing tools such as ChatGPT stand in the way of future profits.
Plagiarism check: A reckoning?Putting machine and handmade products side by side quickly highlights one of the big differences between man and machine.
Machines – computers in the case of ChatGPT – produce output that is perfect. And even if there are defects in production, those marks are likely to be identical across all of the output.
Academics have pointed to perfect spelling, and the absence of grammatical errors as a clear tell that students are using generative AI.
However, penalising students who have correctly spelled all of the words in their essay and understand the rules of grammar, isn’t a practical solution for plagiarism checker tools.
More likely, such software will need to be designed to reward original thought and new ideas. There’s also the question of fact-checking.
ChatGPT is trained on text scraped from the internet, and any errors within that vast amount of training data – of which there are likely to be many – will persist.
It feels like a missed opportunity if plagiarism checker software and other essay authentication tools condemn the use of ChatGPT as blanket cheating.
The use of AI writing aids has become widespread in industry, and many examples of AI-generated text could be sitting in your inbox right now.
Machine-generated text produced in response to human-crafted prompts can provide creative inspiration.
And, with human intervention, advanced chatbots such as ChatGPT can be used to shape much more compelling written arguments, which certainly has educational value.
The question comes down to how AI-writing tools are used – for example, using a calculator in a maths class isn’t seen as cheating. It’s a discussion that will roll on.
And maybe the focus will shift from screening tools, to a re-think on how best to assess students and evaluate their abilities.
Perhaps the power of ChatGPT to generate text capable of passing exams doesn’t represent a failing on the part of the student, but in the nature of the evaluation.
Turning things on their head, forward-thinking teachers and lecturers may even fully embrace AI-writing tools and challenge students to submit the best example of work produced using ChatGPT and its successors.