CNN
—
Two months after OpenAI unnerved some educators with the public release of ChatGPT, an AI chatbot that can help students and professionals generate shockingly compelling essays, the company is unveiling a new tool to help educators adapt.
OpenAI announced this on Tuesday a new feature called an “AI Text Classification” that allows users to check whether an essay is written by a human or by AI. But even OpenAI admits it’s “imperfect”.
The tool, which works on English AI generated text, is powered by a machine learning system that takes input and assigns it to different categories. In this case, after pasting a piece of text, such as a school essay, into the new tool, it returns one of five possible outcomes, ranging from “probably generated by AI” to “very unlikely.”
Lama Ahmad, policy research director at OpenAI, told CNN educators have been asking for such a ChatGPT feature, but warns that it “should be taken with a grain of salt.”
“We really don’t recommend using this tool in isolation, because we know it can and sometimes will be wrong — just like using AI for any kind of assessment purposes,” said Ahmad. “We emphasize how important it is to keep a human being informed… and that it is just one data point among many others.”
Ahmad notes that some teachers have referred to past examples of students’ work and writing style to gauge whether it was written by the student. While the new tool may provide another point of reference, Ahmad said, “Teachers need to be very careful about how they measure up in academic unfairness decisions.”
Since ChatGPT became available in late November, it has been used to generate original essays, stories, and lyrics in response to user queries. It produced summaries of research papers that fooled some scientists. In fact, it recently passed law exams in four courses at the University of Minnesota, another exam at the University of Pennsylvania Wharton School of Business, and a US medical licensing exam.
In the process, it has raised the alarm among some educators. Public schools in New York City and Seattle have already banned students and teachers from using ChatGPT on the district’s networks and devices. Some teachers are now rethinking their assignments with remarkable speed in response to ChatGPT, even though it remains unclear how widespread the tool’s use is among students and how detrimental it could actually be to learning.
OpenAI now joins a small but growing list of efforts to help educators detect when a written work has been generated by ChatGPT. Some companies, such as Turnitin, are actively working on ChatGPT plagiarism detection tools that can help educators identify when assignments are being written by the tool. Meanwhile, Princeton student Edward Tuan told CNN that more than 95,000 people have already tried out the beta version of his proprietary ChatGPT detection feature, called ZeroGPT.
Jan Leike – a leader in the OpenAI tuning team, which ensures the AI tool is aligned with human values – listed several reasons why detecting plagiarism through ChatGPT can be challenging. For example, people can edit text to avoid being identified by the tool. It will also be “best at identifying text that is very similar to the kind of text we trained it on.”
also the company said it’s impossible to determine whether predictable text — such as a list of the first 1,000 prime numbers — was written by AI or a human because the correct answer is always the same, according to a blog post from the company. The classifier is also “very unreliable” on short texts of less than 1,000 characters.
During a demo with CNN ahead of Tuesday’s launch, ChatGPT successfully labeled several types of work. For example, an excerpt from the book ‘Peter Pan’ was deemed ‘unlikely’ to have been generated by AI. However, in the company’s blog post, OpenAI said it mislabeled human-written text as AI-written text 5% of the time.
Despite the possibility of false positives, Leike said the company wants to use the tool to spark conversations about AI literacy and potentially dissuade people from claiming that AI-written text is human-made. He said the decision to release the new feature also stems from the debate over whether people have a right to know if they are interacting with AI.
“This question is much bigger than what we’re doing here; society as a whole has to grapple with that question,” he said.
OpenAI said it encourages the general public to share their feedback on the AI check feature. Ahmad said the company continues to talk to primary school teachers and the like at the collegiate level and beyond, such as Harvard University and the Stanford Design School.
The company sees its role as “an educator for the educators,” Ahmad said, in that OpenAI aims to make them more “aware of the technologies and what they can be used for and what they shouldn’t be used for.”
“We’re not educators ourselves — we’re very aware of that — so our intent is to help teachers use these models effectively in and outside the classroom,” Ahmad said. “That means giving them the language to talk about it, helping them understand the possibilities and limitations, and secondly, equipping students through them to navigate the complexities that AI is already introducing to the world.”