Features Best known for its anti-plagiarism software used by tens of thousands of universities and schools around the world, Turnitin builds tools to detect AI-generated text.
Large-scale language models have gained momentum since the commercial release of OpenAI’s GPT-3 in 2020. Several companies are now building their own rival machine learning systems, fueling a new wave of startups building products powered by generative AI. These models behave like generic chatbots. It responds with coherent and persuasive sentences when users enter instructions.
While students are increasingly using AI tools to complete assignments, teachers are just beginning to consider the impact and role of AI tools in teaching. Opinions are divided. Some believe the technology can hone their writing skills, while others see it as cheating. According to Forbes, schools in California, New York, Virginia, and Alabama have students Blocking access to the latest ChatGPT models on public networks.
The education sector is not sure what academic policies should be put in place to regulate the use of AI text generators. Moreover, given the current lack of effective methods for detecting machine-written works, it would be difficult to apply all the rules anyway. Enter Turnitin. Founded in 1998, the US company sells software that looks for signs of plagiarism by calculating how similar a given essay is to the contents of a large database of articles, web pages and books. I’m here.
Turnitin was acquired by media giant Advanced Publications for $1.75 billion in 2019, and its software is used by 15,000 institutions in 140 countries. With over 20 years of experience, Turnitin’s extensive reach in the education field and amassing a vast repository of student publications make it an ideal company to develop academic AI text detectors.
Turnitin has been quietly building software for years since the release of GPT-3, says Chief Product Officer Annie Chechitelli registerThe push to give educators the ability to distinguish between text written by humans and computers is gaining momentum with the launch of its more powerful successor, ChatGPT. As AI continues to advance, universities and schools need more than ever to protect academic integrity.
“Speed is key. I hear from teachers that it gives them something,” says Chechitelli. Turnitin hopes to release the software in the first half of this year. “It’s going to be pretty basic detection at first. Then we’ll scrap subsequent quick releases to create more actionable workflows for teachers.” The plan is for the company to gather data and user feedback. As we do, we plan to make prototypes free for existing customers.
“Initially, we just want to help the industry and help educators step in and feel confident. And getting as much use as possible early on is critical to creating a successful tool.” After that, we decide how to commercialize it,” she said.
AI lighting patterns
The AI-generated text is compelling, but there are signs that reveal the algorithm’s handiwork. The writing is usually mediocre and unoriginal. Tools like ChatGPT regurgitate existing ideas and perspectives and don’t have a clear opinion. Humans may find AI-generated text, but machines are far better at it.
Eric Wang, vice president of AI at Turnitin, says there are clear patterns in the descriptions of AI that computers can detect. He said, “We feel human, [machines write using] radically different mechanism. We are picking the most likely words in the most likely places. It’s a very different way of building a language. [compared] to you and me,” he said register.
“We unconsciously roll our eyes back and forth as we read, moving back and forth between words, paragraphs, and sometimes pages. I may be writing, and I’m thinking about something, a paragraph, a sentence, a chapter; Linked to be written.”
However, ChatGPT does not have this kind of flexibility and can only generate new words based on previous sentences, he explained. Turnitin’s detectors work by predicting the words AI is likely to generate in a given text snippet. “It’s statistically very bland. Humans don’t tend to use high-probability words consistently in high-probability places, but GPT-3 does, so the detector is actually shows that,” he said.
Wang explains that Turnitin’s detector is based on the same architecture as GPT-3 and is a miniature version of the model. “We are in many ways the [say] Fight fire with fire. A detection component is connected instead of a generation component.So you’re reading the language exactly the way GPT-3 reads it, but instead of spitting out more language, it predicts if this sentence will look like [it’s from] GPT-3.”
The company has yet to determine how best to use the tool to present detector results to teachers. “It’s a difficult task. How do you tell the instructor what you want to see in a small space?” Cechteri said. See a percentage that indicates how much of the essay appears to have been written by an AI, or a confidence level that indicates whether the detector’s prediction confidence is low, medium, or high to assess accuracy may be required.
This software is not designed to ban ChatGPT in academia. While students can be discouraged from using these kinds of tools, Turnitin believes its detectors will instead enable teachers and students to trust each other and their technology.
“I think there will be a big shift in how content is created and how we work,” says Wang. “Certainly it extends to the way we learn. We need to think long-term about how we teach. How do we learn in a world where this technology exists? Put the genie back in the bottle.” I don’t think you can. Visibility into the use of these technologies is valuable because they are fundamental building blocks of trust and transparency.” ®