Artificial intelligence research startup OpenAI on Tuesday launched a tool designed to figure out whether text is written by humans or computers.
OpenAI is the company behind ChatGPT, a chatbot that generates text that looks like it was written by a human in response to prompts, which quickly went viral after its release two months ago. Last week Microsoft announced a multibillion-dollar investment in OpenAI and said it would incorporate the startup’s artificial intelligence model into its consumer and business-facing products.
Since ChatGPT debuted in November, some of the largest U.S. school districts, including New York City, have banned the AI chatbots because they fear students will use the text generators to cheat or plagiarize.
Sam Altman, OpenAI’s chief executive, said education has changed in the past with the advent of technologies such as calculators, but he also said the company may have a way to help teachers spot text written by artificial intelligence.
This new tool from OpenAI isn’t perfect; the company says it’s still being developed and optimized, and that errors can occur during use. “In our evaluation of the ‘challenge set’ of English texts, our classifier correctly identified 26 percent of texts written by AI (true positives) as ‘probably written by AI,’ while identifying texts written by humans 9 percent of the text incorrectly labeled as AI-written (false positives),” the OpenAI staff wrote.
The new tool is not reliable at analyzing texts containing fewer than 1,000 characters, and OpenAI does not recommend its use for languages other than English. In addition, text from an AI can be updated slightly, preventing the classifier from correctly determining that it is not a human work.
OpenAI says it is engaging with educators about the capabilities and limitations of ChatGPT and will continue to work on detecting AI-generated text.
Back in 2019, OpenAI made it clear that identifying AI-synthesized text is not an easy task. openAI staff said, “Our work on detecting AI-generated text will continue, and we hope to share improved methods in the future.”
