OpenAI has developed a tool to detect AI-generated essays and research papers but has delayed its release due to internal debates and user concerns. The company is weighing the benefits against potential negative impacts on non-native English speakers and overall user experience.
OpenAI has developed a tool to detect if ChatGPT is used to write essays or research papers, but the company has yet to release it due to internal debates and concerns about user reactions. According to a report by WSJ, the tool has the ability to detect AI-written text by ChatGPT with 99.9 per cent accuracy.
Another concern OpenAI pointed to was that the tool might unfairly impact non-native English speakers. Despite the tool’s technical promise, the company is weighing risks and researching alternatives. Some employees argue that the benefits of the technology outweigh these concerns.
Surprisingly, the report suggests that OpenAI’s anti-cheating tool has been ready for a year, but internal discussions have delayed its release. The company is torn between transparency and retaining users. OpenAI even conducted a survey with the users, nearly a third of loyal ChatGPT users were opposed to the technology.
How does this tool work?
The anti-cheating tool uses a watermarking method that subtly changes token selection in text generation. These changes create an invisible pattern detectable by OpenAI’s technology, which can score the likelihood that a document was written by ChatGPT. The watermarking is reportedly 99.9% effective when enough new text is generated.
There are concerns that the watermark could be easily removed through techniques like translation or adding and removing emojis. Moreover, determining who should have access to the detector is a challenge. Limiting access could reduce its usefulness, while broad access could compromise the watermarking technique.
Other companies, like Google, are also developing watermarking tools, such as Google’s SynthID, which is in beta testing. OpenAI prioritises audio and visual watermarking over text due to their significant impact, especially during election years.
Teachers and professors are eager for tools to combat AI misuse in education. According to the report, a survey found that 59 per cent of middle- and high-school teachers believe students have used AI for schoolwork. Despite the need, OpenAI is cautious about the anti-cheating tool’s impact on user experience and fairness.
OpenAI has faced internal debates and conducted surveys to gauge public opinion on the tool. While some data suggests public support, there are still concerns about false accusations and the potential impact on ChatGPT’s performance. OpenAI continues to explore less controversial alternatives and seeks to align with public opinion and potential new laws on AI transparency.
