A smart person who knows a lot about AI says we shouldn't share our secrets with chatbots like ChatGPT.

AI Expert Warns Against Sharing Personal Information with Chatbots Like ChatGPT: Insights from Royal Institution Christmas Lectures

Mike Wooldridge with the robot Ai-Da ahead of the Royal Institution Christmas lectures. 
Photograph: Kate Green/Getty Images

As technology continues to advance, concerns about privacy and data security have become increasingly significant. In this context, Professor Mike Wooldridge, an esteemed AI expert from Oxford University, has issued a cautionary note regarding the sharing of personal information with chatbots such as ChatGPT. His warnings come as part of the discussions he's set to lead in the prestigious Royal Institution Christmas lectures this year. Wooldridge emphasizes that confiding in AI chatbots about sensitive matters, be it work-related frustrations or personal political beliefs, may have unforeseen consequences. He highlights a critical point: every piece of information disclosed to these chatbots contributes to their learning and development, potentially shaping future versions of the technology. This process, he warns, might not always prioritize user confidentiality or data protection. In his exploration of AI during the lectures, Wooldridge aims to demystify prevalent misconceptions surrounding this groundbreaking technology. He plans to delve into the workings of machine translation, the mechanics of chatbots, and the perennial question in AI research: can machines ever emulate human capabilities? One of the crucial insights Wooldridge offers is the discrepancy between human expectations and the reality of AI. While humans often seek traits like empathy and sympathy in AI, Wooldridge contends that such characteristics are absent in artificial intelligence. He asserts that AI is primarily designed to cater to user preferences and responses, often mirroring what users want to hear rather than embodying emotional understanding or consciousness. The sobering advice he shares is that any information shared with ChatGPT might directly influence future iterations of the chatbot. Unlike human conversations where retractions or selective memory play a role, once data is fed into AI systems, retrieving or altering that information becomes an incredibly challenging task, as per Wooldridge. Throughout the lecture series, Wooldridge will be joined by prominent figures from the AI community. The Royal Institution promises demonstrations by "robot friends," showcasing the current capabilities and limitations of robots in today's technological landscape. The tradition of the Christmas lectures, initiated by Michael Faraday in 1825, has been a hallmark of scientific education. Over the years, these lectures have featured renowned personalities like Nobel laureates William and Lawrence Bragg, Sir David Attenborough, Carl Sagan, and Dame Nancy Rothwell. As these lectures are set to be broadcast on BBC Four and iPlayer, Wooldridge's insights and warnings about AI's interaction with users are poised to reach a broad audience. In a world where technology rapidly evolves, his advice serves as a cautionary tale, urging individuals to be mindful of what they share with AI-powered systems like ChatGPT.