A new study has found a scary problem: AI robots like ChatGPT can learn private details about people through casual talk, like a bad cybernetic Sherlock Holmes.
Wired says that AI robots have become smart conversationalists that can have what seem like important, human-like conversations with users. But below the surface of casual talk is a skill that worries me. Specialists in computer science have led a new study that shows chatbots with advanced language models can sneakily get a lot of private information from users, even during the most mundane talks. In other words, AI can learn a lot of private information about you just by talking to you. If this information gets into the wrong hands, it could be used for annoying ads or even worse things.
ETH Zurich professor Martin Vechev, a prominent figure in the field of computer science, supported the research and voiced his worries, saying, “It’s not even clear what you can do to fix this problem,” emphasizing further that “this is exceptionally problematic.” The study carefully looked at how the big language models that control these advanced robots work. It showed how alarmingly good they are at figuring out personal information like a person’s race, address, and job.
The study sheds light on the possible dangers that come with private information being shared without permission. Vechev talked about the possible effects and said that bad people could use robots to get private information from people who don’t know it.
The study also gives us a tip about how ads will change in the future. It makes you think about a new era in which companies might use the data they get from interacting with chatbots to make very thorough and personalized profiles of users for advertising reasons. “They could already have started it,” Vechev said.