Back
Automatically translated version. May contain inaccuracies compared to the original.
🫠 ChatGPT independently compiles psychological profiles of users, often without their knowledge, - a study by researchers from Germany, India, and the Netherlands
The researchers analyzed archives of 80 users' ChatGPT conversations and found 2050 records stored in the neural network's memory. In 96% of cases the AI created these records on its own, without a direct request from the user to remember anything.
At the same time, 28% of the records contained personal data subject to European information protection law, and 52% contained psychological characteristics of users: their preferences, habits, interests, and inferred personality traits.
Moreover, ChatGPT is capable not only of remembering what was said but also of drawing its own conclusions. For example, if someone asked for recommendations of bands similar to Nirvana, the neural network recorded that they like Nirvana, even though the user did not state that explicitly.
According to the researchers, chatbots are gradually turning from ordinary tools into interlocutors to whom people entrust personal problems, political views, and concerns. As a result, AI may know much more about a person than typical internet services.
The scientists warn that such personalization carries not only privacy risks but also the danger of reinforcing biases, because if the neural network knows the user's views well and constantly agrees with them, it may only strengthen their misconceptions.
An important caveat: since the data were collected, OpenAI has updated ChatGPT's memory mechanism, so the results do not necessarily fully reflect the operation of the current version of the service.
UNIAN on 🕸Google | 👥 X | 📺 YouTube |🧵 Threads
Telegram publication
