Meta, artificial intelligence and the risks of sharing user data


Institutional Communication Service

26 June 2024

In recent weeks, media outlets across Europe have expressed concerns about Meta, the parent company of Facebook and Instagram, wanting to use the personal data of its users in Europe to train its artificial intelligence. Professor Andrea Emilio Rizzoli, director of the Dalle Molle Institute for Artificial Intelligence (IDSIA USI-SUPSI), tackled the issue on the RSI radio broadcast "SEIDISERA".

In response to significant pressure from various sources, Meta has decided to postpone the programme's launch scheduled for 26 June. However, the topic will continue to be discussed for an extended period. "Essentially, Meta has been using this type of data for years for profiling purposes," says Andrea Emilio Rizzoli, Director of the Dalle Molle Institute for Artificial Intelligence (IDSIA USI-SUPSI).
"The novelty is that the US-based company would now like to exploit the same data to "train" its generative artificial intelligence engines. Meta's interest in this kind of operation is based on the fact that the data in question is of good quality, and therefore extremely valuable with a view to improving AI programmes. With the potential shortage of good-quality data that can enhance the performance of these algorithms in the near future, Meta's decision doesn't surprise me. In fact, I believe it is essential for the industry to have access to this data".

Meta has been disappointed by the delays caused, in large part, by the unfavourable notice issued by the Irish Data Protection Commission. Nevertheless, the US company is determined to continue improving what it calls "innovative technology." The challenge, in the context of an operation that would undoubtedly enhance the quality of the AI product, lies in safeguarding users' privacy. "The algorithm assimilates data and then retains it. As if it 'eats' them and then stores them in its stomach", notes Rizzoli. "It would not be simple, but convincing the algorithm to retrieve data in full, if necessary, using targeted requests would also be possible. This poses significant risks to the security and privacy of user data. Currently, it's unclear whether it's possible to prevent this scenario in any way".

The complete interview with Andrea Emilio Rizzoli, conducted by RSI journalist Laura Dick, can be accessed by clicking this link: