The Right to Be Forgotten, LMM and AI

This post is also available in: עברית (Hebrew)

Privacy on the internet has been an issue for governments worldwide for years, especially when trying to teach the internet, which remembers data forever, how to “forget” data that is harmful, embarrassing, or wrong.

There have recently been efforts to provide help to individuals when damaging information about them constantly resurfaces in web searches.

‘According to TechXplore, the flag scenario was Mario Costeja González who had his past financial troubles continuously turn up in web searches of his name and took Google to court asking to remove private information that was old and no longer relevant. In 2014 the European Court of Justice sided with him and forced search engines like Google to remove links to the hurtful data.

These laws came to be known as the Right to be Forgotten (RTBF) rules.

There have recently been related worries regarding a different industry that can endlessly regurgitate old damaging data. Researchers are warning that large language models (LLMs) risk violation of those RTBF laws.

Experts and researchers argue that even while RTBF focuses on search engines, LLMs cannot be excluded from privacy regulations.

Dawen Zhang, author of the paper “Right to be Forgotten in the Era of Large Language Models: Implications, Challenges, and Solutions” said- “Compared with the indexing approach used by search engines, LLMs store and process information in a completely different way.”

Zhang explains that 60% of training data for models like ChatGPT-3 pulled information from public resources, and as a result “LLMs may memorize personal data, and this data can appear in their output.”

And that is not all- much of generative AI data sources remain essentially unknown to users. This type of risk to privacy violates laws in various other countries as well, and researchers claim these laws should also extend to LLMs.

“The technology has been evolving rapidly, leading to the emergence of new challenges in the field of law,” Zhang said, “but the principle of privacy as a fundamental human right should not be changed, and people’s rights should not be compromised as a result of technological advancements.”