IA: disappointed by the biography delivered by ChatGPT, a deputy seizes the CNIL

The deputy of the presidential majority Éric Bothorel announced that he had seized the CNIL on Wednesday after noting gross errors on his profile presented by ChatGPT.

A real enthusiast of new technologies, Éric Bothorel decided to titillate ChatGPT by asking him to establish his biography and draw up his profile, he who is a deputy of the 5^e constituency of Côtes-d’Armor since May 6, 2017. OpenAI’s chatbot has delivered very amazing answers over several attempts, with many errors and approximations. Consequently, the former elected socialist now affiliated with Renaissance has decided to seize the CNIL.

A biography made in ChatGPT “a bit” approximate

” And the myth of ChatGPT collapsed… wrote Éric Bothorel in a tweet on April 11, posting several screenshots of the AI’s responses to its prompts (instructions or requests). Several notable errors can indeed be identified.

If he is indeed a politician, Éric Bothorel was not born on November 20, 1961 in Morlaix nor on January 2, 1975 in Guingamp as ChatGPT could indicate, but on October 20, 1966 in Paimpol, in the Côtes- of Armor. You can easily find this information both on the National Assembly website and on the MP’s Wikipedia page.

However, ChatGPT failed, and the list of errors here is not exhaustive. The conversational robot lent him an imaginary career as mayor of Saint-Brieuc since 2014. The next day, Wednesday April 12, the elected official decided to file a complaint with the CNIL, which is much more important than we think. believes it.

ChatGPT says itself not to be 100% reliable

The futility of the initial request that led to the complaint raises several issues. First, it is good to remember that ChatGPT is only a text generator, and that in this sense, it is not intended to be a search engine that does not know the error. ChatGPT tells us this himself, when we ask him what he thinks of the information he gives us on certain biographies:

As a language model, I don’t have the ability to know or know if the information I provide is wrong or not. I am programmed to generate responses based on the training data I have received, which is based on publicly available online texts. However, it is possible for errors or inaccuracies to creep into the information I provide, as I am a computer algorithm, not a human being who can understand complex nuances and contexts. That’s why it’s important to always verify the information I provide with reputable sources and to exercise critical thinking when reading any kind of information online.

ChatGPT may well be an editorial tool with certain and promising usefulness, but today it is not intended to provide information as a media would. The problem is elsewhere.

Can ChatGPT fall under GDPR?

This problem is that of the personal data processed (and/or misinterpreted by ChatGPT). The question is indeed to know if the conversational tool must be the object of a framework, of a regulation with regard to the information which it absorbs and diffuses, and especially, if it must be regarded as a tool in front of ensure the processing of personal data.

As Guillaume Champeau (former Qwant and familiar with issues related to personal data) rightly reminds us, the CNIL will have to deliver its response to the inaccuracy of the information provided in the results, which can be linked to the article 5d GDPR. He tells us that ” personal data must be […] accurate and, where necessary, kept up to date “. The text adds that ” all reasonable steps must be taken to ensure that personal data which is inaccurate […] be erased or rectified without delay “.

Not sure if ChatGPT still grants reply rights, but you never know! Still, even if the conversational robot should not be considered encyclopaedic, considering it as a tool processing personal data would create a hell of a mess, in a context where the regulations on AI are already quite vague.

Sources: Eric Bothorel on TwitterClubic