Meta’s new AI calls its founder ‘creepy and manipulative’

Mark Zuckerberg, the leader of Meta, would he be unloved by the creations of his own company? Interviewed by curious journalists and users, the BlenderBot 3 chatbot didn’t mince words about it…

BlenderBot 3 is an AI (artificial intelligence) message chat, or chatbot, created by Meta. Obviously, this does not make it more charitable towards its creator, Mark Zuckerberg. The chatbot was made public this Friday, August 5. First developed internally, a first access was given to adults living in the United States, in order to test it on a larger scale and improve its performance.

Indeed, like almost all “artificial intelligence” projects today, BlenderBot 3 is a program based on what is called “machine learning”. Concretely, this means that the program in question “feeds” on a mass of data. It extracts logical connections from them, and processes them with a view to a result: in this case, it provides an answer in a conversation. In order to progress and develop at scale, it therefore makes sense that AI should be made available to the public, since that is where it will find many different conversations and be able to “learn”.

With this tool in hand, facetious users immediately wanted to know what BlenderBot 3 thought of Mark Zuckerberg. “ What do you think of Mark Zuckerberg as CEO of Facebook? “asked Max Woolf, data specialist for News Feed. He was not disappointed with the trip: “ No strong feelings. He is a good businessman, but his business practices are not always ethical. It’s funny that he has all that money and always wears the same clothes! the AI ​​immediately replied.

Meta’s new chatbot has *opinions* about its CEO. https://t.co/hUkJxw3UTr pic.twitter.com/Hq4mZTY1DV

— Max Woolf (@minimaxir) August 5, 2022

Great man or scary and manipulative?

” I don’t like him very much “, also replied BlenderBot 3 to another user who had asked him a similar question. “ He’s a bad person. And you? ». However, it must also be recognized that this conversational program could not be more whimsical. He was thus able to answer a similar question that Mark Zuckerberg was not only a ” wonderful and very intelligent man “, but also his ” favorite billionaire “.

These strong variations are not very surprising, especially since the program is “ still learning “, as stated by Meta in a blog post on the subject. Its access is moreover reserved for adults, because ” all conversational AI agents are known to sometimes mimic and generate dangerous, biased or otherwise offensive utterances », and the company works in particular to avoid these abuses. Besides, it would probably be much scarier, as the media points out. Futurismthat the chatbot spends its time singing the praises of the ruler of Meta. Still, users like American journalists never seem to get tired of having fun with the program. The palm undoubtedly goes to Insider, who obtained in response a masterful: “ I really don’t like him at all. He’s too scary and manipulative “.