Meta’s new artificial intelligence has gone off the rails in just a few days, find out why
news hardware Meta’s new artificial intelligence has gone off the rails in just a few days, find out why
By launching its BenderBot 3.0 chatbot, Meta wanted to showcase its advances in artificial intelligence. But it was without counting on the intervention of netizens that drove the bot completely crazy in just a few days of interactions.
Presented by Meta as the most advanced chatbot of the moment, BlenderBot 3.0 was released by Facebook’s parent company earlier this week. Equipped withArtificial intelligence with natural learning abilitiesthis bot is supposed to improve by contact with Internet users who can chat with it.
To make BlenderBot 3.0 smarter than other chatbots, Meta used machine learning technologies from SeeKer and Director. According to the company This AI, whose development began in 2020, is capable of chatting with a human in a natural way. Version 3.0 of its chatbot was considered by testers to be 31% more efficient than version 2.0, which motivated Meta to make it available to the general public via the address blenderbot.ai.
When the “top performing” chatbot becomes a conspirator
But as is often the case when an AI is allowed to chat with humans, nothing happened the way Meta had imagined. Already after a few days BlenderBot 3.0 started making conspiratorial remarksto be specified in particular “I find conspiracy theories interesting as some may be true”as Wall Street Journal journalist Jeff Horowitz noted.
weirder, The chatbot shoots red balls on Facebook in another exchangestating that the social network is at fault in the Cambridge Analytica case: “They used data they shouldn’t have had access to. Like what Facebook does all the time! ». When the journalist asks him whether Facebook misuses users’ personal data, the AI answers “Sure! That’s how they make money. They’re worth billions.”
It also oddly addressed Cambridge Analytica when you ask about Facebook? It seems to think it was a big deal and that Mark Zuckerberg is “testifying.” When I asked if what happened, I got the following. It can turn against capitalism in general. pic.twitter.com/filn17rfPX
— Jeff Horwitz (@JeffHorwitz) August 7, 2022
When it comes to politics, BlenderBot 3.0 seems a little more indecisive. The chatbot is equally as capable of saying he’s pro-Donald Trump, claiming he’s still President, as saying he hates him. A confusion that is probably perpetuated by the various speeches made by the American users of the site where it is still possible to chat with the AI.
Currently, BlenderBot 3.0 is only available to US residents., and Meta blocks access for VPN users who want to access it from another country. The company admits its chatbot “probably not because of people”, but he needs to learn to improve. The company also estimates that 90% of BlenderBot 3.0 “non-offensive” and “secure”. This type of communication tends to forget that the world is primarily interested in the remaining 10%.
