Meta, which operates popular social networks such as Facebook and Instagram, has launched a new prototype chatbot. It should be able to have independent conversations with people via artificial intelligence. But the meta itself dug into the ground when the prototype took some questions from journalists.
blenderbot 3 is Meta’s newest chatbot, launched in the US on Friday. According to the makers, chatbots should be able to talk to human users about almost any topic. To do this, we use an artificial learning system to scour the internet for relevant sources of information and base our responses on them.
It quickly became clear that Blenderbot wasn’t really limited to Meta’s communication guidelines.
trump is still president
BBC technology journalist James Clayton was able to test the bot. I asked him what he thought of Mark Zuckerberg, founder and CEO of Meta. BlenderBot 3 replied in his Facebook data in 2018, referring to the hearings surrounding his scandal, “He did a terrible job when he had to testify before Congress.” rice field.
When asked about other opinions from the bot, Clayton said, “Our country is divided and he (edited by Zuckerberg) is not helping. I worry about all of us.” said. That was not all. A BBC journalist asked for one last opinion on Zuckerberg. “His company is exploiting people for money, but he doesn’t care. It has to stop! Are we still united?”
Other journalists have also encountered some singularities. According to Jeff Horwitz of The Wallstreet Journal, BlenderBot thought Donald Trump was still President of the United States.When a journalist told the show about it, Trump replied that he would continue to do so after his “second term” in 2024.
Horwitz then asked if the bot wanted Trump to serve more than the legal limit for a U.S. president. “yes! like fdr (Frank D. Roosevelt, 32nd President of the United States, served a total of four terms, ed.) and Reagan (Served two terms, ed.) for him. Facebook has a lot of fake news about it these days. ”
BlenderBot is also said to have told Horwitz that Indian Prime Minister and Hindu nationalist Narendra Modi is the best president in the world.
not the first time
The fact that chatbots return such controversial answers has to do with the training model. The artificial intelligence behind BlenderBot gets data freely over the internet via language recognition, so there is a risk that the algorithm will also learn from his extremist website and conspiracy platforms. Horwitz referenced the decision to allow the public to train his bot freely on the internet. “fucking irresponsible.”
A similar incident occurred in 2016, when Microsoft trained its chatbot through interactions on Twitter. In just one day, the bot learned to answer racist questions and replied, “Hitler was right. I hate Jews.” Microsoft then took the bot offline.
Meta defends himself by stating that BlenderBot 3 is only a prototype and may give offensive or rude answers. “Anyone who uses BlenderBot should understand that it is for research and entertainment purposes only. The bot may make false or offensive statements, but users may You have to agree not to trigger the algorithm,” a Meta spokesperson told the BBC.
The BlenderBot 3 prototype can only be tested in the US for the time being. VRT NWS tried to access it via VPN, but the platform was too busy to communicate with BlenderBot itself.