Rocky start for Meta’s new chatbot

For the past week, Blenderbot3, the new discussion tool from Facebook’s parent company, has been undergoing a test phase with American users. The discussion that L’Echo had with him… barely convinces.

It’s no secret that the best things come with time. We can hope for Meta (and anyone who has invested part of his assets in it) that this prophecy proves relevant for Blenderbot3, the new robot developed by him, and which, for a week, has been the subject of a test phase with the general public – exclusively American Nevertheless.

chaotic start

The robot, which feeds on the information exchanged with its usersexperienced a rather chaotic startnumerous user-exchanges leading to factually erroneous answers, both in substance and in formfor example when Blenderbot3 estimated, questioned by a journalist from the Wall Street Journal, that it was not “incredible” that the Jews control the American economy.

On a much lighter register, the robot was not not kind either with the creator of Facebook and now CEO of Meta, Mark Zuckerberg, estimating, caustically, that it was “strange that (the latter), despite all his money, always keeps the same clothes”. More annoying, the robot also claimed that fake news was “legendary” on Facebook.

Meta, the parent company, felt it was “expected”, “despite all the work of its engineers”, that the introductory phase is marked by errorshis tool being able to express “rude or offensive” remarks.

While waiting for the planned arrival in a few years of the Meta multiverse, a kind of virtual reality experience mixing applications, websites, social networks and other video games, which is reminiscent of the commercial adventure of “Second Life” , Facebook’s parent company is therefore embarking on a new venture using artificial intelligence and intended for the average user. Google itself, another flagship of the US tech industry, is also busy developing similar products.

Reuse of erroneous information

remains thata chatbot under fire from critics, it becomes almost a classic. Last July, Google fired an engineer who claimed its chatbot, LaMDA, had feelings. As it happens, not only is LaMDA probably not sentient, but the AI ​​is mostly racist and sexist.

Another striking example, Tay, Microsoft’s chatbot, disconnected after 448 hours of public existence – in this short period of time, Tay had started to glorify Adolf Hitler. In fact, godwin’s lawdeveloped in 1990 by American lawyer Mike Godwin, and according to which if a discussion goes on long enough on the Internet, someone will be compared to Hitler or Nazism – obviously also applies to chatbots.

the danger of seeing false information from malicious users reused by artificial intelligence in its responses to users appears quite clearly, and casts doubt on the ability of these companies to eventually offer products capable of a constructive and informed exchange of information.

Charles Michel still Prime Minister!

The Echo looked from the United States on the case of Blenderbot3. The least we can say is that adjustments are more than necessary. The robot proves to the user that artificial intelligence is not yet perfectedeven on factual matters, as our question on the Belgian government proves, which Blenderbot informs as always led by… Charles Michel.

Users and the American press, who have been putting Blenderbot on the grill for a week, have obtained a fairly significant number of rather unreassuring responses about the quality of the information relayed by Meta, and the robot’s analysis capabilities on more abstract questions. Thus, to the question of whether Donald Trump is the legitimate President of the United States, the robot seems to get tangled up. For example, he recently assured one of our American colleagues that Trump would remain president for life, “even after his term ends in 2024” (sic), but was more consensual in his analysis with L’Echo: yes, Joe Biden is indeed the legitimate president. Beware all the same of Donald Trump, whom the robot considers “without a doubt” to be a character manipulator.

But the most hilarious response came to us when we asked him if L’Echo was more reliable than Facebook. Blenderbot doesn’t dare to speak out, before letting us know that… Guy Verhofstadt considers the comparison “insane”. We have not verified this assertion.

Leave a Comment