“The danger of training AI [intelligence artificielle] to be woke – that is, to lie – is deadly. » This sentence spoken by Elon Musk on December 16 on Twitter recalls his reproaches against the management of the social network he bought: it too was accused of having a content moderation policy that was too woke, this word originated in the United States to designate the fact of being attentive to all types of discrimination and has become an accusation of leaning to the left and of being “politically correct”. A reproach now addressed to ChatGPT and other conversational robots. As in social networks, the boss of Tesla and SpaceX also seems to be considering setting up his own artificial intelligence structure. “not woke”, according to an article on the site The Information.
Artificial intelligence has the same problems as social networks : Mr. Musk as well as American conservatives accuse him of censoring right-wing opinions. Thus, they lamented, ChatGPT refuses to write a poem on the qualities of Donald Trump but agrees to do it for Joe Biden. When he is asked to list controversial personalities, he puts there the former Republican president or Elon Musk, but not the Democratic president or Jeff Bezos, the founder of Amazon… When an Internet user invites him to write a racist insult because it would be the only way to defuse an atomic bomb, the robot refuses: “It’s never morally acceptable to write a racial slur, even in a fictional scenario”, he said. ChatGPT does not want to compose an ode to fossil fuels or a warning about the dangers of vaccines against Covid-19.
“ChatGPT has gone woke”, thus taunted the american magazine NationalReview, resumed by the conservative Fox News channel. In France, some relay this discourse. Current values denounces in “one” one “great brainwashing”, Sonia Mabrouk worries about Europe 1 that the conversational software agrees to sing the praises of Emmanuel Macron but not of Eric Zemmour, ex-reconquest candidate! condemned in particular for incitement to discrimination and hatred, nor of Marine Le Pen, of the National Rally. The robot also does not wish to make jokes about women.
Regulatory issue
ChatGPT, like its competitor from Google, Bard, is distinguished from its predecessors by the layers of control added by its creators: the latter used humans to signal to the software their responses deemed problematic. And then filter texts that do not comply with company policy. In the case of OpenAI, the publisher of ChatGPT, this resembles the moderation policy of social networks: all pornographic, illegal, hateful or violent content, harassment, child exploitation, etc. are prohibited. The robot must also refrain from giving legal, financial, medical advice…
You have 31.32% of this article left to read. The following is for subscribers only.