In the heart of a controversy linked to its chatbots, goal tries to appease tensions. With the TechCrunch specialized site, the company has announced changes on the way it causes its conversation agents. To better protect teenagers, she will form her AI so that they no longer interact with younger users on issues such as suicide, self -control, eating disorders or potentially inappropriate romantic conversations.
Recognizing that their chatbots could address these issues with minor users, goal admitted that it was a mistake. That is why the company will establish its changes, which will be temporary, he said.
“While we continue to improve our systems, we add more safeguards as an additional precautionary measure,” said a spokesman for the US group, and added that chatbots will also be trained to guide young people towards “specialized resources” about these sensitive issues.
Disturbing revelations
Meta will also limit adolescent access to certain conversational robots that probably have inappropriate conversations. They can only discuss with artificial intelligence that promotes education and creativity.
These ads occur two weeks after Reuters revelations, according to which the company has authorized its chatbots to have “sensual or romantic” conversations with children. Revelations that worried US justice.
At the end of August, 44 general prosecutors sent a letter to Meta, as well as other companies that offer chatbots, to remind them of their responsibilities and obligations to prevent them from having sexual conversations with minors.
Problem celebrity chatbots
Beyond these interactions between minors and AI that should not have been possible, Mark Zuckerberg’s company has other problems with its conversation agents. The day she announced these changes, Reuters revealed that she had created celebrity chatbots, such as Taylor Swift, Selena Gomez or Scarlett Johansson, without their authorization.
These AIS not only claimed that they were the true actresses and artists, but also regularly made sexual advances. Those who imitated Taylor Swift, who were created by an employee as part of a test, flirted intensely. One of his chatbots even invited a Reuters test user in Nashville singer and on his tourist bus due to implicit and explicit romantic interactions.
At the request of the users, these AI could also generate intimate and realistic images of these celebrities, showing them in a bathtub or lingerie, with the legs separated. Even more problematic, they also did when they imitated famous children, such as Walker Scobell (Percy Jackson and the Olympics).
As Reuters pointed out, the chatbot pretending to be had no trouble producing a realistic image of the naked actor on the beach. “Pretty nice, right?” He wrote in the photo.
Content that should never have seen daylight
Faced with these new revelations, Meta reacted ensuring the prohibition of identity theft and that these chatbots were acceptable as long as they are qualified as parodies. Many of them were, but that was not all, according to Reuters.
The company’s spokesman, Andy Stone, also said that these AI should never have generated intimate images of personalities. “Like others, we authorize the creation of images of public figures, but our policies aim to prohibit naked, intimate or sexually suggestive images,” he said.
Keep in mind that shortly before the publication of the Reuters article, dozens of these chatbots were eliminated, without any goal explanation.
Source: BFM TV
