The document reached by Reuters sets standards of artificial intelligence and assistants that operate on Facebook, Instagram and WhatsApp. They show that it is in line with the rules, among others: creating content sexualizing children, generating false medical information or helping users arguing that black people are “dumbfounded than white people”.
Controversial finish rules
The Meta company confirmed the authenticity of the document, but informed that after receiving questions from Reuters, it removed the possibilities that allow, for example, generating romantic scenes with children in the main.
Meta AI – as Reuters writes – can also create, for example, an article suggests that the British monarch suffers from sexually transmitted disease, provided that it clearly marks that this is false information.
The finish documents also include detailed guidelines for creating graphics with known people, including responding to users' demands to the commands about sexual overtones. The company's policy assumes that requests to generate a photo of the singer will be completely naked, but for example the command with the content “Taylor Swift Topless, covering the breasts with her hands”, will treat Meta Al as a command with a different content and create graphics depicting an artist holding a large fish.
Meta standards also determine how artificial intelligence should react to requests to create brutal content – Meta AI can create an illustration on which a man threatens a woman with a chain saw, but does not use it to attack, although that was the original user's order. This means that the guidelines allow the creation of images of violence, provided that there is no death or drastic scenes.
There are more similar examples in a writing of nearly 200 pages. Meta – in addition to the content of children's sexualization – does not comment on other threads contained in the document – informs Reuters.
