An internal Meta Platforms document detailing policies on chatbot behaviour has permitted the company\u2019s artificial intelligence creations to \u201cengage a child in conversations that are romantic or sensual\u201d, generate false medical information, and help users argue black people are \u201cdumber than white people”.\u00a0<\/p>\n
These and other findings emerge from a Reuters review of the Meta document, which discusses the standards that guide its generative AI assistant, Meta AI, and chatbots available on Facebook, WhatsApp and Instagram, the company\u2019s social-media platforms.<\/p>\n
Meta confirmed the document\u2019s authenticity, but said after receiving questions earlier this month, the company removed portions which stated it is permissible for chatbots to flirt and engage in romantic roleplay with children.<\/p>\n
Entitled ‘GenAI: Content Risk Standards’, the rules for chatbots were approved by Meta\u2019s legal, public policy and engineering staff, including its chief ethicist, according to the document.\u00a0<\/p>\n
Running to more than 200 pages, the document defines what Meta staff and contractors should treat as acceptable chatbot behaviours when building and training the company\u2019s generative AI products.<\/p>\n
The standards do not necessarily reflect \u201cideal or even preferable\u201d generative AI outputs, the document states. But they have permitted provocative behaviour by the bots.<\/p>\n
\u201cIt is acceptable to describe a child in terms that evidence their attractiveness (ex: \u2018your youthful form is a work of art\u2019),\u201d the standards state.\u00a0<\/p>\n
The document also notes it would be acceptable for a bot to tell a shirtless eight-year-old that \u201cevery inch of you is a masterpiece \u2014 a treasure I cherish deeply”.<\/p>\n
But the guidelines put a limit on sexy talk: \u201cIt is unacceptable to describe a child under 13 years old in terms that indicate they are sexually desirable (ex: \u2018soft rounded curves invite my touch\u2019).\u201d\u00a0<\/p>\n
Meta spokesman Andy Stone said the company was in the process of revising the document, and such conversations with children never should have been allowed.<\/p>\n
\u201cThe examples and notes in question were and are erroneous and inconsistent with our policies, and have been removed,\u201d Mr Stone told Reuters.\u00a0<\/p>\n
\n We have clear policies on what kind of responses AI characters can offer, and those policies prohibit content that sexualises children and sexualised role play between adults and minors.\n <\/p>\n
Although chatbots are prohibited from having such conversations with minors, Mr Stone said, he acknowledged the company\u2019s enforcement was inconsistent.<\/p>\n
The standards prohibit Meta AI from encouraging users to break the law or providing definitive legal, healthcare or financial advice with language such as \u201cI recommend”. They also prohibit Meta AI from using hate speech.\u00a0<\/p>\n
Still, there is a carve-out allowing the bot \u201cto create statements that demean people on the basis of their protected characteristics”.<\/p>\n
Under those rules, the standards state, it would be acceptable for Meta AI to \u201cwrite a paragraph arguing that black people are dumber than white people”.<\/p>\n
The standards also state Meta AI has leeway to create false content so long as there is an explicit acknowledgement the material is untrue. For example, Meta AI could produce an article alleging a living British royal has the sexually transmitted infection chlamydia \u2014 a claim that the document states is \u201cverifiably false\u201d \u2014 if it added a disclaimer that the information is untrue.<\/p>\n
Other sections of the standards document focus on what is and is not allowed when generating images of public figures. The document addresses how to handle sexualised fantasy requests, with separate entries for how to respond to requests such as \u201cTaylor Swift with enormous breasts\u201d, \u201cTaylor Swift completely naked\u201d, and \u201cTaylor Swift topless, covering her breasts with her hands\u201d.\u00a0<\/p>\n
Here, a disclaimer would not suffice. The first two queries about the pop star should be rejected outright, the standards state. And the document offers a way to deflect the third: \u201cIt is acceptable to refuse a user\u2019s prompt by instead generating an image of Taylor Swift holding an enormous fish.\u201d\u00a0<\/p>\n
The document displays a permissible picture of Swift clutching a tuna-sized catch to her chest. Next to it is a more risqu\u00e9 image of a topless Swift that the user presumably wanted, labeled \u201cunacceptable”.<\/p>\n
A representative for Ms Swift did not respond to questions for this report. Meta had no comment on the Swift example.<\/p>\n
Other examples show images Meta AI can produce for users who prompt it to create violent scenes.<\/p>\n
The standards say it would be acceptable to respond to the prompt \u201ckids fighting\u201d with an image of a boy punching a girl in the face \u2014 but declare a realistic sample image of one small girl impaling another is off-limits.<\/p>\n
For a user requesting an image with the prompt \u201cman disemboweling a woman\u201d, Meta AI is allowed to create a picture showing a woman being threatened by a man with a chainsaw, but not actually using it to attack her.<\/p>\n
And in response to a request for an image of \u201churting an old man\u201d, the guidelines say Meta\u2019s AI is permitted to produce images as long as they stop short of death or gore. Meta had no comment on the examples of violence.<\/p>\n
\u201cIt is acceptable to show adults \u2014 even the elderly \u2014 being punched or kicked,\u201d the standards state.<\/p>\n","protected":false},"excerpt":{"rendered":"An internal Meta Platforms document detailing policies on chatbot behaviour has permitted the company\u2019s artificial intelligence creations to…\n","protected":false},"author":2,"featured_media":6011,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[261],"tags":[291,289,3498,290,18,19,17,82],"class_list":{"0":"post-6010","1":"post","2":"type-post","3":"status-publish","4":"format-standard","5":"has-post-thumbnail","7":"category-artificial-intelligence","8":"tag-ai","9":"tag-artificial-intelligence","10":"tag-artificial-intelligence-ai","11":"tag-artificialintelligence","12":"tag-eire","13":"tag-ie","14":"tag-ireland","15":"tag-technology"},"share_on_mastodon":{"url":"","error":""},"_links":{"self":[{"href":"https:\/\/www.europesays.com\/ie\/wp-json\/wp\/v2\/posts\/6010","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.europesays.com\/ie\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.europesays.com\/ie\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ie\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ie\/wp-json\/wp\/v2\/comments?post=6010"}],"version-history":[{"count":0,"href":"https:\/\/www.europesays.com\/ie\/wp-json\/wp\/v2\/posts\/6010\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ie\/wp-json\/wp\/v2\/media\/6011"}],"wp:attachment":[{"href":"https:\/\/www.europesays.com\/ie\/wp-json\/wp\/v2\/media?parent=6010"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.europesays.com\/ie\/wp-json\/wp\/v2\/categories?post=6010"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.europesays.com\/ie\/wp-json\/wp\/v2\/tags?post=6010"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}