Facebook Caught Hosting AI-Powered Hitler

The bare minimum: don’t create RoboHitler!

RoboHitler

Fresh on the heels of Meta-formerly-Facebook issuing a sweeping — not to mention both inexplicable and all-around awful — series of changes to its “hateful conduct” standards allowing for a much-expanded array of horrendous speech on its platform, an NBC investigation has revealed that the social media giant’s platforms are literally hosting AI-powered Hitler chatbots.

Last year, Meta launched an “AI Studio” allowing Instagram, Facebook Messenger, and WhatsApp users to create and converse their own AI chatbots. Per the company’s AI Studio policies, these bots are supposed to follow a fairly strict set of rules: the chatbots can’t impersonate a living person, and if they do evoke a real human, that person needs to have been dead for more than 100 years; they can’t be based on fictional characters from copyright-protected media like movies, books, or videogames; and they can’t embody religious figures or leaders. Known terrorists, mass murderers, hate groups, and other violent or explicit figures and topics are off-limits as well.

What’s more, Meta claims to review all bots before they go live to ensure they follow these guidelines, which the company touts as a defense against chatbots producing outputs considered “inaccurate, offensive, or otherwise objectionable.”

But according to NBC’s reporting, a lot of chatbots are slipping through those purportedly human-monitored cracks — including chatbots impersonating the Nazi leader Adolf Hitler, arguably humanity’s most notorious mass murderer and undoubtedly a violator of Meta’s very clear AI Studio policies.

The discovery calls the social media giant’s promised AI safety practices around its public-facing chatbot characters into question. And backdropped by its deeply controversial move to roll back accepted standards around hate speech, hosting Hitler-styled chatbots is somehow an even worse look than it would be on any given day.

Violations on Violations

NBC’s review further discovered a number of other policy-violating chatbots, including AI bots created in the likeness of stars like Taylor Swift and Justin Bieber, and many modeled after copyright-protected fictional characters like Harry Potter and the “Frozen” franchise’s Elsa. There’s also, according to the reporting, a plethora of bots designed to emulate religious figures like Jesus Christ, Muhammad, and God.

In a statement, a Meta spokesperson told NBC that the “AIs in question that violate our AI studio policies have already been removed, and we’re continuously improving our detection measures to prevent creation and publication of AIs that violate our policies.”

But reassurance that bots were deleted after they were discovered by journalists doesn’t exactly sow comfort in Meta’s attention to ensuring that its much-publicized AI Studio bots are up to code, especially given how easily users were able to skirt by the company’s existing defenses.

More on Meta: Zuckerberg Announces New Measures to Increase Hate Speech on Facebook