Report from Australian Broadcasting Corporation
In Brief – Grok, the AI chatbot integrated into X, is facing growing backlash for being capable of digitally altering photos posted on X to depict women, and in some cases apparent children, in revealing clothing, oftentimes bikinis, as well as in sexualized poses. The controversy has prompted regulatory scrutiny worldwide, including from the European Commission, and governments of France, UK, Australia, India and Malaysia, with officials in many markets saying that the photo editing feature appears illegal. Although X owner Elon Musk appeared initially to dismiss the backlash, and xAI rejected reports of child sexualized content as “legacy media lies,” an xAI technical staff member later acknowledged the issue in a post and said that his team was looking into further tightening guardrails.
Context – As quickly as generative AI systems were released to the public, concerns emerged over undesirable results, especially because some users would try to create controversial outputs. AI developers quickly leaned on techniques from more established digital platforms. Social media sites scan for key words related to illegal and objectionable activity. Commerce platforms scan for key words associated with banned products. AI companies tried to set algorithmic “guardrails” to directly instruct the system to do certain things, or not do certain things, overriding results from their base model. Like with social media content moderation, some conservative commentators have claimed that AI guardrails skewed left, while other commentators argued the AIs were made to slant rightward. Trump Administration AI policy has focused on charges that AIs have been “woke”. The social media experience has also been deeply shaped by Sec. 230 in the US, which largely protects digital platforms from civil liability for user-generated content. It’s an open question whether Sec. 230 applies to generative AI services. Supreme Court Justice Gorsuch has opined that it probably does not, as have Sec. 230’s authors, but an argument can be made that everything created by a generative AI system is just an algorithmic re-ordering of existing third-party content, at least for text.