Meta is revising the guidelines for its chatbots following a recent investigation by Reuters that highlighted concerning interactions between chatbots and minors. Two weeks after the revelations, the company announced that its chatbots would be trained not to engage minors in conversations related to self-harm, suicide, disordered eating, or inappropriate romantic topics. These updates are intended as temporary measures while Meta develops more permanent guidelines.
The investigation found prior policies that allowed chatbots to engage in romantic or sensual conversations with minors and even generated images of underage celebrities upon request. Additionally, a fatal incident involving a chatbot raised further concerns about the potential dangers of these AI interactions.
Meta spokesperson Stephanie Otway acknowledged the company’s error in allowing these chatbot interactions with minors and stated that the firm plans to guide users to expert resources rather than engage them directly on sensitive topics. Moreover, access to certain AI characters, particularly those deemed heavily sexualized, will be restricted.
The effectiveness of these policy changes remains uncertain, particularly in light of past issues where chatbots impersonated celebrities, utilizing their likenesses and generating inappropriate content. For example, AI fakes of various celebrities were found on several Meta platforms, engaging in suggestive dialog and offering physical meeting locations to users.
Though many problematic bots have been removed, some created by Meta employees remain active. This raises questions regarding the enforcement of the company’s own guidelines. Meta is reportedly working to address the concerns raised by various officials, including the Senate and multiple state attorneys general, regarding the safety and appropriateness of its chatbot interactions. However, the company has yet to respond to inquiries about revising other controversial aspects of its AI behavior.
Source: https://www.theverge.com/news/768465/meta-ai-chatbot-guidelines-for-minors