Facebook CEO Admits AI Struggles More With Hate Speech Than Nudity
Mark Zuckerberg acknowledges that Facebook's AI systems currently detect nudity much more easily than hate speech due to the complexity of natural language processing. The company plans to shift to a proactive moderation model by combining AI advancements with tens of thousands of human reviewers.
Mark Zuckerberg admits that Facebook's artificial intelligence systems currently detect nudity much more easily than they identify hate speech. He explains that while computer vision easily spots a nipple, natural language processing struggles to understand the nuances, dialects, and context required to flag hateful language accurately.
Despite these linguistic challenges, Zuckerberg emphasizes that AI successfully removes 99 percent of ISIS-related content before any human views it. To bridge the gap in handling complex text, Facebook plans to combine ongoing AI improvements with the hiring of tens of thousands of human workers over the next few years.
This massive investment is part of a broader shift toward a proactive moderation model across the social network. Facebook hints that additional AI services for its family of apps are likely to debut next week at its annual F8 developer conference.