How AI will fragment the internet

The technolibertarian dream of a borderless internet may soon get its biggest reality check in years. Meta’s Oversight Board—a governance body launched in 2020 that continues to independently make binding content moderation decisions and policy recommendations—released a report suggesting that large language models could be “stifling political speech.” A study of 10 commercial LLMs found that the models may be essentially censoring themselves based on the speech laws of countries around the world. What the board seems to find more concerning, however, is that the AI chatbots appear at times to be applying one country’s more repressive speech restrictions globally. For instance, multiple chatbots refused to create flyers critical of Saudi Arabia’s Crown Prince Mohammed bin Salman or China’s President Xi Jinping. Yet they would do so for U.S. President Donald Trump and U.K. King Charles III.