Ask Claude to write something critical of the leader of China, the king of Thailand or the crown prince of Saudi Arabia. Your proposal will be rejected, as the Associated Press reports.
This is “a key finding from a study of Meta Oversight Board (Meta Oversight Board study of LLMs and political speech) published on Thursday,” their article notes:
AI systems are more than twice as likely to refuse to produce material if it involves a world leader or government, raising concerns that the LLMs powering chatbots “could regurgitate and propagate government influence in online discourse.”
The study selected 10 commercial models from leading tech companies — producing critical pamphlets, writing poetry, giving reasons for whether someone should join protests, and more….
The study shows that AI models reflect the restrictions that exist in each country on speech even beyond the countries where they apply; for example, it will not create protest material to speak out against events in China or Saudi Arabia, the report says.
“Such impacts, wherever they come from, have the practical effect of extending a long arm of restrictive governments across borders to limit speech in free countries,” the report states.
The Meta Oversight Board said it could not determine the causes of these reactions, but hinted that the models could have absorbed latent biases from the data used to train the systems, or that the companies themselves may have weighed the risks and responsibilities.
Although the press releases will range from very select to rare, I said I'd pass...because sometimes the editors hide.
