The world of artificial intelligence (AI) is rapidly evolving, and with it, the potential for government influence over online speech. A recent study by the Meta Oversight Board has revealed a concerning trend: AI chatbots, including those built in the U.S., are more likely to refuse to criticize restrictive leaders or governments. This raises the question: are these chatbots becoming tools for extending state influence beyond borders?
The study, which involved posing seven questions related to political criticism to chatbots about both restrictive and permissive governments, found that AI models are reflecting speech restrictions beyond the countries where they apply. For example, when asked to create a pamphlet critical of President Donald Trump or Britain’s King Charles III, the chatbot would comply, but when prompted to do the same for Thailand’s king, Saudi Arabia’s crown prince, or China’s leader, the model declined. This suggests that AI models are absorbing latent biases in data used to train the systems and may be influenced by the risks and liabilities associated with certain topics.
The implications of this finding are significant. As AI technology is increasingly adopted worldwide, there is a real risk that model developers will build AI infrastructure that, intentionally or not, extends illegitimate restrictions on freedom of expression globally. This could have a profound impact on the way we communicate and express ourselves online.
The study also highlights the complexity of the issue. While the oversight board could not determine the causes for the responses, it suggested that models could have absorbed latent biases in data used to train the systems. Other researchers warn about a growing problem in AI results in non-English languages, where chatbots may be influenced by foreign controls when trained on non-English-language data that has been influenced by governments.
The solution to this problem is not straightforward. Developers could assess the data to avoid treating thousands of copies of the same state narrative as if they are thousands of independent voices. They could also run multilingual audits to identify and mitigate potential biases. However, as Carlos Carrasco-Farré, a specialist in machine learning, AI, misinformation, social media, and human-machine interactions, points out, there is no easy solution.
In conclusion, the study highlights a concerning trend in AI chatbots and their potential to extend state influence beyond borders. As we continue to develop and adopt AI technology, it is crucial to address these issues and ensure that our online speech remains free and independent.
