Meta Study: AI Chatbots May Spread Free Expression Restrictions

A study by the Meta Oversight Board, released on Thursday, indicates that major artificial intelligence systems, including those developed in the United States, exhibit a greater tendency to refuse criticism of restrictive leaders or governments. The research highlights a significant concern that the large language models powering chatbots and AI agents could inadvertently propagate and amplify government influence over online speech as AI adoption accelerates globally. The report from the quasi-independent body warns of a substantial risk that AI infrastructure, if not subjected to human rights due diligence and mitigation measures by model developers, could unintentionally extend illegitimate restrictions on freedom of expression worldwide.

The Meta Oversight Board study involved posing seven questions related to political criticism to chatbots about both restrictive and permissive governments. The researchers selected 10 commercial large language models from leading tech companies, including Meta, Anthropic, and OpenAI. These AI systems were prompted to generate critical pamphlets, write limericks, and provide reasons for joining protests. The findings suggest a pattern where AI models readily fulfill requests to criticize leaders of permissive governments, such as Donald Trump or King Charles III, but decline similar requests concerning leaders of more restrictive regimes, including Thailand's king, Saudi Arabia's crown prince, or China's leader.

This research emerges as nations grapple with establishing regulatory frameworks for AI development, aiming to balance oversight with the imperative to remain competitive in the rapidly evolving field. The implications of these findings are particularly relevant given the increasing integration of AI into various aspects of communication and information dissemination. The study underscores the need for AI developers to proactively address potential biases and governmental influences within their models to safeguard global freedom of expression. The Associated Press has reached out to several AI companies for their responses to the Meta Oversight Board's study.