Three AI agents, two countries, and one uneven world wide web (royapakzad.substack.com)

A recent experiment by a technology and human rights researcher tested how three AI agents—Meta’s Muse, Anthropic’s Claude, and OpenAI’s GPT—handled data retrieval and processing for the World Bank's country profiles of the U.S. and Iran. The researcher aimed to examine the role of language in AI's decision-making and information retrieval, particularly as it relates to differing access challenges in English and Farsi. The agents faced significant disparities in performance; they retrieved valid data from U.S. sources much more effectively than from Iranian sources, revealing issues with language representation, contextual understanding, and source prioritization. This study is significant for the AI/ML community as it highlights the implications of AI agents on digital rights and accessibility, particularly in censored environments. It raises critical questions about the potential of LLMs to act as tools for circumventing censorship and ensuring better access to information. The experiment revealed stark differences in how each agent managed permissions, transparency in their processes, and their ability to adapt to access barriers. This raises awareness about the need for language diversity in AI systems and scrutinizes the responsibilities of AI developers in ensuring ethical data access and representation across different languages and regions.