LLMs as glocal inequality crisis by Cristina Godoy Bernardo de Oliveira et al. published by G20 Brasil 2024, T20 Brasil 2024, CEBRI, Fundação Alexandre de Gusmão, Ipea (2024).
“The urgent need to address global inequalities, particularly amidst the challenges posed by the COVID-19 pandemic, climate change, geopolitical tensions, and rapid advancements in Artificial Intelligence (AI), has prompted Brazil to prioritise this issue during its presidency of the Group of 20 (G20). Despite discussions and efforts, little progress has been made within the G20 regarding the regulation and governance of AI, leading to exacerbating existing disparities, especially in the most vulnerable countries. Large language models (LLMs) have emerged as pivotal tools with the potential to both exacerbate and mitigate global inequalities. While they offer opportunities for democratizing access to education, fostering cultural diversity, and enabling economic prosperity, they also deepen disparities by privileging technologically advanced regions and individuals, exacerbating existing inequalities in labour markets, and undermining cultural diversity. Moreover, the proliferation of LLMs raises significant concerns regarding privacy, misinformation, and algorithmic bias, necessitating urgent collaborative responses from the G20 to mitigate these impacts.
In response to these challenges, this Policy Brief proposes recommendations for the G20 to consider, including the establishment of a Global AI Observatory to facilitate informed discussions and policy recommendations on AI, the development of an international framework for AI ethics and workers’ rights, and initiatives to promote linguistic and cultural inclusivity in AI development. Additionally, regulations on AI generated content, data privacy, and property rights are suggested to ensure ethical AI usage. At the same time, efforts to strengthen academia and enhance critical computational literacy1 aim to foster a more informed and capable society that engages with AI technologies responsibly. The brief concludes by presenting potential outcomes and trade-offs associated with each recommendation…”
