Communication Bias in Large Language Models: A Regulatory Perspective

Large language models (LLMs) have become integral to numerous applications, raising concerns about bias, fairness, and compliance with emerging regulatory frameworks. This article provides a review of some of the most significant risks associated with biased LLM outputs and their broader societal implications. We discuss how regulatory initiatives such as the European Union’s (EU’s) Artificial Intelligence Act (AI Act) and the Digital Services Act (DSA) seek to address these challenges by introducing legal requirements for transparency, accountability, and harm mitigation. In addition, we explore various approaches for assessing and mitigating bias, along with the limitations of current methods. While regulatory frameworks play a crucial role in governing artificial intelligence (AI), they also present challenges in balancing innovation with legal compliance. As oversight mechanisms continue to evolve, we emphasize the fact that LLMs are increasingly regarded as a distinct subset of AI systems and that LLMs are particularly noteworthy due to their immediate interaction with consumers, functioning as the primary interface between users and AI. This distinguishes LLMs from many other AI applications operating behind the scenes and renders them unique in the sense that they are becoming integral tools for personal decision making, offering validation, guidance, and support across various aspects of everyday life, including healthcare, finance, and even choices made in relation to politics. While the current landscape of AI-generated content is addressed by several new regulatory frameworks, none of them have been devised with the possibility in mind that LLMs may affect citizens’ fundamental worldviews and social perspectives, including their individual and collective voting decisions. Our analysis highlights the need for a greater focus on both competition and technology design governance as a complement to value chain and content-based regulation, in order to foster fair and trustworthy AI systems.