Algorithmic Neo-Institutionalism and AI Language Models

Large language models (LLMs) should not be viewed as neutral repositories of information or simple recommendation engines. Instead, they function as potent ideological and geopolitical instruments, reflecting the inherent biases and worldviews of the entities that finance and develop them. António Covas, the author of this opinion piece, contends that these AI models are deeply embedded within the power structures that shape their creation. Consequently, they inevitably mirror the tensions and dominant ideologies of their creators and financial backers. This perspective challenges the notion of AI as an objective tool, highlighting its role in propagating specific viewpoints and agendas. Covas suggests that understanding LLMs requires recognizing their function as ideological devices. Their outputs are not merely data-driven but are influenced by the geopolitical contexts and the specific interests of those who invest in and design them. This critical lens is essential for navigating the complex landscape of AI-generated content and its potential impact on public discourse. The piece urges readers to move beyond a superficial engagement with LLMs and to consider the underlying forces that shape their capabilities and outputs. By framing LLMs as geopolitical actors, Covas encourages a more discerning approach to the information they provide and the narratives they help to construct.