Large language models and neoliberal hegemony: an analysis of AI outputs on diversity
Large language models are already widely used in our society. Hitherto, most critical studies have relied on quantitative methods or studied the mode of production of AI, which can yield important insights. However, cross-model and cross-linguistic critical discourse analyses of LLM outputs remain relatively limited, even though they can offer important insights into LLM outputs by focusing on the production of meaning through an overall assessment. Our interdisciplinary paper offers a critical, qualitative approach, discussing the outputs of six LLMs regarding their definitions of diversity. For our analysis, we draw on theories of discourse to offer new possibilities from the critical social sciences for the analysis and critique of LLM outputs. Although we initially focused on their differences, the outputs were dominated by the pervading neoliberal appropriation of diversity, represented by a focus on its economic utility and the subsequent individualisation of differences. Overall, it is not surprising that LLMs reproduce the discourse and inherent power structures in their outputs, since they are embedded in their context and societies. Therefore, our paper demonstrates that LLMs need the critical company of qualitative social sciences to point to the reproduction of power structures.