The Role of Ai's Linguistic Habitus in Reproducing and Combating Social Inequalities: A Sociotechnical Perspective on Natural Language Processing Models
When examining the technical workings of these machines, a sociological framework based on Bourdieu's concepts of habitus and linguistic markets can be applied. The argument is made that AI models are not neutral tools, but rather socialized agents that internalize and reproduce existing social structures. These systems, trained on a corpus of texts that, by its “legitimate” nature, reflect dominant cultural and socioeconomic norms. As highly respected "actants" and professional commercials of their discourse whose outputs are often accepted unquestioningly, AI models contribute to the social construction of reality and influence users' language use by reinforcing the meanings and discourses, linguistic habitus, they incorporate from the training data. Operationalized in their embeddings and generative neural networks as schemes of perception and action, the two dimension of the habitus.
However, AI also holds immense potential as a crystallizer of these worldview, collective memory, cultural tradition, and discourses embedded in its linguistic forms. If trained appropriately, AI big data models potential could democratize language by offering a more equitable representation of meanings, beyond the capacity of any individual. Moreover, AI can serve as a powerful tool for preserving cultural heritage, integrating the unique perspectives of certain traditions, generations, or social groups. This potential to conserve diverse worldviews, once maintained and shared primarily through oral tradition, is crucial in an era of globalization, individualization and rapid industrial and technological change, where traditional forms of knowledge risk being overshadowed by dominant, homogenizing narratives.