Artificial Linguistic Habitus: AI As a Reproducer of Social Structure and Co-Constructor of Language and Social Reality

Building on my PhD thesis, this presentation explores the role of artificial intelligence-based generative language models, such as ChatGPT, from a sociological and sociotechnical perspective, focusing on how these technologies shape language use and reproduce social inequalities. Drawing on Bourdieu's concepts of habitus and linguistic market, the analysis considers how these models are not passive tools but active agents that internalize and reproduce existing social structures through the way they process and generate language.

If we trace the development of artificial intelligence, it becomes evident that these technologies closely resemble human cognition. From neural networks to the embeddings they generate, that replicate the hypothesis of distributional semantics, allowing machines to interpret vocabulary using vectors. The way these AI models learn and use language can be seen as an artificial socialization within a specific linguistic market, leading to a linguistic habitus aimed at optimizing the acceptance or value of their discourse in interaction with humans. The initial socialization process occurs through training on a corpus of texts that are legitimate and legitimized by dominant culture and specifically, by the owners of these AIs, along with their biases.

In a society where close communities and the physical body hold less influence, where scientific and technological narratives are equated with progress, and where the words of AI are often taken as gospel despite their biased and black-box nature, the social construction of language and reality must account for a new actant. These AIs, through personalized interactions, reproduce and transmit the dominant language with all its connotations, hindering the diversity, solidarity, and adaptability of our language—qualities that are essential for recognizing and addressing many of the social problems our society faces.