A new preprint on arXiv, titled The Mathematical Collective Unconscious: Jungian Archetypes within the Latent Space, proposes a conceptual link between analytical psychology and machine learning. The abstract opens by recalling Carl Jung's notion of a collective unconscious—an inherited framework populated by universal archetypes—and then turns to large language models (LLMs) trained on massive text datasets.
The paper's central suggestion, as far as the abstract reveals, is that the latent spaces of LLMs may exhibit structures analogous to Jungian archetypes. The authors appear to argue that patterns recurring across human cultural texts could surface as shared, abstract representations within these models.
Because only the abstract is available in the source, the specific methods, evidence, or conclusions are not detailed. The announcement does not yet include a full paper, so the claims should be treated as an early hypothesis rather than a demonstrated result.