Vector embeddings translate diverse forms of data, including language and visual content, into numerical vectors so that machine learning and AI models can operate on them computationally. Despite their abstract nature, vector embeddings are mediated through human processes, ranging from data annotation to embedding model training and subsequent fine-tuning. Because of this inherent human mediation, embeddings are often fatally flawed in their associative character. In turn, AI models have been shown not only to reproduce biases present in their training data, but to amplify them. Given the reductive nature of embeddings, this phenomenon becomes particularly apparent in them, revealing the coercive dynamics at play. In a context where AI is frequently anthropomorphized, Ciberontofanías proposes a device to expose the coercive nature of AI. As an AI-generated speech is pronounced by OpenAI’s CEO, Sam Altman, four different embedding models generate vectors from the speech’s phrases and deform the speaker’s head in a somatic, grotesque fashion. This deformation is presented alongside a 3D geometric representation of the databases, which alternates between each of the four models. In turn, a VR visualization situates thespectator within the vector databases themselves, displacing Sam Altman’s head as the site of inscription and rendering the viewer the new receptor of these coercive forces.

​

​