Vector Media

It is difficult to overstate the importance of text and image embedding in contemporary machine learning. Yet many current critiques of artificial intelligence overlook this paradigm in their efforts to understand generativity as a cultural phenomenon. In this talk, Offert presents a new theory and history of embedding as a medium. He argues that, whereas previous technological revolutions produced new media on the basis of existing ones, contemporary machine learning systems instead aim to dissolve prior media into a universal space of commensurability. This transformation becomes clearer when embedding is situated within the history of computer vision, where it emerges as the provisional endpoint of a longer series of attempts to theorize intelligence as compression.