What is the technical essense of AI systems, whether we’re discussing GOFAI, early connectionist models, or the deep learning methods prevalent today?
Within the wider arc of Gilbert Simondon’s philosophy, invention stands out as a stage in the development of mental images, rather than a faculty distinct from perception and memory. Of particular interest is a phase of transition that precedes invention in the development of mental images. Simondon compares this transition to the process of metamorphosis occurring in some species. This transition in the development of mental images is marked by the dedifferentiation of the dominant organizing principle. This dedifferentiation paves the way for the possible reorganization of mental images at a higher level of development. The free play of mental images corresponds to this transition, enabling the discovery of a new organizing principle with unprecedented possibilities of adaptation. It is enlightening for contemporary debates about machine learning processes, like those operative in Large Language Models or of image generation, to think carefully about this transition. In this article, we will look at the significance of this transitional dedifferentiation in living beings. This will lead us to argue against the use of the term Artificial Intelligence. A better alternative seems to be the term ‘Automated Optimization’ (AO) – suggested by the engineer and philosopher Yagmur Denizhan. Denizhan defines intelligence as the «border activity between the modelled and the unmodelled», i.e. between what is admissible in our model of reality and what is excluded or not yet encompassed by it. Intelligence thus conceived, I propose, is directly relevant to Simondon’s analogy between invention and metamorphosis. The «border activity» between the modelled and the unmodelled, at the level of cognition, may thus correspond to free play of mental images in the strong sense, namely, as involving transitional dissolution of their organising principle. Without it, I argue, we cannot begin to understand the historical recasting of our mental worlds, including paradigm shifts in the arts
What is the technical essense of AI systems, whether we’re discussing GOFAI, early connectionist models, or the deep learning methods prevalent today?