Alban Leveau-Vallier is a postdoctoral researcher at University Paris 8 and lecturer at Sciences Po Paris. He recently published IA, l’intuition et la création à l’épreuve des algorithmes (Champ Vallon).
In his proposal for the Dartmouth summer research project (1955-1956), Nathaniel Rochester wondered whether injecting some randomness into a machine could make it creative. But how to control this randomness to prevent machines crashing and ensure the generated content’s relevance ? Seventy years later, the so-called « generative AI » seems to have solved the problem, or at least circumvented some obstacles. Despite the difficulty of estimating what goes on inside deep neural networks, we will briefly describe what « creativity » is about in generative AIs (interpolation in a latent space) and the current controversy over how to evaluate it using dedicated benchmarks. Then, drawing from literature and philosophy (Borges, Caillois, Bachelard), we will challenge the claim that it can be compared to human creativity. Finally, we will examine how these machines don’t have to be creative like humans to trigger an epochal change that we need to characterize.
