AI Distillation: Between Innovation, Intellectual Property, and Sovereignty

Azeem Azhar's article addresses the complex issue of AI model distillation, where laboratories, notably Chinese ones, reuse the outputs of American models. He compares this situation to the story of Samuel Slater who, in 1789, memorized British industrial plans to launch the American textile industry, highlighting a parallel between the transfer of industrial know-how and AI model distillation.

Distillation, a machine learning technique allowing a large model to train a smaller one, poses major legal and ethical challenges. Currently, the law does not consider AI outputs to be protected by copyright, creating a gray area. This situation fuels geopolitical tensions, particularly between the United States and China, where China seeks to achieve autonomy in AI chips, stimulated by American restrictions.

The author highlights the gap between technological advancement and the legal framework, and questions the need to regulate this practice. He suggests a targeted approach, focused on real harms, rather than broad regulation that could favor dominant players. He concludes by emphasizing the importance of nuance in the AI debate, in the face of extreme and polarized discourse.