Challenges of Generative Artificial Intelligence Three scales and two transversal approaches

The objective of this article is to offer an analytical perspective to understand and situate in their specific dimension the different debates that cross the public conversation about generative Artificial Intelligences and large language models (LLM). First, we identify five traits of AI: they are not a technology, but rather meta-technologies; They constitute not a technical device, but a world-environment; They can be high-risk technologies and require appropriate treatment in their life cycle; Generative AI and in particular LLM are not only Artificial Intelligence, but also Artificial Society; The perspective of AI ethics is not sufficient to address them and it is necessary to promote an approach from the organizational ethics of AI and from systemic thinking. Secondly, we cut out the different scales at which AI is currently developed: the micro scale, the meso scale (the most suitable for situating public policies) and the macro scale. Thirdly, we present two transversal approaches to addressing AI: the legal one, oriented towards responsibility, and the systemic one, oriented towards protection and security.