The sciences are searching for their new foundations. We show how epistemic matrices can offer the sciences of knowledge the opportunity to find their complex thinking with cognitive temporalism. Complex thought finds its full expression in cognitive temporalism. This is why the systemic practice of epistemic matrices implements the Method of exercising complex thought, to borrow the phrase of Le Moigne [14]. The matrices invite the researcher to think at the limits of neuro-cognitive. They allow us to understand the cognitive framework proposed by AI, as the uniduality of a quantitative-temporal dialogic. Big Data possesses this formless form which is that of the world. Thinking about Big Data through a quantitative-temporal dialogic means thinking that the times of knowledge permeate the world, and that the learning of this knowledge is a reciprocal function. The project of AI is to make knowledge accessible to as many people as possible, by restoring it to the world in the most varied forms. Morin asks us to understand this new world through the notions of order and disorder, outside of any logic. In cognitive temporalist terms, this injunction translates into saying that the emergence of AI in the world must choose, to ensure its sustainability, between synchronies that deconstruct before reconstructing, and diachronies that construct before deconstructing. A paradox that Morin summarizes with the Heraclitean statement: Live from death, die from life. We have seen that with the logic of cognitive temporalism it is possible to define an ethics of cognition with AI which is an ethics of the living, in accordance with the self-eco-organization of Morin.