Rédiger un PREreview

This paper gives the first definition of AUGMANITAI, a way of thinking that intensive AI use has made possible but that has had no name. A person first uses several AI language models as a deliberately diversified council, then fades that support step by step until the structure of the multi-voiced dialogue keeps working inside the person without a device: often running by itself, interruptible whenever it is noticed, owned as one's own thinking. Judgement authority stays human while the frequency of human judgement falls; thinking stays coupled to the real world, where the consequences of a thought, not agreement among models, show whether it holds; and the self expands visibly through what it brings about. The mode exists only while its build-up stays slow, the inner loop can be ended at will, device use can be reduced or stopped, and all of this remains compatible with the integrity of the person, their social world and their environment. The paper defines the construct, separates it from neighbouring concepts (extended mind, scaffolding, supervisory control, cognitive offloading), draws contrast types, and puts it through a structured stress test of 20 objections from 20 disciplinary standpoints, each answered. The novelty lies in one combination: internalisation of a plural artificial dialogue, planned fading of that support, world coupling as the test of truth, and judgement condensation protected by the internalised loop. The definition is declared living: it will be revised as AI systems, agents, robots and human-AI teams develop, under the same concept DOI. Conceptual paper with testable hypotheses. English with German summary.