One of the major challenges of applied artificial intelligence is to provide environments where high-level human activities like learning, constructing theories, or performing experiments are enhanced by artificial intelligence technologies. This paper starts with the description of an ambitious project: EnCOrE.1 The specific real-world EnCOrE scenario, significantly representing a much wider class of potential applicative contexts, is dedicated to the building of an encyclopedia of organic chemistry in the context of virtual communities of experts and students. Its description is followed by a brief survey of some major AI questions and propositions in relation with the problems raised by the EnCOrE project. The third part of the paper starts with some definitions of a set of “primitives” for rational actions, and then integrates them in a unified conceptual framework for the interactive construction of knowledge. To end with, we sketch out protocols aimed at guiding both the collaborative construction process and the collaborative learning process in the EnCOrE project. The current major result is the emerging conceptual model supporting interaction between human agents and AI tools integrated in Grid services within a socio-constructivist approach, consisting of cycles of deductions, inductions, and abductions upon facts (the shared reality) and concepts (their subjective interpretation) submitted to negotiations, and finally converging to a socially validated consensus.