Using AI to Reconstruct Claude Bernard's Empirical Investigations

We propose an Artificial Intelligence architecture dedicated to the reconstruction of old scientific discoveries. This architecture is based on the notion “core models” upon which the system draws out the logical consequences of scientific hypotheses. “Core models” are based on an initial ontology that contains entities and concepts recognized by scientists. Hypotheses are tentative explanations about entities functions. Those hypotheses have to be tested by a valid model, which may realize virtual experiments, i.e. “in silico”, or real-world experiments. A third module, under construction, is aimed at comparing the expectations to the experimental data and then building new working hypotheses using abduction as inference. The architecture is validated on historical data: the ultimate goal is to reconstruct parts of Claude Bernard’s empirical investigations. More precisely, it is to reconstruct the main steps of his scientific inquiry. The paper presents the general architecture and the way it helps to understand some of Claude Bernards scientific steps.