Fossil wood is one of the most common types of plant fossils and often preserves fine features of the tissue (e.g., growth-ring boundaries) and of the individual cells themselves (wall thickness, ornamentation, etc.). Fossil wood provides information on the systematic affinities of extinct plants but also on their physiology and on the environmental conditions in which they grew. However, their study is time consuming and relies on a small number of experts worldwide, meaning that the paleobiological and paleoenvironmental information recorded in fossil woods remains largely unexploited. The exploratory project Foss-AI aims to automate the extraction of relevant qualitative (e.g., cell types) and quantitative (e.g., cell size) data from fossil wood sections using deep learning, the part of artificial intelligence that replicates the way humans gain certain types of knowledge. Recent studies have already demonstrated the feasibility of this approach on histological thin-sections of extant plants, but fossil woods pose a few additional challenges due to their variable preservation and to the way they are prepared. In the first stage of the project, we are concentrating on the analysis of transverse sections of Devonian, Carboniferous, and Permian woods from key localities. High-resolution images of slides acquired with a digital microscope are labelled by human experts and used to train the AI. Results of the AI based analysis of new images will then ben compared to analysis by the human experts to assess the validity of the results on different types of fossil wood material. The long-term goal of this project is to develop tools that can help and promote the work of fossil wood experts by facilitating large-scale analyses, with significant implications for our understanding of past plant communities and their dynamics.