Two observations can be made. The first concerns the promised integration of AI in the operating room, and, as we know, this integration requires a lot of data. The second is that this integration is not as fast as hoped. We will try to identify some reasons for this "slowness", which for the most part are specific to the field of surgery. Variability of data, difficulties of the extraction task, are two of the main reasons for this slowness. We will examine these points first from our field as part of our research within the Bopa innovation chair – which means augmented operating theatre – (hepatobiliary surgery at Paul Brousse Hospital in Villejuif and at the National Institute of Oncology in Rabat, orthopaedic surgery at Trousseau Hospital in Paris and Hautepierre in Rabat, and the ARC research center at Tel-Hashomer Sheba Hospital in Ramat Gan) and second from the literature. I will limit myself as far as possible to the intraoperative phase (in the operating room), but since it depends on the preceding, preoperative phase, during which the planning of the operation is done, this latter must therefore be taken into account.