For most of this stretch, calling your work “artificial intelligence” was a way to get looked at funny. The term carried decades of overpromising, so the people doing the actual work in France mostly called it something else: machine learning, statistical inference, computer vision, robotics. INRIA and a handful of university groups kept at it through years with no hype to ride and no capital chasing them. Looking back from the end of 2012, that unglamorous stretch of AI research in France is the part that will matter, assuming it matters at all.
The backdrop was a research system being rewired. The 2006 programme law and the ANR’s shift to project-based money changed what got funded and how, pushing labs toward work you could scope and defend in a dossier. For machine-learning groups that cut both ways. Method work with a clear deliverable did fine. The open-ended theory, the kind whose payoff you cannot name in advance, had to fight harder for air. INRIA’s machine-learning and vision teams came through the period strong regardless, which says something about the depth already in place.
What France built in these years was a deep pool of people and almost nothing for them to do. The labs turned out researchers who knew the methods cold, and the country had very few places to put them. Dedicated AI companies barely existed here. A freshly minted expert in statistical learning faced a thin domestic market and a fat stack of offers from American labs and from finance, which pulled in quantitative talent without blinking. So the question hanging over 2012 is whether the training France paid for stays in France, or whether it has been quietly subsidising hiring in California and the City of London.
The one thing the research held onto was a seat at the international table. French groups stayed in the global conversation through the conferences and the collaborations, so nobody here was going to be blindsided by where the field went. That sat alongside the structured part of the ecosystem, the competitiveness clusters like Cap Digital and Systematic trying to wire research to industry, and the robotics and perception depth down in Toulouse. The pieces existed. Whether they add up to companies is a separate test, and not one France has passed yet.
There are stirrings now, at the close of 2012, that the neural-network ideas written off years ago are working again on problems that used to look hopeless. Maybe this is the turn. I have watched AI promise the world and hand back a letdown enough times that I am not ready to call it, and the honest position is that the gap between a real inflection and another false dawn is invisible while you are standing in it. What I will say is narrower. France spent these six years building a bench it mostly could not play. If the game ever starts, the bench is real. Whether anyone here ends up owning the team is the part nobody has solved.