Fireside Conversation with Yann LeCun | Talk | VivaTech

I get even more invitations that I have to refuse if I want to do some real work, which means real research. Other than that, not so much. It's a great honor, of course, it's a bit of a surprise as well, and it's very good for the field of machine learning and AI because in the history of the Turing Award, a lot of them are focused on core computer science things like systems, compilers, programming languages, theory, which is your expertise actually, not mine. I don't know anything about that. And then AI, but very often more logic-based AI, which is also close to core computer science. So things like machine learning have received very little attention from the Turing committee in the past. It's a bit of a new thing. I'm really happy that they chose to give it to the three of us: Yoshua Bengio, Geoffrey Hinton, and myself.

Yeah, I think it's a very important question. So I'm myself part-time at Facebook. I spend part of my time as a professor at New York University, so I share my time between the two. I think the two worlds are very complementary. People work with different assumptions in the two worlds, and so that drives them to work on different topics. They also have different kinds of resources at their disposal. So a lot of good ideas come from academia or from public research. A lot of the practical things, like the good results in machine translation or computer vision, come from industry because you need to scale it up to make it work really well in practice. And there is very good research coming out of industry as well, mostly from Facebook, Google, Microsoft, a little bit from IBM, and a few others. I think the partnership between industry and public research is very fruitful, and it has to be done in a way that is not detrimental to any of the two. One model that we have pushed, and that we are using in the US and Canada, as well as a little bit in the UK and in France as well, is to have people part-time. Younger professors generally spend only a small amount of time at Facebook, most of the time in the university. And then people like me, you said I'm a young Turing winner, but I'm not that young, can spend more time because they have passed the sort of junior step. We have a lot of collaborations in France with INRIA and various universities. We have 20 PhD students under the CIFRE status. For those of you who are not from France or don't know, there is a very special thing in France called CIFRE. It's a particular status for PhD students that allows them to spend most of their time in industry under co-advisorship from research scientists in industry as well as in academia, and it fosters collaboration. We have 20 of them at Facebook AI Research, and it's extremely productive. The students are fantastic; they've produced amazing results that found their way into Facebook products, but also were published and open-sourced. So that's been incredibly fruitful. The first batch of students is graduating; some of them are going to academia, some are working for Google, some are working for French companies, starting startups. So I think it helps a lot in building the local ecosystem. You can add that if your company has PhD students under this CIFRE system, you get help from the state, from the government, though this is not why we're doing it. I am really convinced that these part-time positions are very useful and really helpful to develop relations between France and its academy and industry.

I think it's a wonderful thing. There is an issue in France, and I hope you're going to hit me for saying this, but researchers in France are very badly paid.

Okay, I knew you did. And this is also part of the new report on AI. It's very difficult for France to attract top talent in public research or in universities because salaries are so low compared to the equivalent in North America, Switzerland, some parts of Asia, and certain types of positions in Germany. So it makes it very difficult to attract talent to France. I wish when I was a young PhD student graduating that there were research opportunities in France that would have allowed me to stay in France, and it wasn't the case. So now there are labs like Facebook AI Research, like Google Research in Paris, that offer this kind of opportunity for young researchers, and perhaps with part-time positions in public research to foster interactions. I think French industry also should probably take a little more risk in investing in long-term research. This is not a well-established tradition in France, at least not in information technology. It has been in other areas like aerospace and nuclear energy, but not so much in information technology. So outside of a very small number of French companies, there is essentially no long-term advanced research in information technology in France.

France in particular, but Europe in general, has a very good education system, so there is a lot of talent. The quality of the education system is the most difficult thing to build. There are frictions and the kind of institutions of research certainly that make it difficult for the best talents to really flourish. A lot of people realize this and want to do something about it, but it's very difficult given the inertia of decades of history. But the talents are there, and that's central. What I've been impressed by is the growth of the technology ecosystem, particularly in Paris. France is a very centralized country, so a lot of things happen in one place. This is built around high-quality technical science and technology schools and engineering schools. What we need is a mixture of industry research lab, public research, attracting the best people from universities and engineering schools, and startups. I think it's happening in Paris. It could be faster, but Paris is becoming one of the top places in Europe for VC investments, particularly in AI-related industry.

Super important. So I have two hats: I'm Chief AI Scientist at Facebook and also professor at NYU. What I did at NYU just before joining Facebook was create a center called the Center for Data Science. The purpose was to bring people from domain sciences like physics, chemistry, biology, social science, into a single place and collaborate to invent new methods based on machine learning and AI and statistics to discover and derive new knowledge from data. A lot of sciences are transitioning towards data-centric science, and I find that transition fascinating. Certainly in high energy physics, astrophysics, chemistry, genomics, that started many years ago, and we see the consequences in medicine and various other domains. Social science, I think, is the new frontier. The fact that we now have large sources of data on human behavior could revolutionize the way we do social science. A lot of this data is private, so it's difficult for scientists to access it. Facebook has an initiative that allows social scientists to access that data through Facebook itself, managed by Professor Gary King at Harvard. So yeah, I think we're going to learn more about humans through AI. One of the quests for AI is really understanding human intelligence. It's not just building intelligent widgets; it's also discovering what are the underlying mechanisms of intelligence and learning. If AI can help social science, psychology, and neuroscience, it's very important for science.