Bertrand Braunschweig and artificial intelligence
© Inria / Photo G. Scagnelli
A few days ahead of the publication of the Villani report on artificial intelligence, Bertrand Braunschweig, coordinator of an AI white paper published in 2016 and head of the Inria research centre in Saclay Île-de-France, reviews the work being done on the subject.
Inria published an artificial intelligence white paper in 2016. Since then, have the generic challenges for AI changed?
The major challenges don’t change that fast! In the white paper, we included subjects on which our teams are working and which will drive research for many years to come. One need simply look at our new scientific strategic plan for the period 2018-2022, which identifies a number of the AI challenges. With the spectacular growth in interest for AI and with the development of its applications in all sectors, these challenges have become even more important.
I would also point out the growth in the questions of verification, validation, certification of AI systems and their corollary, the ability to produce explanations that are understandable to the users: behind these technical subjects lies the problem of confidence in AI, one which will become a major issue if not dealt with correctly.
Robotics and self-driving cars are among the two AI fields attracting the most attention. Do you think that they will remain central to the discussion?
Le robot humanoïde iCub - © Clotilde Verdenal / LoeilCreatif
I consider that the self-driving car is a robot, so as far as I am concerned, these two areas are inextricably linked. Yes, obviously, when you look at the colossal investments being made by the car makers, equipment manufacturers and suppliers of mobility solutions, it is highly likely that this field of AI application will remain central for several decades.
But alongside this, there are other enormous markets, in particular that of the digital society, which was historically the first to bring AI in from the exile to which it had been banished in the 1990s and which has major economic, technological and social implications.
More generally, we can see that all economic sectors are gradually adopting AI, starting with health, which has understood the considerable progress in the diagnosis and treatment of diseases that can be expected, but also the energy, finance/insurance, defence and security, environment, games and leisure and other sectors. There are no longer any sectors that will not be impacted by AI in the medium term. For example, just look at the list of winners of the “Innov’up Proto IA et Robotique” competition in the Île-de-France region, reflecting the considerable diversity in AI today. At Inria, we are receiving increasing numbers of collaboration proposals from companies in all fields.
There are no longer any sectors that will not be impacted by AI in the medium term.
Have other AI applications emerged in the past two years? Is it possible to draw any conclusions from these two years?
Apart from the economic impact I have just spoken about, we could mention some spectacular achievements and major international announcements.
Some spectacular achievements in the past two years have hit the headlines: for example AlphaGo Zero by Google DeepMind, which demonstrated the power of reinforcement learning by scoring a further victory against the world’s best players; or Libratus, the artificial poker player from Carnegie Mellon University, which beat human champions in a game where the adversary’s hand is not known, unlike chess and go for instance. Others did not make the front pages but I consider them to be just as important: optical flow labelling by Nvidia through deep learning which, at the end of 2017, obtained better scores than all conventional technologies, a bit like in 2012 when a convolutional neural network beat the traditional competitors in the ImageNet campaign; or the progress made by the French start-up Therapixel which last year beat 1200 international teams in breast cancer recognition on mammograms.
These past two years have seen significant international announcements, firstly China’s AI plan which devotes enormous resources to its aim of becoming the world’s AI leader by 2030, which is actually credible given the current situation and the potential of this great country; our Canadian friends have continued to invest in three key AI centres (Montreal, Toronto and Edmonton), for example very recently with the announcement of the Scale. AI initiative of more than $200M for AI in logistics chains. We have seen may strong declarations about the Franco-German couple, in particular from our German neighbours, but on this point, I believe that we should wait for the publication of the report by Cédric Villani this Thursday. On the other hand, there is clear reticence on the part of the European Commission, which I hope will soon be a thing of the past, at least in the preparation for the 9th R&D framework programme.
What do you expect from the Villani report being published on next 29th March ?
© Inria / Photo G. Scagnelli
I’m extremely impatient! We submitted the #FranceIA report to the previous Government more than a year ago now and at the time François Hollande had promised 1.5 billion euros to support AI in France. At the beginning of 2017 we already considered that it was urgent for our country to invest massively in artificial intelligence, because the others wouldn't wait for us; I therefore hope that the publication of this report will be accompanied by support decisions on a par with our ambitions.
To talk a little more about the content and taking account of the broad outlines of the report that Marc Schoenauer (DR Inria Saclay, scientific adviser to the Villani mission) and Cédric Villani himself sketched out in the media in recent weeks, I hope that the measures adopted will enable us to remain on an equal footing with our international partners and competitors in this field. We need research support in order to further amplify our actions and measures to encourage the transfer of this research to companies.
I also hope that these measures will draw on the specific features of our country and our society, which emphasise questions of ethics, trust, respect for privacy, security: not only is this a moral obligation, but it also represents a competitive edge because by working on these subjects, we can set an example and assume leadership in what could constitute the essential precondition for the deployment of AI in the economy, worldwide.
Finally, I also hope to see support for deployment of AI infrastructures for computation and data sharing: I was joint coordinator for the working group of the Allistene alliance, which proposed setting up a major national facility for artificial intelligence (GENIAL) so that our researchers and our companies would have access to computing tools but also to shared data enabling them to keep up with the best, internationally. I hope that this will be one of the announcements made this week.
Finally, several key areas for AI applications, such as those I mentioned earlier and which were covered in the interim communication from the Villani mission last December (health, transport, environment and the defence-security sector), deserve special support in order to develop the competitiveness of European companies and I hope to see tangible measures in this respect, notably in partnership with our German and European friends.