[CONVERSATION] Juliet Matteoli: Artificial intelligence, a new growth lever for economic intelligence? [1/2]

EI Portal (PIE) : What is the definition of your artificial intelligence? Is it possible to present a simple and “obscene” idea?

Juliet Matioli (JM) : AI provides cognitive power in an artificial system. These various cognitive abilities are consistent with perception, learning, abstraction, reasoning, decision making as well as communication and action.

We find two instances which are conflict but which in my opinion will hybridize. On the one hand, data-driven AI, conventional through learning, adapted to the world of computer vision (artificial intelligence techniques for image analysis) perception and prediction. On the other hand, there are more symbolic examples of AI, which are based on knowledge. It covers combinational optimization, planning, multi-parameter decision making techniques but also includes knowledge engineering along with ontologies. These two paradigms often contradict each other when I think they are complementary: a combination of the two approaches to hybrid AI is needed to solve a complex problem.

PIE : How do you define “machine learning”, “deep learning” and what are their differences?

J.M. : Machine learning, also known as “automatic learning” in French, includes technologies such as deep learning or reinforcement learning. Overall, learning is learning from examples, much like a child learns to recognize apples by looking at examples of apple varieties.

Today, there are many technologies based on neural networks. Thanks to linear algebra we can infer these models. Deep learning is a specialized network with many layers, which makes it possible to do better than the neural networks of the 90’s. For example, this technology enables one to recognize faces and this is especially true of this application which has gained its reputation. .. However, in order for it to work well, many, many examples of this technology are needed.

There are also increasingly complex technologies that require work from examples, less static than images, such as time series. We can cite CNN (Convolutional Neural Network).

PIE : What are the main tasks that AI currently uses? AI is strategic for which sector and why?

J.M. : AI can help solve problems in all cases. It is significantly present in all verticals of Thales. In aeronautics, for example: it helps air traffic controllers better estimate landing delays. In the case of railways, this makes it possible to identify obstacles. Earth can be observed from space to identify the effects of climate change, identify targets, or plan defense activities.

AI is also used in engineering, such as to design test batteries automatically or to capitalize and manage business knowledge through knowledge engineering. Similarly, for production, AI optimization makes it possible to develop Industry 4.0 In the case of supply chain automotive, AI applies to autonomous vehicles, in health and agriculture, not a single field escapes AI

However, it will not be able to solve all the problems. You need to be able to show its extra value compared to the solution based on more “classic” technology.

In short, many application areas of Thales (and elsewhere) use and continue to use AI, whether based on machine learning, symbolic AI, or hybrid AI.

PIE : AI has been a popular topic of discussion for several years. What’s new today? Has anything changed since its popularity began?

J.M. : AI was already popular more than thirty years ago, for example, AI defeated Kasparov in 1996. Since then, a number of things have changed: Computing power 7 in the first place It allows you to do more complex things and much faster. The availability of information has made it possible to design new approaches such as deep learning. The Internet and open data also contributed to the renaissance of AI. These innovations led to a direct explosion of database-based AI systems, especially thanks to the work of LeCun, Bengio and Hinton.

In my opinion, year and year algorithms have enabled library capitalization (skit-learn). The advent of the semantic web has also democratized scientific work (Google Scholar, etc.) which has allowed to break the scientific silo. All of these, for example, lead to a reduction in barriers between AI scientists and health professionals

As a result, the exchange of knowledge is much more fluid. Finally, much more money is injected, especially through GAFAM and of course through the video game industry, making AI known to a wider audience.

PIE : What is the position of France in this regard today? What are our strengths and weaknesses?

J.M. : In 2018, France positioned itself in AI for the sake of humanity (i.e. moral AI), especially through the mission of Cedric Villani. This has given a great impetus to academic research.

At the time, major French manufacturers such as Renault, Michelin, Saffron and Thales were not too concerned about the strategy. Some then sign The IA Manifesto in 2019, today brings together 15 French manufacturers. Indeed, trusted AI was their concern. The government heard this wave of emotion. Later the Grand National Challenge of Trusted AI or Grand Challenge of AI in Health was launched. Today, the strength of France lies in being the leader of opinion on trusted AI, i.e. AI for mission-critical systems.

The main weaknesses of France in this case are based on the fact that various professionals in the field do not really know how to sell themselves despite having real knowledge. Direct Consequences: The French are not very visible internationally and if they are, it remains an individual success around an individual, a lab or an entity, but not a fully recognized French ecosystem. Second, in terms of salaries or research teams, we don’t have the same options as China or the United States. Finally, the education of AI is not necessarily well organized and remains unequal among the various training courses available. In short, France must bet on training AI, visibility and “knowledge”. However, it is important to emphasize that there is a “know-how”, because we have very bright scientists in this field.

Pie: We often hear about brain drain, what do you think?

J.M. : Whether in favor of GAFAM or BATX, both recognize the French structure as well as its talent, mentioned above, as most were installed in France. Their research center (FAIR for Facebook, Huawei), one of the main sources of this brain drain is the problem of salary attraction. The salary of a CIFRE PhD student and the salary of a Facebook PhD student have about two multiplier factors.

The AI ​​manifesto mentioned above is also interested in the problem of attractiveness: how to keep one’s talent in French art. A working group is also studying the issue and an event was organized in November 2021, inviting all engineering schools in France to discover French industries that have major problems with AI and to remedy this leak.

Pie: How can France take advantage of this technological advancement?

JM: France must continue to innovate, make itself more visible, file patents and build open source around AI so that France’s uniqueness, its specialization in trusted AI, its ethics for critical systems, its healthcare and B2B’s intellectual property This culture of property can be developed.

Interviewed by Yassin Ivalitin For Club Information-intelligence

The second episode is 06 May

For the future:

Leave a Comment