The CNRS aims to attract the AI talent of tomorrow
The organisation's 'Choose France CNRS AI Rising Talents' programme is intended to attract the best researchers in this burgeoning field. Massih Reza Amini, the deputy scientific director of CNRS Informatics, explains how.
Could you explain the thinking behind the 'Choose France CNRS AI Rising Talents' call?
M.R.A.: The 'Choose France CNRS AI Rising Talents' programme started in 2020 and offers talented young researchers in France and abroad a unique opportunity - to launch and lead a four-year artificial intelligence (AI) research project in a CNRS laboratory. The programme is open to researchers under the supervisory authority of any of our Institutes whose projects focus on AI, including AI for science.
The initiative is co-financed by France's Ministry of Higher Education and Research and the 'IA' PEPR research programme co-steered by the CNRS. This is the fourth annual call and it is considered of such strategic importance for the CNRS that the organisation's Chairman and CEO Antoine Petit played an active role in drafting the text of the call which will be subsequently coordinated by CNRS Informatics.
One of the main novel features of this year's call is that it's being opened up to talented young French researchers with an international background, the aim being to attract and hopefully retain such talent in France. Previous editions of the call were limited in that only projects focusing exclusively on core AI topics like machine learning were accepted. The second new feature of this year's programme's is that projects at the interface with other scientific disciplines can now apply which is specifically intended to support the development of projects of this sort. Finally, the programme can offer selected projects a competitive budget of up to €1 million each. The call opened on September 9th and will close on March 31st 2025.
The 'CNRS AI Rising Talents' programme is part of France's National Strategy for Artificial Intelligence (SNIA). Could you explain more about this strategy and the issues it involves?
One of the main issues involves avoiding a brain drain to other countries or technology hubs. Right now we're in the middle of a worldwide AI race, with countries like the United States and China investing massively. So it's crucial for France to stay in the competition because if we don't we'll end up marginalised as regards scientific and technological progress in the 21st century.
The National Strategy for Artificial Intelligence (SNIA) was launched in 2018 to strengthen France's position in AI following on from the Villani report. The SNIA is based on four main pillars - a massive investment of €1.85 billion that was used to fund AI research projects and infrastructure between 2018 and 2022; encouraging collaborations between the academic sphere and industry to drive research; attracting talent; and finally bolstering international cooperation. This strategy is currently going into its second phase which provides an additional €1.5 billion of funding for projects between now and 2025. This gives an overall total of over €3 billion.
How is the CNRS involved in this race?
The CNRS has made a proactive commitment to the development of AI through our support for interdisciplinary research initiatives, more targeted projects and by managing intensive computing resources for AI. There are several flagship actions that illustrate that commitment well:
Firstly there's the AISSAI Centre (AI for Science and Science for AI) which is a cornerstone of the CNRS's AI strategy. Its main objective is to stimulate interdisciplinarity by bridging the gap between AI researchers and their counterparts in other scientific fields. The idea is to develop collaborative models in which AI serves science with science also helping to develop AI. The AISSAI centre creates links with renowned international institutions like IVADO in Quebec, Stanford, MIT and the University of Chicago in the US and finally European universities like Berlin and Cambridge. This kind of international cooperation is essential for us to remain competitive on the global stage.
The CNRS has also invested in four Interdisciplinary Institutes for Artificial Intelligence (3IA) created in 2019 in different parts of France. Their mission is to structure research around AI's basic foundations and practical applications, to train researchers and engineers and finally to work on technology transfer and cooperation with industry. The programme's initial budget of 80 million euros has driven the emergence of centres of excellence. Today, another five centres have been set up, forming a national network or 'AI clusters' with total overall funding of €350 million.
The CNRS has also set up a research support system with its PNRIA Engineers' network made up of around twenty specialist AI engineers. These technical experts are recruited straight after their studies for these key roles supporting researchers in their projects. One notable example is in collaboration projects working on the Jean Zay supercomputer which the CNRS operates. This network has enabled an attractive new salary scale for data scientists to be created that is competitive on the job market so we can help more engineering talents to stay in France.
The Jean Zay supercomputer is one of Europe's most powerful calculation machines and is a major asset for the development of AI at the CNRS. It boasts 126 petaflops1 of computing power and means researchers can learn more about extremely complex and large-scale AI models, like large language models or LLMs. This infrastructure needs to be regularly updated to remain cutting edge and thus guarantee that French scientists have the right resources to compete with the best research centres around the world.
Finally, the CNRS continues to invest in collaborative initiatives like the PEPR IA (Priority Research Programme and Equipment in Artificial Intelligence). The aim of this national research programme is to drive synergies between different and varied AI research stakeholders in work on ecological sustainability issues. The focus is on involving more mathematical fields, developing new hardware architectures or computing paradigms, system security and robustness and finally explicability and interpretability.
AI's rapid development of AI has brought up a lot of questions. Currently, what are the main research subjects in this field and how is the CNRS working on these?
The artificial intelligence boom has indeed led to a lot of questions being asked, particularly involving ethical, ecological and societal issues. Initially, AI made people frightened because of science fiction-style scenarios of AIs taking over our societies but such fears have tended to fade with AI now seen as a decision-making tool rather than as actually likely to replace humans. Nonetheless there are still major issues to deal with, particularly involving the transparency of decisions taken by these systems, the explicability and interpretability of the models, and the ethical responsibility of using them.
Training large-scale models like LLMs also involves major ecological issues because the systems involved required considerable energy resources. This means it's crucial to adopt a sustainable and responsible approach to developing AI.
Currently, there are three other major issues in AI research. Firstly, there's the critical challenge of mastering foundation models capable of generating texts, images or videos, particularly as regards the quality and veracity of the information produced. Secondly, there's the challenge of making sure the decisions these models make can be explained because these systems can't always justify their choices transparently, unlike humans. Finally, ethical concerns remain about using these models along with the environmental issues associated with their development that we discussed earlier.
AI is multidisciplinary by nature as it involves fields as diverse as computer science, mathematics, physics, neuroscience, psychology, language science, as well as the humanities and social sciences. This multidimensionality is exactly what makes its development so crucial and stimulating for the future.
- 1126 million billion floating-point operations per second.