Research

Table des matières

Research in Artificial Intelligence

AI is transforming our societies, but AI, as a paradigm for knowledge discovery, also has an impact on how research is produced across all disciplines. AI methods, particularly machine learning, now represent a major challenge for modeling increasingly complex and multi-scale systems, as well as for exploiting the massive increase in the flows, volumes, and diversity of multi-source data from large instruments, observational systems, experimental platforms, etc.

This leads to new research practices that vary widely in terms of maturity and organization from one discipline to another, and to new needs in terms of expertise and international collaborations. Within CNRS, AI methods are now rapidly deployed as part of major research infrastructures and national and international experimental platforms that CNRS supports or contributes to (for example, in biology, high-energy physics, climate modeling…).

The aim of the AISSAI center is to promote dialogue between disciplines, tackle new scientific questions, and establish radically new modes of collaboration between sciences. 

Research fields

Artificial intelligence (AI) is a field of research in its own right, but it is also a gas pedal of scientific discovery in many fields, provided that it fosters closer ties between different scientific communities. Thanks to its national and multi-disciplinary dimension, the CNRS must play a leading role in fostering these interactions between disciplines, which not only concern the appropriation of AI methods (particularly machine learning), but are also necessary for the removal of obstacles specific to this field on subjects as strategic as explicability, the incorporation of expert knowledge or frugality.
AI is the latest of the profound changes brought to our societies by digital science and technology since the second half of the 20th century (after the computer itself and the Internet). AI draws on the availability of vast datasets related to almost all human activities (from health to education, business to science), and takes advantage of recent conceptual and practical advances in machine learning and hardware-accelerated computing devices.

However, despite recent successes in this field, today’s most powerful algorithms – deep neural networks to name them – face numerous challenges and suffer from serious shortcomings that call into question their deployment in sensitive and critical areas with a high societal impact. These shortcomings and challenges include information leakage and privacy preservation, social biases in training data, lack of explicability, accountability or robustness, but also their ecological impact. This raises questions about the social responsibility of these algorithms and, more profoundly, about the future we are helping to shape.

The CNRS, France’s leading multidisciplinary basic research organisation, firmly believes that it is essential to approach AI from a holistic point of view. Indeed, all sciences, including the humanities and social sciences, are either impacted or reshaped by AI, but also offer new opportunities that can contribute to understanding the possible consequences of AI and ultimately improve the design of AI-based tools. Scientific bodies such as the CNRS have a major responsibility to help policy-makers and public opinion understand the implications, benefits and possible risks of AI.

In the context of the national AI strategy launched by President Emmanuel Macron, the CNRS has been involved in several actions, with AI chairs, participation in the four selected Interdisciplinary Institutes for AI (3IA) or actions to attract the best talent (Choose France program) in a fierce international competition, and the creation of the AISSAI center on which we will focus shortly. CNRS operates the Jean Zay national supercomputer for AI, which has become one of the most powerful computing facilities freely accessible to public AI research in Europe. CNRS is also coordinating, with other university partners, the new PEPR in AI, a priority research program on the foundations of AI, focusing on reliable, integrated and decentralized AI.