“Universities are ideally placed to train students in Artificial Intelligence”


On October 2, the Paris Artificial Intelligence Research Institute, known as PRAIRIE, opened its doors at Paris Dauphine - PSL, which was a primary impetus in its founding. As an official French Institute of AI, whose members include the CNRS, Inria, Institut Pasteur, Université de Paris and PSL with support from industrial partners, PRAIRIE aims to become a center of excellence for artificial intelligence in Paris.
Here, three PSL researchers bring us up to date on the state of our current knowledge and the scientific challenges in store as well as current and future educational programs in artificial intelligence and data management.

Chloé Agathe Azencott (MINES ParisTech - PSL), Jamal Atif (Dauphine - PSL) et Pierre Senellart (ENS - PSL)

An interview with:


PSL: How would you define Artificial Intelligence?

Chloé Agathe Azencott: The field of artificial intelligence is devoted to building systems that “imitate” cognitive functions that we, as humans, associate with being living creatures. One essential component of artificial intelligence is machine learning, or statistical learning, in which we construct algorithms that can independently use data to determine how to model a phenomenon or behavior.
Jamal Atif: What you have to keep in mind is that it’s a multidimensional field of study in constant flux. It’s not easy to give a precise definition of artificial intelligence, but I’d agree with Chloé-Agathe’s suggestion. It’s the ability to give machines what we would consider a degree of intelligence and the ability to resolve complex problems.
Pierre Senellart: I like to refer to the Turing Test, which is the thought experiment that Alan Turing proposed in 1950 in response to the question of whether machines can think. A system with the ability to behave in ways that are indistinguishable from how a human would behave during a complex interaction with another human – that’s artificial intelligence. Of course, we’re not exactly there yet! A number of problems have been identified as probably being AI-complete, meaning that we can only solve those problems completely if we have artificial intelligence in the way Turing defined it. Examples of that include computer vision, fully self-driving cars, an understanding of natural language, solving problems in the face of unexpected events and so on. Today, the goal of research in artificial intelligence is to advance our knowledge of those various fields, as well as their mathematical and computer underpinnings, so that one day we can create true artificial intelligence.

PSL: In your view, how and why should universities embrace the task of training students in artificial intelligence?

Jamal Atif: If there’s any area of research these days that’s critical in multiple respects, insofar as it’s the object of intense interest within the digital industry and among governments, it’s artificial intelligence. By its very nature, it has implications not just for research and education but for society at large as well. Already – and this will be even more true in the future – some decisions about whether to make a purchase, whether to issue a medical prescription and so on, aren’t and won’t necessarily be understood by humans. We’ll need to educate and inform members of the community about those issues, both now and in the future, through undergraduate education, doctoral study and executive education. The only place where all those forms of education are available and informed by ongoing research is at a university.
Pierre Senellart: The fact is we’ve made some impressive progress recently in multiple fields that we’ve thought of as AI-complete. For example, we haven’t entirely solved problems like image classification and machine translation yet, but modern techniques are yielding some very good results. We’ve really seen some profound changes, driven by methodological advances in particular (such as deep learning techniques) but also by our access to very large quantities of data and new computing architectures. We need to train students in those new techniques, in part because they’ll generate plenty of practical applications in the industrial world and the rest of society, but also in order to refine those techniques and make further progress toward artificial intelligence. Universities, with their research professors at the cutting edge of research in those fields, are ideally placed to train those students.

PSL: How do today’s PSL students learn about artificial intelligence?

Jamal Atif: Currently at PSL, AI is being taught to a wide range of students, from the undergraduate to the PhD level. That instruction can take various forms, depending on each student’s profile and educational level. So for example, in order to educate future business executives about the challenges posed by AI, we organized a digital semester for undergraduates in the organization science Bachelor’s program (Dauphine - PSL). For students with more specialized backgrounds, we’ve created a new Bachelor’s degree in Mathematics and Computer Science for Decision-Making and Organizations (Dauphine - PSL). Meanwhile, engineering students at ESPCI Paris will have a completely new curriculum in AI along with a new instructor, since a PSL professor position has been established. PSL Master’s degrees now also include state-of-the-art programs that have proved highly popular with students, such as the new Artificial Intelligence, Systems, Data academic path and the MASH program in Machine Learning and the Humanities. In executive education, the executive Master’s degrees in Artificial Intelligence and Statistics and Big Data and the Data Sciences Certificate Program at Dauphine - PSL have all been a big success. We could go on giving examples, but the point is that education in Artificial Intelligence is a major focus of our attention and will be a significant component of our new course offerings.
PSL will soon be offering a PhD track for the university’s various Master’s programs in artificial intelligence. In particular, we’re working on setting up a cross-disciplinary AI/DS (artificial intelligence/data science) program for the 2020-2021 academic year that would encompass all of the students at the Master’s and PhD levels in each discipline. With that program, which will be included in PSL’s new graduate programs curriculum, students will be able to choose a major and minor along the lines of, say, “Biology/Humanities.” As a forerunner of that, this year Gabriel Peyré and Léa Saint Raymond at ENS have created a research seminar on AI and the Humanities that we hope will be an example of a major or minor, depending on the background of the students. We’re currently working on organizing an intensive, week-long course on “Artificial Intelligence in the Humanities” for early April that will be open to all PSL students. Our aim will be to offer similar courses at least twice a year, so we can disseminate knowledge of artificial intelligence more effectively in every scientific field that generates data.

 Interview de Bruno Bouchard (Dauphine - PSL) sur le programme Dauphine Numérique

PSL: What do you think are the major scientific challenges in artificial intelligence? And how will the 68 researchers and research professors that do work in AI at PSL, including 26 with ERC grants, will be able to meet those challenges?

Chloé-Agathe Azencott: There are so many challenges! It’s still a young discipline. Right now AI is at a turning point in terms of:

  • the computing power required, the environmental impact of putting it to use, the ability to implement our current solutions on embedded systems;
  • algorithmic biases and how we can quantify them;
  • applying these techniques to noisy, scarce, heterogeneous data that’s difficult to gloss.
Inauguration de l'Institut PRAIRIE © Inria

Opening day for Prairie at Dauphine - PSL ®Inria

PSL: What will you be doing at PRAIRIE?

Jamal Atif: I’m the Deputy Scientific Director for Education. My role will be to tie in PRAIRIE’s educational programs with the PSL curriculum.
Chloé-Agathe Azencott: I hold a junior chair to develop methods for generating wide-ranging genomic data so we can discover biomarkers, meaning areas of the genome that can explain a characteristic we see at the macroscopic level. There are numerous potential applications, such as precision medicine, the discovery of risk factors and our understanding of biological mechanisms. In addition, as co-founder of the Paris branch of Women in Machine Learning and Data Science, I plan to be playing an active role at Prairie in promoting diversity in education, hiring and representation.
Pierre Senellart: My research involves data management: how can we model, organize, process and analyze large volumes of data in an effective, relevant way. In some ways you might consider it the “memory” aspect of artificial intelligence. However you conceive of AI, you need a way to store, access and explore a huge quantity of data. At Prairie I’ll be working more specifically on managing uncertain, imprecise and heterogeneous data. As a research professor, I’ll also be overseeing a number of courses for the new Artificial Intelligence, Systems, Data Master’s program at PSL, which is being jointly funded with PRAIRIE.


Présentation des deux chaires de recherches PRAIRIE "Intelligence Artificielle" et "Santé" à MINES ParisTech - PSL

 Le rôle clé de l'Institut Curie (membre associé de l'Université PSL) dans le développement de l'IA