Research and startups: France’s strengths in the race for generative AI

Xavier Lazarus
5 min readSep 11


Photo de Thomas T sur Unsplash

Digital technology has its roots in the great mathematical philosophers of the 17th century. While Descartes, Leibniz and Pascal laid the foundations of computing three centuries ago, over time mathematics developed and engineering invented the sophisticated machines we use today.

For a long time, digital innovation stemmed from research, with applications that were first military and then professional, before finally integrating into common use. This cycle brought us the personal computer and the Internet. This innovation cycle changed at the dawn of the 21st century: digital innovation development arrived mainly via very large — typically American — companies, prioritizing public consumers, followed by commerce and government, while moving further away from a focus on research and mathematics.
The rise of e-commerce, the iPhone and social networks is the clearest example of this shift. It’s for this reason that Artificial Intelligence was initially developed on a massive scale by American big tech.

However, the Copernican revolution represented by GenAI (aka Generative Artificial Intelligence) spearheaded by chatGPT, could once again reshuffle the cards.

The development of GenAI is based on the confluence of capital, skills and innovation. Whereas historically, tech innovation relied heavily on the approaches of resourceful engineers, GenAI requires cutting-edge scientific skills, particularly in mathematics, and mastery requires extensive training and research. On this last point, France has a lot to offer.

France was a driving force in the birth of the Web, notably through Inria, one of the three founders of W3C, the World Wide Web’s regulatory body. Despite this, France has never been recognised as a major computer science research country. Our compatriots have only received one Turing Award, the computer science equivalent of the Nobel Prize, far behind the Americans who are the primary recipients the list of medal winners. [AB1] Yann LeCun, considered to be one of the three godfathers of modern AI, alongside Yoshua Bengio (Canada) and Geoffrey Hinton (United Kingdom), was the first French Turing Award recipient in 2019, for his work on deep neural networks, where the combined approach of mathematics and computer science reached its upper echelons. However, unlike the Turing Award, France places second in the list of most awarded countries for the Fields Medal, the Nobel Prize equivalent for mathematics. France is also well represented at the highest level in American Big Tech AI teams, with Yann LeCun as the long-term head of Meta’s AI department, before being replaced by another French man, Jérôme Pesenti.

We are on the cusp of a new era in tech, in which our French talent can excel on the world stage. It’s up to us to capitalise on these talents, so that France and Europe more broadly, can develop a powerful socio-economic ecosystem and so that we’re not excluded from the international podium, as during the first wave of global digitalisation.

This new wave of innovations align with the Schumpeterian vision: entrepreneurs will be the vectors of development and propagation. We see two main areas of development for GenAI startups: the foundation (models and their environment) and usage.

For the foundation of GenAI, it is important to understand that the current domination of models such as OpenAI and ChatGPT does not mean that the race is over. It is still possible to rapidly develop new LLMs (Large Language Models) alongside additional models that are more agile, less cost-intensive, independent and more respectful of data confidentiality. In fact, the centralised black-box approach offered by OpenAI, with its large energy consumption and expensive cost per usage, is not suited to all needs.

We will also see the development of infrastructure requirements linked to these LLMs: from management to optimisation of necessary computing infrastructures.

As for usage, the revolution appears limitless. We are facing a GenAI tsunami rather than a wave.

To support the emergence and growth of these new players, we will need a substantial amount of capital, as these models are not only expensive to train, but also high in cost of internal implementation given the infrastructure disruption generated. The support and growth over the last ten years towards developing the French Tech ecosystem will be key to elevating these future champions.

There is also a big question around the costs (economic, societal, environmental) of GenAI if we are to push back against Big Tech’s centralisation and privatisation of this new technology. At the heart of this issue will be the type of systems (open source, distributed or controlled) and data ownership, either by governments, companies or individuals. Finally, we will require more transparency from major corporates, SMEs and public services, to assist the large-scale commercial launch of our champions’ solutions. Not only will these initial questions and user feedback be key to getting these startups off the ground, but they will also enable our economy, our society and our government to fully embrace the progress that GenAI will undoubtedly bring, while managing the associated risks, particularly due to our technological dependence.

If we do not succeed in aligning these different interests — capital, modes of use and consumers — it is unlikely that a French national champion will emerge “alone in the face of global competition”.

While this is a classic recipe for innovation, responsible for Silicon Valley’s success, what we need above all is more talent, more science and more research. Our country needs to start by investing in training, particularly for the very young. Many students already have access to these tools in their private lives, and any attempt to resist them appears futile. We also need to train more PhDs, particularly engineers, in a wide range of fields and incorporate the needs and tools of GenAI into their research. We must continue to develop innovative entrepreneurship among these technical talents, to stay on the cutting edge and wi new science into useful and usable products.

Beyond the technology, there are clearly ethical issues and a lack of understanding of the limits of GenAI. Again, it is through a multidisciplinary approach and pursuing the scientific method that these questions can be best answered. Finally, we need to be able to address the new scientific frontiers that will inevitably emerge or become accessible which often appear through fundamental research.

For example, current GenAI systems need external memory and as they learn, they require external retention methods. The question is therefore to develop a memory base ‘internal’ to the model, in much the same way as the human brain integrates both its reasoning capabilities and its memories. Mastering technology will be one of the major challenges if we are to increase the effectiveness of GenAI.

Finally, if the goal of AI is to reproduce the human brain, then current systems are far away from achieving this, from electrical power to the amount of data training required. There will be new machines, or rather biological machines to invent, and new algorithms to get there.

GenAI is not a wave but a tsunami, and its effects will last. We are facing one of those moments in history when the world can be turned upside down, especially as the current economic crisis increases instability and offers undeniable opportunities for the establishment of new world orders. This time round, France is well equipped to be one of the driving forces behind this revolution. And if we do our job properly, the traditional Silicon Valley FOMO (Fear of Missing Out) will apply, but towards French startups!

This article originally appeared in L’Opinion on 16th August, 2023.