AI, the invisible tsunami. An evening with Laurent Alexandre

Laurent Alexandre analyzes the speed of automation in image industries. The economic stakes of the sector in the face of algorithmic computing.

Doctor, entrepreneur, and controversial voice on AI in France. An influential author and polarizing commentator, his stances on education, transhumanism, and human intelligence in the face of machines divide public opinion. In 2025, as generative AI transforms society and Europe seeks to regulate it, we interview this provocateur whose contributions, whether one agrees with them or not, enrich the debate on our digital future.

Mistral's most advanced model (Mistral Large 2) has just fallen out of the global LLM top 10 ("Artificial Analysis" benchmark). What is your take?

With a European Union that does not understand the real stakes of AI, Europe must work immensely hard to hope to return to the top rankings of the best language models. The excellence of Mistral's teams is not enough to allow the company to play in the big leagues over the long term.

We are witnessing a real economic war, with DeepSeek shaking up the Californian giants and OpenAI responding with its new model. The DeepSeek case is fascinating: they have reportedly caught up in a few months with what took OpenAI years to build. Does this call into question your vision of an AI race dominated by financial resources?

The Chinese team at DeepSeek accomplished a software feat, both in pre-training and inference. They leveraged approaches that had been considered in the United States, but abandoned at the time due to a lack of conclusive results. DeepSeek achieved a genuine breakthrough and an improvement to the Transformer model. On the other hand, they lied about their budget. They claim $5 million to train their model, when in reality, it likely required between $1.5 and $2 billion in GPUs. The question that arises is: did they smuggle in Nvidia H100 chips? Did they use Western data centers through cloud services? We do not know. What is certain is that this model was not trained on just $5 million. Furthermore, DeepSeek's primary paper is signed by 200 top-tier researchers. So, this is not a small startup operating on a shoe-string budget.

As a staunch advocate for educational reform in the face of AI, if you were appointed Minister of Education tomorrow morning, where would you begin?

I do not aspire to be a minister. If I were the Minister of National Education, my track record would be extremely thin. Just because I know a little about AI doesn't mean I could break neuro-genetic determinism. If educational technologies existed that genuinely reduced intellectual inequalities, helping less quick or less innovative children catch up with the most gifted, we would know about it. Yet, today, no educational technology truly works. School selects intelligent children; it does not produce them. In the era of galloping AI, and likely AGI within 18 to 24 months—even if any technological prediction remains risky—the question becomes even more serious. School faces two major problems. First, it does not know how to leverage digital tools to allow each student to reach their full potential—even if that potential is not unlimited. Second, and this is undoubtedly the most worrying part, it does not prepare children to go into professions where they will be truly complementary to AI. This second point is crucial. We long believed that AI would complement human intelligence. But in fields where this complementarity has been studied, such as medicine, we observe the exact opposite: AI + doctors is systematically less performant than AI alone. In other words, school, as it exists today, is producing workers who degrade AI results. It is imperative to rethink education with a clear objective: ensuring that "AI + humans" is more effective than AI alone. Otherwise, we are heading toward an accelerated "Yellow Vest-ification" of society, with major social unrest to follow. Investing heavily in education is therefore essential. Not because it will reduce inequality—it will only do so very marginally—but because it must limit the damage. School must take advantage of new technologies to help each child reach their own potential. But it will not increase this potential, because neuro-genetic determinism is powerful—and, undoubtedly, one of the greatest political taboos. Finally, it is urgent for education to seriously reflect on the complementarity between AI and human intelligence. Yet, in the only field where this question has been thoroughly examined—medicine—the results are anything but conclusive. Today, the education system is training doctors who are anti-complementary to AI. A perfect recipe for severe disillusionment. What we are seeing in coding will generalize to other professions. Many technical professions will be crushed by AI, and today we have no clear answers to offer young people regarding career guidance. School is not equipped to anticipate these upheavals: there are almost no AI experts at the Ministry of Education, and the situation does not seem likely to improve anytime soon. We are therefore facing a major educational crisis. A large portion of children are being sent down dead-end paths, with the exception of those from the elite, who will possess what will become the most valuable skill in the age of AGI: broad general knowledge. This ability to grasp complex subjects will remain an asset when facing AI. I therefore remain very humble about my answer: we are heading straight toward educational chaos. AI is evolving so fast that any strategy defined today will be obsolete in six months or a year. We would need a mobile and responsive education system, which is currently not the case. For a long time, I have been saying that we are heading toward a cognitive crisis: society has neither the tools nor the capacity to regulate the tsunami of artificial intelligence that is about to submerge us. This applies to primary and secondary schools, universities, and adult continuing education. The problem is simple: by the time an adult is trained for a job of the future, AI will have already made that training obsolete. For the middle