Artificial intelligence

Yann Le Cun, a leading figure in artificial intelligence, offers a vision that challenges the most alarmist narratives surrounding the technology. In his view, claims focused on existential risks do not stem solely from scientific or ethical concerns. They can also serve commercial strategies designed to slow the development of open source and strengthen the dominance of a handful of private players. Faced with this concentration of power, he urges Europe to build its digital sovereignty on two foundations: transparency and knowledge sharing. This means avoiding overly restrictive preventive regulation that could stifle European initiatives before they have a chance to mature. Yann Le Cun also points out that today’s language models remain technically limited. Their ability to manipulate words does not mean they genuinely understand the physical world. His research into World Models is specifically intended to overcome this barrier and could usher in a new stage in the evolution of AI. This approach reframes the debate: the challenge is no longer simply to improve systems capable of producing language, but to design intelligent systems able to form representations of their environment. Yet this transformation must not remove human beings from the decision-making process. Here, artificial intelligence is conceived as a means of augmenting human capabilities, while citizens retain control over decisions and the ethical framework governing them. The sovereignty he advocates is therefore rooted neither in isolation nor in fear. It grows from open access to knowledge, Europe’s ability to develop its own technologies, and a collective determination to keep these tools at the service of people. It is a path that brings together scientific ambition, digital independence, and democratic responsibility.

Faced with the darkest predictions about artificial intelligence, Andrew Ng offers a decidedly different interpretation of the transformation now underway. In his view, the alarmist narratives promoted by some large companies do more than express technological anxiety: they can also serve corporate interests by encouraging regulations restrictive enough to curb competition. AI, therefore, would not trigger a jobs apocalypse. Instead, it would act as an economic complement, increasing the value of human judgment rather than making it obsolete. In this new environment, the advantage would not simply go to those who use an AI model, but to those who possess detailed knowledge of their field and know how to integrate it into automated workflows. This “contextual advantage” turns a tool available to everyone into a genuinely distinctive capability. Andrew Ng therefore urges students and professionals to become “AI native,” meaning they should develop both the technical skills needed to work with these systems and the domain expertise that gives meaning to their outputs. He encourages everyone to prepare actively for this new market, where innovation will depend on this dual command of technology and professional knowledge. Yet his argument does not idealize today’s models. He notably highlights their limitations when it comes to deep, long-term learning. This reservation does not weaken his central message: rather than being paralyzed by unfounded fears, people must learn to build, experiment, and combine automation with genuine expertise. The future would then belong less to passive users of AI than to those capable of directing its possibilities through their understanding of context, their command of the tools, and the quality of their judgment.

In a lecture at Stanford University, Andrew Ng and Lawrence Moroni examine how artificial intelligence is transforming technical careers and product development. Their conclusion is straightforward: by accelerating code production and lowering its cost, AI is gradually shifting value toward the stages that come before implementation. Programming skills still matter, but the ability to understand a need, write clear specifications, and design a relevant product is becoming more decisive. For engineers, this change means developing genuine empathy for users so that speed of execution is not mistaken for real usefulness. It also means building strong professional networks, which can prove invaluable when navigating a rapidly changing job market. The speakers nevertheless stress that AI-generated code does not eliminate technical challenges. On the contrary, it can create substantial technical debt when architecture, maintainability, and quality are neglected. The goal, therefore, is to adopt a pragmatic approach that pays as much attention to business value as it does to the product’s technical foundations. The discussion also highlights the rise of small AI models, which open up new possibilities without necessarily pursuing ever greater computing power. In this shifting environment, Moroni encourages future professionals to remain authentic, rigorous, and capable of stepping back from industry hype. This mindset helps distinguish lasting applications from passing trends while maintaining professional standards despite the accelerating pace of new tools. Success no longer depends solely on mastering a particular tool or programming language, but on the ability to connect technology, human understanding, and economic purpose. AI makes building easier, but above all, it places greater responsibility on people to decide what is genuinely worth building.

Researchers at Harvard and MIT have unveiled MatrAIx, a platform capable of simulating 8.3 billion profiles, effectively creating a digital representation of the entire human population. Its ambition is to enable companies and institutions to test products, interfaces and even policies on artificial intelligence agents without involving real people. Each virtual profile is assigned specific social and psychological characteristics designed to reflect the diversity of human behaviour. At first glance, the promise is considerable: market research could become faster and more efficient, while designers would gain access to a vast experimental environment before deploying anything to the public. Yet this synthetic humanity also raises profound ethical questions. Can a simulation, even one conducted on such an immense scale, truly capture the unpredictability that defines individuals? The widespread use of these agents could above all create a closed loop in which artificial intelligence designs, tests and evaluates the very solutions that will later be offered to humans. MatrAIx also opens the door to more troubling applications. An extremely detailed algorithmic understanding of behaviour could make it easier to manipulate choices or encourage automated price discrimination, with offers and prices tailored to the anticipated reactions of each profile. Behind the promised time savings, then, lies a broader transformation: market research would no longer seek merely to understand individuals, but to reduce them to a vast exercise in statistical optimisation. MatrAIx therefore raises a question that reaches far beyond product testing: what remains of our individual uniqueness when our preferences, hesitations and decisions appear reproducible through billions of virtual counterparts?