Publications — IA

Humanoid robots attract most of the attention, but many of the demonstrations are preprogrammed, remote-controlled or generated by AI. Physical AI, meaning machines that perceive, decide and act in the physical world, is already a market of about $18 billion, of which humanoids account for only $2–3 billion. The report explains why value will first come from specialized systems, describes the state of the technology and the positions of China, the United States, Europe and Japan, and lays out future scenarios and what executives should do today.

Les grands acteurs de l’IA promettent que ses bénéfices pour le climat dépasseront son empreinte. Cédric Villani et moi en doutons : l’usage quotidien des modèles concentre l’essentiel de l’énergie et de l’eau consommées, et la pression se concentre localement sur les réseaux électriques et les ressources en eau. L’IA aidera pourtant à s’adapter à un monde à +3 °C. Encore faut-il la mesurer : nous demandons des données comparables et auditables.

AI seems almost weightless to its users, but every model rests on an industrial backbone of minerals, manufacturing, electricity and water. The report examines three vulnerabilities that businesses have to take into account: environmental impacts, energy supply and the grid, and the choke points of compute infrastructure. It sets out the critical uncertainties for the future and strategic recommendations. The study is based on three months of work and more than 50 expert interviews.

AI has been used in research, development and innovation for a decade, and the arrival of generative AI and large language models has sped it up over the last two years. Yet its integration into R&D&I processes is still immature, especially where results must be error-free. The report looks at what AI can do at each step of the process, how to make its implementation succeed, the landscape of tools and providers, and future scenarios, and closes with six strategic actions. The study was carried out with LVMH, Avril, GRTgaz, CNRS and INRAE, from 40+ interviews and a survey of more than 200 companies and institutions.

ChatGPT is only the tip of a much larger generative AI iceberg. This report explains how AI became more human, what generative models can now do, and how they change the value of different kinds of human intelligence inside companies and beyond. It covers the impact on business, the value chain and competition, the limits and risks, and the critical uncertainties, up to the question of artificial general intelligence. Two interludes give the floor to an artist and a philosopher. The study combines in-depth research, an expert survey and interviews with players from across the AI ecosystem.

In January 2023, Laurent Alexandre was telling me that AI was dumb as a rock. Two months later, he had changed his mind. We explain how large language models work and the emergent properties that may lead to AGI. Three uncertainties make the future hard to predict, but risks and opportunities are already tangible. Companies that sense new developments early and respond fast will have an edge.

Tens days after its release, one of my daughters, tired of waiting for my help with her physics homework, asked ChatGPT instead, and got a flawless guidance. Written by me and proofread by an AI, this article explains what GPT is, how it is already disrupting business, and what its limits and risks are. My conclusion: AI will not replace you, but a person or an organization using AI will.

I co-wrote this article with an AI, after posting on LinkedIn an article entirely generated from a single prompt the day of ChatGPT’s release. It looks at how ChatGPT and AI are set to disrupt business, in the short and long term, and at the challenges of adoption, from data privacy to bias. My take: AI will augment humans rather than replace them. The follow-up, “My kids have replaced me with ChatGPT”, is written by a human.