At the entrance to the auditorium, there was no longer the classic headset service for simultaneous translation, where it was even exciting and somewhat scary to leave your ID or Passport in exchange for the “ gadget ” with the headphones… Instead, I found a QR code inviting attendees to download an application that allowed them to follow the presentation in real time, with a synthesized voice and a surprising range of languages —including Catalan, and not with a beta test accent, but with an impressive naturalness. I thought that this small gesture symbolized quite well how far generative artificial intelligence has come: from research to service, from the laboratory to our daily lives . But it also raised an uncomfortable question: if the machine can do this so well, what will be left for us to learn?
We are living in a time when the fascination with artificial intelligence seems unattainable. In just two years, large language models have transformed expectation into vertigo and placed this technology at the center of the global economic and political debate . Behind the media noise, however, there is a deeper structural change: AI has ceased to be a tool to become state infrastructure. The world’s major powers are investing trillions in it with the same urgency with which energy or military resources have historically been protected. According to the Stanford AI Index 2025, global investment in generative AI exceeded 33.9 billion dollars in 2024 , while the cost of inference of models has been reduced 280 times since 2022. This combination of massive investment, falling costs and accelerated business adoption (from 55% to 78% in a single year) could break the classic technology adoption curve. Traditionally, every new technology goes through a phase of enthusiasm or “ hype ”, a valley of disillusionment or “bubble burst”, and a more gradual and constant growth of maturity. But with AI the pattern could be different: its ability to self-accelerate — thanks to continuous innovation, product integration and competition between states—can generate sustained growth without significant setback. As long as material constraints—energy costs, semiconductor production or computational capacity deficits—do not slow down its development, AI could mark a historical turning point: becoming a new form of digital sovereignty that redefines the rules of technological maturity.
Amidst this technological effervescence, there is another question that perhaps we are not asking enough: what is the cognitive price of this massive delegation of intelligence? Returning to the real example I cited at the beginning, that real-time translation service with more than 20 languages is wonderful, but it also exemplifies a new form of dependency. If we let the machine think, translate and remember for us, what does the act of learning become? Cognitive psychology had already warned about this years ago. Studies such as those by Sparrow or Risko and Gilbert describe what is called “ cognitive “offloading “: externalizing mental processes can be useful, but it can also change the very nature of learning. We remember less about the content and more about where to find it. We gain efficiency, but we may lose depth.
The educational challenge of our time is, therefore, to redefine literacy. It is not about stopping learning languages or formulas, but about learning to delegate with judgment: knowing when it makes sense to use AI, when it doesn’t, and how to verify the result. This new technological literacy will require teachers, parents, and students to learn to live with AI as with any other thinking tool. It is not about banning it, but learning to make it work for the brain, not instead of the brain.
This reflection on cognitive debt finds its parallel in the world of business. Generative AI has captured all the headlines, but 95% of the value that AI generates in organizations today still comes from traditional machine learning models: demand prediction, fraud detection, logistics optimization or personalized recommendations. Generative AI brings additional value, yes, but only when it is intelligently integrated into processes and combined with good data governance. Without this, AI becomes an aesthetic mirage: it produces a lot but transforms little. Productivity studies in real environments confirm this: the gains are substantial —15% more in customer service, up to 55% more in programming—, but only if they are accompanied by organizational changes. The difference between a company that “uses AI” and one that “learns with AI” is the same as between copying and understanding.
Ultimately, perhaps the real challenge is not so much understanding what AI can do, but what we want to do with it. AI will not replace us ; will accelerate us . The difference between noise and progress will be how we delegate intelligently… and what we do with the time we recover. And if the technological adoption curve has accelerated like never before, our duty is to ensure that our capacity to think, learn and decide has also done so. Perhaps the future will depend not so much on the power of algorithms as on the lucidity with which we know how to use them . And perhaps the true intelligence—the one that must be preserved—will be, precisely, that of learning not to stop learning.
