“To use well a technology that thinks fast, we first need to have learned how to think slowly.” Artificial intelligence has made speed one of its greatest virtues. But learning has never been a fast process. And that tension between the logic of algorithms and the logic of learning is the starting point for our conversation with Joaquín Rodríguez, researcher at the Institute for the Future of Education at Tecnológico de Monterrey.
While the public debate revolves around which chatbot to use, which assignments to allow, how to assess students, or which digital skills should be taught, Rodríguez introduces words that rarely appear in conversations about artificial intelligence: effort, memory, attention, friction. Because learning has never been about finding the shortest path.
All these ideas are grounded in the article, written together with Juan Freire, Slow First, Smart Later: A Framework for Cognitive AI Integration in Education, which proposes a model for integrating artificial intelligence based on the progressive development of cognitive capacities. In this conversation, the author reflects on how we learn and on the role that schools and teachers should play in the age of AI.
Necessary Difficulties
In contrast to a technological culture that associates progress with speed, automation, and the absence of effort, Rodríguez defends the importance of friction. Learning requires retrieving information, making mistakes, comparing ideas, sustaining attention, and reorganizing knowledge. These difficulties do not slow learning down; they are part of it.
This is why his proposal could be summed up in a very simple formula: slow first, smart later.
His approach does not consist of indefinitely delaying the arrival of artificial intelligence in the classroom, nor, of course, of banning it. It consists of introducing it at the right moment. First, as an AI-free environment in which to build foundational cognitive capacities. Later, as an assistant capable of questioning arguments, pointing out inconsistencies, or suggesting new questions. Only then does it become fully integrated into the learning process.
In this model, artificial intelligence ceases to be a machine that provides answers and becomes instead a tool that helps us think better. A kind of Socrates 2.0, capable of guiding us in the construction of knowledge.
The conversation explores some of the most thought-provoking ideas in the current debate on artificial intelligence and learning: the paradox of tools that can weaken the very capacities we need in order to use them well; the “cognitive laziness” produced by shortcuts; the defence of effort and of “desirable difficulties”; the differences between generations that have learned within very different technological ecosystems; and the possibility of designing an AI that does not provide answers, but instead helps us formulate better questions.
We invite you to watch the full interview.
