As generative artificial intelligence produces increasingly convincing texts, images, and reasoning, a crucial educational question emerges: what does understanding still mean? In the age of information overload, developing a “well-formed mind” no longer consists of accumulating knowledge, but of learning to judge its depth, validity, and relevance to reality.
Students today can produce work that is flawless in form: structured, well-argued, sometimes even brilliant. Yet, when questioned, a sense of unease surfaces. They struggle to explain what they have truly understood, to justify their choices, to connect their work to a lived experience or a concrete situation. Generative artificial intelligence (AI) is not always the direct cause of this situation, but it is a powerful catalyst. Because while producing information has never been easier, understanding what one is doing has never been so demanding.
To know, to understand, to comprehend: a distinction that has become central
In the age of AI, education can no longer be conceived in terms of knowledge accumulation. It requires clarifying what we mean by knowing, understanding, and comprehending, and examining how these dimensions are articulated in the learning processes.
Two major epistemological traditions help to illuminate this distinction. The scientist Michael Polanyi demonstrated that all human knowledge contains an irreducibly tacit component : it is rooted in the subject’s experience, action, and engagement. “We know more than we can say,” he wrote, emphasizing that understanding often precedes its explicit formulation. This knowledge in action, often implicit, is constructed through doing, trial and error, and confrontation with reality.
Conversely, the philosopher Gaston Bachelard established that scientific knowledge does not proceed from a simple extension of experience. It requires a break with initial assumptions and with opinion, at the cost of a process of rational, critical, and abstract construction. “Science does not proceed from opinion,” he reminded us, insisting on the need to train the mind to pose problems rather than to accumulate answers.
Developing a “well-formed mind” is therefore neither about accumulating abstract knowledge nor being satisfied with raw experience. It is about learning to hold together these two dimensions: lived experience and conceptual construction, action and reflexivity.
What AI can do – and what it cannot do
Artificial intelligence systems excel precisely where knowledge can be formalized: calculation, synthesis, reproduction, and formatting. They take on an increasing share of explicit, stabilized, and computable knowledge. But they operate within a specific framework: that of statistical correlation and the production of plausible statements.
AI does not know the world; it does not understand it. It has no experience, no embodied relationship with reality, and no access to the conditions of pluralism of the phenomena it describes. The information it generates is by nature probabilistic—it relies on probability calculations derived from statistical correlations rather than on an understanding of causes; contingent—it depends on the data, the contexts of enunciation, and the technical parameters; and revisable—in the sense that it can be corrected, contradicted, or reformulated at any time without this implying an internal progression of understanding.
This distinction is now at the heart of contemporary work on the educational uses of AI , which shows that the automation of certain cognitive tasks can, if poorly managed, impoverish the exercise of critical judgment .
The more convincing AI productions become, the greater the risk of confusing formal coherence with real understanding, namely appearing truthful instead of a cautious statement that opens up dialogue.
Measuring complexity: a skill that can be learned
Faced with this situation, a major educational challenge emerges: the ability to grasp the complexity of things. To distinguish what is superficial information from what requires structured understanding. To appreciate the levels of depth of a problem, a system, or a situation.
But this capacity cannot be decreed. It is built gradually through experience of reality. It presupposes an active process of comparing what is theoretically anticipated with what is revealed by the test of concrete realization. It is in the gap, always instructive, between the model and experience that the criteria for judgment are refined and that a truly situated intelligence develops, in the sense that it articulates formalized knowledge and lived experience .
Knowledge only becomes effective when it is tested, brought into contact with reality, and readjusted in light of its resistances and surprises. Conversely, raw experience, if not processed within a reflective and conceptual framework, remains mute and difficult to transmit. Education must therefore organize the conditions for this demanding interplay between theory and practice, abstraction and embodiment.
A pedagogical as well as a technological revolution
Artificial intelligence is not just transforming our tools. It intervenes at the very heart of higher cognitive functions: externalized memory, instant access to information, and the generation of apparent reasoning. Where previous technologies amplified existing human capabilities, AI is now reconfiguring the balance.
The educational focus has shifted accordingly. It is no longer primarily about learning to produce or disseminate information, but about learning to assess its depth, coherence, validity, and real-world impact. This shift aligns with sociologist Edgar Morin ‘s analyses of complex thought, which emphasize the need to cultivate minds capable of connecting, contextualizing, and confronting uncertainty, rather than reducing reality to simplistic answers.
Recent work in cognitive science and educational science shows that the substitute use of AI can lead to a form of excessive cognitive delegation, reducing intellectual engagement and long-term memorization, whereas reflective and critical use can instead strengthen learning.
To train engineers – and citizens – capable of judging
Developing a well-rounded mind in the age of AI thus requires distinguishing between cognitive delegation and intellectual abdication. It is about training individuals capable of using powerful systems without being subservient to them, capable of maintaining a sense of meaning where the machine produces only form.
At IONIS, the development of the IONIS Institute of Technology (I2T) on our campuses stems from a strong conviction: while our engineering students must master artificial intelligence technologies, they must equally learn to test their limits through real-world application. The laboratory, the workshop, and experimentation then become central spaces for developing critical thinking skills.
Training good engineers—and more broadly, informed citizens—means cultivating a critical, measured, and evolving mindset, nurtured by concrete experience, by doing and undoing. In the age of AI, the essential question is therefore not only what we expect from machines but also what we expect from humans: their capacity to understand, create, and make discerning decisions in uncertain and technologically augmented environments.
Author Bio: Clément Duhart is Director of Strategy and Innovation at IONIS Education Group
