What Happens to Human Development When AI Joins Our Thinking?
Thinking, learning, judging, orchestrating: four capacities for developing the mind in an AI-mediated world
Guy Levi
Act I — A Different Educational Question
As Generative AI becomes a pervasive presence in the cognitive activities through which people think, learn, create, solve problems, and make decisions, a different educational question begins to emerge.
For the past three years, much of the conversation has focused on AI itself: its capabilities, limitations, risks, and what people need to know about AI to use it effectively and responsibly. Those questions remain important. Yet I have gradually come to believe they are no longer the most important ones. The more deeply AI becomes woven into our cognitive lives, the less I find myself asking what people should know about AI, and the more I find myself asking a different question:
What does it mean to continue developing the human mind when AI increasingly participates in cognition itself?
This piece explores that question — not by proposing another framework for AI literacy, but by stepping back and asking how the conditions under which people develop intellectually are changing.
When the Conditions Change
Human development does not occur in isolation. It unfolds within the cognitive conditions in which people think, learn, solve problems, and make sense of the world. When those conditions change, development changes with them.
Generative AI is changing those conditions in a fundamental way. For the first time, people can think with an intelligent system that responds, questions, explains, challenges, and generates alternatives in real time. Cognitive work that once required another person — or long periods of solitary effort — can now unfold in continuous dialogue with AI. This is a condition of hybrid cognition: thinking that is neither fully human nor fully artificial, but distributed across the ongoing exchange between the two.
The question is not whether cognition is becoming hybrid. It already is. The question is what kind of human development becomes possible, or necessary, within it. Because AI participates in cognition, intellectual development increasingly occurs as people think with AI — and the educational challenge becomes helping them continue to develop their own thinking within that process.
Act II — Four Capacities for Intellectual Development
My answer is built around four capacities that I increasingly see as a developmental progression. They are not independent competencies; each prepares the ground for the next.
Thinking
Thinking comes first because intellectual development begins with engagement.
Generative AI can produce explanations, arguments, summaries, and solutions with remarkable fluency. That is one of its greatest strengths. But it also invites a kind of cognitive surrender: the gradual handing over of the work of understanding — accepting the framing, the categories, and the conclusions offered, one reasonable answer at a time. Surrender rarely feels like giving up. It feels like progress.
Thinking means resisting that surrender: remaining actively engaged in constructing understanding — asking questions, exploring alternatives, challenging assumptions, and developing ideas rather than adopting those generated by AI.
Resisting has a cost, and the cost is the point. The effort involved — the slowness, the struggle to articulate, the discomfort of not yet understanding — is what I call cognitive friction. It is tempting to treat friction as an obstacle that AI mercifully removes. Developmentally, it is closer to the opposite: friction is one of the mechanisms through which understanding is built. A system that removes all of it removes something we need.
The goal is not to think without AI. It is to continue thinking while thinking with AI.
Learning
Thinking alone is not enough. Intellectual development depends on what happens after thinking.
Get Guy Levi’s stories in your inbox
Join Medium for free to get updates from this writer.
Learning is the process through which ideas gradually become understanding — richer, more connected, and more durable than the immediate results of a conversation with AI.
Generative AI can accelerate learning in remarkable ways. But speed is not the same as development. Anyone who has ended a brilliant AI conversation only to find, a day later, that nothing of it remains has felt the difference. The challenge is not simply to learn more efficiently; it is to ensure that what emerges from thinking with AI becomes part of one’s own understanding — available tomorrow, without the thread open.
Judging
Understanding alone is not enough. Intellectual development depends on the ability to decide what deserves to be believed, questioned, pursued, or rejected.
In an AI-mediated world, understanding is rarely the end of the process. It is the beginning of a series of choices: Which interpretation makes the most sense? Which evidence is convincing? Which direction should I pursue? Which recommendation should I trust?
Judgment is the capacity that transforms understanding into responsible decisions. And AI raises its stakes in a particular way: it delivers every answer — strong or weak, grounded or invented — with the same confident fluency. When confidence is no longer a signal of quality, the work of evaluation shifts entirely to the human side of the exchange.
Orchestrating
Judgment alone is not enough. Thinking, learning, and judgment increasingly unfold within a world where intelligence comes in three forms: the human intelligence a person brings, the artificial intelligence now woven into their work, and the collective intelligence of the people and networks around them — a triad I describe elsewhere as HI-AI-CI.
Orchestrating is the capacity to bring these three forms of intelligence together intentionally and responsibly in pursuit of meaningful goals. It is not simply about using AI effectively. It is about deciding when to rely on AI, when to challenge it, when to think independently, when to collaborate with others — and how to combine these possibilities in ways that strengthen rather than diminish one’s own development.
In an AI-mediated world, perhaps the highest expression of intellectual development is not knowing more, but learning how to orchestrate intelligence wisely.
A Different Way of Looking at Intellectual Development
This framework is not a checklist of competencies. It describes a progression.
Thinking engages us in the construction of understanding. Learning transforms that engagement into durable understanding. Judging gives that understanding direction.
Orchestrating brings these capacities together into purposeful action in a world where intelligence is human, artificial, and collective at once.
None of this makes AI literacy unimportant. But literacy and development answer different questions: literacy asks what people need to know about AI; these capacities describe how people continue to grow while thinking with it. And if development is the goal, the next educational question is not how to teach four capacities — it is how to design learning environments in which they can develop.
Act III — Designing Learning in an AI-Mediated World
For me, this framework is not primarily a way of describing intellectual development. It is a way of thinking differently about educational design.
Educational design has typically started from one question: How can AI help students complete this task? It should start with another question: How should this learning experience help students continue developing intellectually while working with AI?
That shift affects almost every aspect of practice: how we design learning activities, how we think about assessment, what kinds of questions we ask students, when AI should provide support — and when it should deliberately step back. If cognitive friction is developmentally essential, then designing AI-mediated learning is not only about removing obstacles. It is a matter of Friction Design: deliberately deciding which friction to preserve — and designing it back in where fluency has erased it.
Over the past three years, this question has gradually shaped my own work in designing AI-mediated learning environments. I have come to see educational design less as the organization of content and more as the design of the conditions under which intellectual development can continue.
I do not see the four capacities presented here as a finished model. They are a working framework that emerged from practice and continues to evolve through it. What matters most is not whether they prove definitive, but whether they help us ask better educational questions — because that is where educational innovation begins.
This essay was developed through extended dialogue with ChatGPT and Claude — a process that deliberately kept the friction in: challenge, disagreement, and dozens of iterations rather than fluent first answers. It is, in that sense, an attempt to practice what it argues. The concept carrying that weight — cognitive friction — is explored in depth in Cognitive Friction in AI-Mediated Environments and its practical implications for higher education in Friction Design #1: The Future of Academic Learning.
