What's Left of Learning When the Answer Arrives Instantly?

Picture a student sitting in front of a hard exercise. He tries for five minutes, gets stuck, opens ChatGPT or another chat window, and gets an excellent solution. The exercise is submitted. He can even explain to himself, more or less, why that solution works. But tomorrow, when he has to start from zero without ChatGPT, will he really know?

This isn't a question against AI. I use it - yes, including to write this article - and I'm even quite the alarmist when it comes to the power we're handing it. But this time I want to talk about something else: not how to stop the revolution in education, but how to make sure it doesn't leave us out of shape.

In August 2026, an MIT committee published a report on the use of AI in teaching, learning, and researcher training. It wasn't an experiment measuring students' brain activity (that was a different study), and it wasn't proof that AI "atrophies thinking." It was institutional committee work: five months of meetings, listening groups, reviews of practices at other institutions, and surveys on campus. One of the committee's surveys received 1,632 responses, a 12% response rate. The report's appendix also summarizes a broader survey on quality of life on campus and a separate survey by the student newspaper. Wait, this is important to say: these surveys measure what people feel and do, not what AI actually does to their ability to learn. Those are not the same thing.

So what did the committee see? Students use these tools a lot, and at the same time they express both enthusiasm and concern. The tools can help them learn, practice, and build projects that couldn't have been pulled off before within a semester. On the other hand, the committee describes signs of erosion in mastery of the material and in self-confidence (I'll devote a completely separate article to that), difficulty assessing what a student really knows, and possible harm to social learning. It presents the decline in face-to-face study groups as anecdotal evidence, not as a measurement proving the tool caused it. In the student newspaper's separate survey, 90% of undergraduate respondents reported being somewhat worried about over-reliance on AI. That too is a description of worry among respondents, not necessarily a measurement of damage already done.

The report's most important sentence, in my eyes, is simple: the right answer on the screen can create an illusion of learning. Sometimes, in order to learn, you have to push yourself, get it wrong, fix it, and try again. If every time the difficulty starts we hand the work over to the machine, maybe we've solved the task - but we've skipped the workout. Not every difficulty is sacred; there's also pointless frustration. The question is which effort builds ability, and which effort we can drop without losing it.

Not a blanket ban. A training program.

The idea I call "Humanity-Fit" (a gym for humanity) was born from this: fitness culture - marathons, CrossFit, Pilates, and gyms in general - emerged as a "compensating culture" for our physical decline, because we outsourced our mobility and because so many jobs stopped being physical. So we had to manufacture "friction" artificially, and that's how the gym was born.

I'm not suggesting we throw away the tool (just as I'm not suggesting we throw away motorized transportation). I'm suggesting we decide what must stay active in us even when the tool can do it for us: asking a good question, formulating a position, spotting a mistake, explaining an idea to another person, deciding who bears responsibility for the outcome. Fitness isn't preserved because we have a crane that can lift the weights for us...

But not every effort is the right workout. I call it "targeted friction" - training what life no longer trains on its own. A tour guide who does cardio all day at work will go to the gym to work other muscles. A porter who lifts heavy things all day might actually need cardio. You can take the stairs instead of the elevator, but if it's 25 floors, that could hurt, or eat time that's needed for something else. Same with learning: the thinking work required in order to understand has to stay with us. That doesn't mean calculating a thousand Excel rows by hand.

The committee recommends a similar principle: build up human capability, don't automatically replace it. It also doesn't say there has to be one uniform solution. An exercise meant to build foundations is nothing like a project where you learn to work with an AI tool (which later, when we get to the workplace, we'll be expected to use!). So the question isn't "allowed or forbidden?" but "what are we trying to learn here, and which use of the tool will serve that?" It suggests teachers set a clear, reasoned policy for each course or assignment, bring in more projects, discussion, and face-to-face work, and check understanding through live conversation (not in front of an AI agent!), a portfolio, and an explanation of the work process. This isn't just a war on copying. It's a way to see whether the person submitting the work grew while doing it.

I'm talking about this from the side that built the tools, not only the side that writes about them. I've worked in learning technologies (EdTech) since 2018, and the pattern that repeats there is constant: the tools change, the condition for success doesn't.

At Time To Know (today: enabley), before the AI era even began, we helped teachers, content experts, and content creators produce content faster and more easily, so that learning would be effective and measurable. At Academe+, we built a film-based learning platform with broadcast rights to Hollywood movies: a teacher could put together a fascinating lesson from clips students already knew and loved. At Melingo AI, a subsidiary of the global Britannica (yes, the encyclopedia one), I developed two courses for teachers around the world: how AI helps them produce content and assessment, how it saves them time that comes back to them so they can strengthen the personal connection with students, how to understand what AI even is, and how to adapt the material not only to the level of the class but to the level of the individual student (personalization).

In all three of those stops I saw the same thing: technology in education works when it frees the teacher up for the human connection, and fails when it tries to replace it. Over those years the tools changed from end to end. The condition didn't budge. That is exactly the committee's conclusion, and that is exactly the point Humanity-Fit comes to strengthen.

I don't yet know what the long-term effects of these tools will be on our learning. The committee doesn't claim to have a final answer either; it calls for measurement, experimentation, and ongoing correction. But we don't have to wait for absolute certainty to ask what we're training and what we're neglecting.

I'm not against using AI for learning. Not from the side of those who create the learning materials, not from the teachers' side, and not from the students' side. But there's a new challenge here that needs to be understood. The encyclopedia didn't require every student to go out and conduct original research; it let them understand processes and meanings. The calculator didn't require us to do every simple addition in our heads; it required us to show the way. AI, too, should raise learning a floor: less memorization and retrieval of details, more deep understanding, critical thinking, the ability to defend a position and find where the tool is wrong or where we think differently. The shape of learning has to fundamentally change because of this tool. And until the education system and the Council for Higher Education change, here are things you - as parents, as students, and as teachers - can do as early as tomorrow morning.

One important note: These are rules I'd try myself, not instructions research has already proven for every age and every subject.

• For parents: Before you let your child ask ChatGPT for a solution, have them explain out loud what they tried and where they got stuck. Then you can ask the tool for a hint or a guiding question, not a ready-made answer. At the end, close the screen and ask whether they can solve a similar problem on their own. If not - we got an answer, but we didn't get learning.

• For students: You can let the tool prepare a draft, summarize material, or write code. But before you adopt the answer, try to formulate for yourself what the question is, what might be wrong with the output, and how you would check it. Every once in a while, do one round without help - specifically in the skill you don't want to lose. Not out of nostalgia. So that on the day the machine offers you something convincing and wrong, you'll still be able to notice.

• For teachers and lecturers: State explicitly what the assignment's goal is. If the goal is independent formulation or proof, keep one stage that's done without tools. If the goal is critiquing a solution, let the tool propose one and ask the students to find a weakness in it and defend another position. Also ask for an oral explanation or a short record of the decisions made along the way. Don't build everything on AI text detectors: the committee itself warns against relying on them.

And if this topic stirs something in you that's hard to put into words - a feeling that this revolution is exciting and frightening in the same breath - you're not alone. I feel it too. Let's build Humanity-Fit together, and preserve what makes us human. Write to me.