title: "How to Learn in the AI Era Without Being Left Behind—Is Knowledge Really No Longer Important?" slug: "fanchunping-how-to-learn-in-ai-era.en" date: "2026-06-26" category: "education" author: "Fan Chunping (Zhigeng)" excerpt: "The AI era doesn't mean knowledge is no longer important—it means the importance of knowledge is heightened, and its structural nature is more demanding. The knowledge ocean of large models has unfolded before you, but what you can fish out depends entirely on your cognitive schema."

How to Learn in the AI Era Without Being Left Behind—Is Knowledge Really No Longer Important?

Author: Fan Chunping (Zhigeng)

The future is already here, just unevenly distributed. AGI is already here, just with jagged edges around its capabilities.

In the face of the AI era, how ordinary carbon-based humans should learn so as not to be left behind—this is no longer merely a question of development; it has become a question of survival.

I. A Rumor: Knowledge Is No Longer Important; the Ability to Ask Questions Is What Matters

The most widespread claim about learning or education in the AI era is probably this: with AI, knowledge is no longer important; what's important is the ability to ask good questions.

Is this claim correct? It is—but only halfway.

Its first half is wrong; its second half is right—very right. And precisely because the second half sounds so thoroughly correct, the error in the first half gets treated as truth as well.

There is a skipped-over question here: how does one acquire the ability to ask good questions?

A person who has never read physics or cosmology will not ask, "Why do dark matter and dark energy account for over 95% of the matter in the universe?"—because they don't even know dark matter exists.

A person who has never read history will not ask, "What was the relationship between the Ming Dynasty's maritime prohibition policy and the first wave of globalization?"—because they don't even know the Ming Dynasty had a maritime prohibition.

A person who has never encountered molecular biology will not ask, "How great is the variation in CRISPR off-target effects across different cell types?"—because they don't even know what CRISPR is.

Being able to ask good questions has a precondition: possessing a sufficiently broad, rich, and deep knowledge structure, along with a logically clear, firmly grounded cognitive model. Only then can one realize "there's a question here" and describe that question with precision—and, upon receiving an answer, evaluate its quality.

Stanford education professor Lee Shulman proposed a classic teaching framework: domain expertise is not simply an accumulation of facts—it encompasses content knowledge, pedagogical knowledge, curricular knowledge, and contextual knowledge, among other dimensions.[1]

The most critical point: you can only recognize problems in domains you already know. Problems beyond your knowledge boundary—you don't even know they are problems.

In other words: knowing your own ignorance requires a great deal of knowledge. Without sufficient knowledge, you can't discover problems, nor can you even locate them.

As the saying goes: "You can never earn money beyond your cognition." Likewise: "You can never accomplish things beyond your cognition."

II. A Type of Legend: Geniuses Crowned Through Questioning—Knowledge No Longer Matters

A certain middle-school prodigy used AI questioning to complete, in two hours, a course that normally takes a full semester. A certain MIT overachiever used 48 hours—pouring in materials, asking questions, solving problems, and following up with deeper questions—to complete a course that normally requires a semester.

I believe these stories are true. What no one emphasizes: that middle-school prodigy had already won two national hackathon championships before completing a course in two hours. And the "MIT overachiever" is an "MIT overachiever" precisely because his knowledge base and capability structure are already beyond what ordinary learners can match—his interrogation method reflects his profound knowledge foundation and powerful methodology.

Using such legends as evidence that knowledge is useless will, in the end, mislead both oneself and others.

III. "There Is a Chicken"

In the 1930s, Soviet film director Vsevolod Pudovkin conducted an experiment that later became widely cited. He shot a documentary about urban life and screened it for a group of African tribal people who had never seen a film. Unexpectedly, after the screening, the tribespeople didn't discuss the skyscrapers, cars, or streets on the screen—they animatedly discussed a chicken.[2]

The director found this very strange, because he didn't think his film contained a chicken. But the tribespeople were so insistent that he checked the footage frame by frame and finally found it—the chicken had appeared in a corner of the frame for less than a second, an accidental intrusion during filming that the entire production team had missed.

But the tribespeople saw only the chicken.

Why? Because their cognitive system contained no schema for "skyscraper," "automobile," "asphalt road," or "roaring machinery"—these modern artifacts were merely unremarkable patches of color and noise to their eyes. The only strange thing they could "recognize" was that chicken—that living, breathing chicken.

German film theorist Siegfried Kracauer, in his 1960 classic Theory of Film: The Redemption of Physical Reality, cited this story to illustrate a profound cognitive principle: people can only see what they already understand.[3]

Not "see but not understand"—rather: things you don't understand, your brain simply won't let you see at all.

This principle is precisely the key to understanding constructivist learning.

IV. The Nature of Constructivist Learning

In every course I teach, I introduce learning methods in the opening session, with constructivist learning as the core.

A central thesis I repeatedly emphasize: learning is not pouring water into an empty bottle, not stacking books into a bookcase—it is using your existing cognitive structures to discern new knowledge, and using new knowledge to construct new cognitive structures—a process of assimilation, accommodation, and equilibration. Knowledge that cannot be assimilated by your cognitive schema finds no foothold in your mind and will eventually drift away on the wind. Only knowledge that can be positioned within your schema becomes knowledge truly learned.

Swiss psychologist Jean Piaget's "schema theory,"[4] proposed in the first half of the 20th century, holds that human cognition is not a process of passively receiving external information, but an active process of construction—your brain first possesses a set of "schemas" (think of them as cognitive scaffolding), and new information is integrated through assimilation and accommodation with these schemas.

Assimilation: new information aligns with your existing cognitive framework → incorporated into existing schemas.

Accommodation: new information conflicts with your existing framework → you need to adjust or even rebuild your schemas.

This forms a progressive process: use existing schemas to absorb new knowledge, enrich and develop the schemas, then absorb more knowledge...

This also became the theoretical foundation of constructivist learning.

An example:

Someone with a cognitive schema for "AI," upon seeing "Sora generated a fully physically realistic video," enters a thinking pathway of "What architecture did this model use? How was the training data obtained? What is the underlying logic of image generation?"

Someone with zero understanding of AI, seeing the same video, reacts: "How amazing," or perhaps "Nothing special"—and then the moment is over. Not because they don't want to think deeper, but because they have no cognitive tools to "process" this information—they can't form questions, and they can't even generate the motivation to ask.

This is why Piaget said: cognitive development is not the accumulation of knowledge—it is the evolution of cognitive structure itself.

Returning to the AI-era context: large models are a "massive repository" of human knowledge—data pools, data lakes, data clouds. But what you can fish out of this repository depends on what tools you use. The most essential tool is your knowledge structure and thinking framework—your cognitive schema. Without a good schema, you can only ask superficial, generic questions and receive generic answers in return.

V. Popper's Three Worlds and the Newly Emerged and Expanding New World

According to philosopher of science Karl Popper: learning is the process of using our own minds (World 2) to reflect upon and know the objective external world (World 1), forming objective knowledge (World 3). Now, beyond Popper's three worlds, a virtual data world has emerged (World 4—formed by devouring World 3 and merging with big data).

At this point, human learning, beyond the original observation, experience, and study of the objective external World 1, now draws primarily upon World 4 as its information source.

VI. How to Turn the Super-Database That Is the Large Model into Your Own Thought Partner?

Many people understand AI-era learning as: the large model is an all-encompassing ocean of knowledge, waves of knowledge churning within; whatever you lack, it can give you.

Correct—it can give it to you. But you need to tell it what you lack. If you don't know what you lack, it is stagnant water, without the slightest ripple.

Without the ability to ask questions, a large model is, to you, less useful than a small reference library. Because walking into a reference library, you can at least pull a book off the shelf and read it. Facing a large model with an empty mind, you won't know where to begin—returning empty-handed from a mountain of treasure.

The Essence of Large Models

However much they may appear human—for now, they remain "reasoning engines with vast stores of knowledge"—"brains in vats." They can:

Retrieve relevant information from their training data in milliseconds
Perform pattern matching and reasoning based on that information
Generate outputs that resemble human thought
Complete tasks you can clearly specify and that are achievable within the data world

But they cannot:

Articulate what you want to say but haven't yet organized into words—you need to express the question precisely yourself
Build your own knowledge structure for you—they give you answers, but answers don't automatically become your cognitive schema
"Know what you don't know" on your behalf—if you yourself are unclear where your blind spots are, the questions you ask will also be blind

AI Needs Question-Driven Engagement

A good question framework is like a map; the large model is a treasure hunter on this map. Without a map, it can only wander aimlessly. Give it a rough map, and it can only find rough directions. Give it a carefully drawn map, marked with landmarks and paths, and it can lead you to where the treasure lies.

Good questions also have structure: what is the question, and for what purpose? What is the background? What are the interconnections? What needs to be avoided? The more clearly you define these, the more precise and brilliant the AI's answer.

Compare two approaches:

Poor question: "Tell me about machine learning."

The AI gives a generic definition and historical overview, akin to a Baidu Encyclopedia entry.

Good question: "I'm a middle-school physics teacher and I want to use AI to demonstrate an experimental simulation of Newton's Second Law in class. I need an interactive demonstration tool that runs in a browser, with sliders allowing students to manually change mass and force values and see real-time acceleration changes. Can you help me design this teaching plan? It should include operational backend code and teaching guidance suggestions."

The AI immediately understands your identity, context, requirements, and desired format. The output is implementable, not a science popularization article.

This is the power of question framing: the more precisely you define, the more complete and useful the AI's output.

VII. The Three-Layer Learning Method for the AI Era

Based on constructivist learning and large model characteristics, whether cultivating one's children or students, or pursuing adult self-development, roughly these three layers may serve as a framework:

Layer One: Strengthen Foundations—Reinforce and Optimize Your Cognitive Schema

A person without knowledge cannot ask good questions. The AI era doesn't mean knowledge is no longer important—it means knowledge's importance is heightened, and its structural nature is more demanding, because knowledge itself is the background and raw material for questioning.

How to do it: read classic, primary-source, systematic books and literature; take substantive, deep courses; pursue first principles; plan your own learning—this is also where the necessary value of universities and teachers lies.

Don't just scroll short videos and fragmented articles. Those things, when you already possess a sufficiently solid, broad cognitive schema, can illuminate certain hidden currents. But when your cognitive schema isn't yet strong enough, they will lead you astray—toward superficiality.

I can offer myself as a case study: I am a complete technical novice. To keep pace with the times, and especially this year in order to raise a lobster and make it my learning and thinking partner, I have been frantically cramming AI knowledge—reading multiple related books and widely following AI news and technical developments. Only then could I understand the broad arc of AI's evolution, grasp the logic behind many AI phenomena, and find effective ways to connect them to my own knowledge structure.

Layer Two: Master Tools—Cultivate an AI Partner Suited to You

Once you have a good knowledge structure, learn to use AI tools to extend your cognitive boundaries, not to replace your cognitive process.

In this process, you and your Agent (in my case, a little rabbit-lobster—Ximiao) will progress together, and your mutual coordination will become increasingly seamless.

Layer Three: Use Good Questions to Drive AI

This layer is truly "the ability to ask questions." A person who can ask good questions has usually already delved deeply into their domain for a long time and accumulated profound depth.

How to do it:

Continuously accumulate domain knowledge, until you can "see" those sparks and highlights invisible to others
Transform your vague intuitions into actionable questions
Use question frameworks to guide your Agent in deep research

The relationship among these three layers is progressive: without Layer One, Layers Two and Three are castles in the air.

Here, a real-world example to share:

The "Future Spark Research Institute," an educational organization dedicated to cultivating future-ready youth for the AI era, has proposed an educational philosophy: educating children must begin with educating parents. If parents' mindsets don't change, their children's education is unlikely to succeed, statistically speaking.

So they have created a dedicated "Computing Parents" course—training parents before training children.

This is, in essence, a spontaneous application of Piaget's schema theory: parents are a child's first teachers. When parents' concepts and behaviors change, it most directly manifests in the influence and guidance they provide to their children. And the foundational cognitive schema from one's family of origin is a child's "factory configuration"—the starting point for all further progress.

For this reason, I am willing to emphasize once more: learning in the AI era not only requires the learner to possess knowledge, but demands that knowledge be structured. Only then can one ascend to the level where one can drive large models to empower you—asking good questions.

The underlying logic of AI-era learning tells us: knowledge is not only important—it has never been more important.

On the foundation of a solid knowledge structure and cognitive framework, strengthen the ability to ask good questions; drive AI with questions; progress in concert with AI—only then will you avoid being left behind by the era.

References:

[1] Shulman, L. S. (1986). Those who understand: Knowledge growth in teaching. Educational Researcher, 15(2), 4–14.

[2] Pudovkin, V. (1929). Film Technique and Film Acting. Vision Press.

[3] Kracauer, S. (1960). Theory of Film: The Redemption of Physical Reality. Oxford University Press.

[4] Piaget, J. (1952). The Origins of Intelligence in Children. International Universities Press.

Fan Chunping (Zhigeng) · June 2026