By Jane Sonnenschein September 16, 2026

Some time ago, a friend of Chris’s was working on a piece of software that could run locally on a computer. More precisely, he was developing a local workflow for managing a company’s internal documents.

The idea was actually quite easy to understand. Many companies have large amounts of internal information, including business data, contracts, customer records, and other documents that shouldn’t simply be uploaded to external platforms.

If those files could be kept locally and organized through a piece of software, a company owner who wanted to know something about the business would no longer have to search through dozens of documents. They could simply ask a question, and the system would look for the answer in the company’s own files.

It sounded convenient, and it also seemed to address a growing concern among businesses: data security.

The interesting part was that the person developing this software wasn’t a programmer. He knew almost nothing about coding.

In the past, that would probably have been the end of the story. Even with an excellent idea, someone who couldn’t program would have found it very difficult to turn that idea into a working product on their own.

But now there is AI.

He could describe what he wanted to build, ask AI to generate the code, and gradually put the whole program together.

And something quite interesting happened. A field that once required considerable professional knowledge to enter had become accessible to someone without any formal training. He could actually build a product that worked.

After listening to his friend explain the project, Chris asked several questions about security.

How exactly did the software handle company documents? Did the data really stay on the local computer? Which third-party components did the program use, and were those components secure? Could the AI-generated code contain vulnerabilities? And if companies eventually started using the software, what would happen when something went wrong?

His friend couldn’t really answer those questions.

I don’t think he was necessarily trying to avoid them. He might simply not have known how to answer.

After all, when someone doesn’t understand programming, AI can help them write code. But it cannot also install into their mind all the professional judgment that an experienced programmer has developed over many years.

That conversation left me thinking about something:

AI can lower the barrier to doing something, but it cannot take responsibility for you when you do it wrong.

A program that runs is not necessarily a program that is secure. A product that looks useful is not necessarily ready to be used by a company.

This becomes especially important when the software involves internal business data, access permissions, privacy, security vulnerabilities, copyright, or software licensing.

The hardest part is often no longer getting a particular feature to work. It is knowing whether something is wrong with what you’ve built, and whether anyone will have the ability to identify the problem when something goes wrong.

In the past, professional barriers also served as a kind of filter.

Someone who couldn’t write code usually couldn’t build a program at all, so many potential risks never reached the stage where a product could actually be offered to others.

Now AI has opened that first door. Even without a professional background, someone can quickly gain part of the practical ability that once took years of training to develop.

That is, of course, an extraordinary change.

But it also creates another problem.

When making something becomes increasingly easy, understanding what you’ve actually made becomes even more important.

A few days ago, I came across another online conversation, this time between a content creator and a man in his forties.

The man hadn’t received much formal education when he was young, and he had always felt that his limited qualifications had held him back.

He had spent many years living and working in Beijing, away from his hometown and separated from his wife and child for long periods. He was one of the many people sometimes described in China as běipiāo¹.

Eventually, because his parents were in poor health and his child was about to enter senior high school, he decided to end those years in Beijing and return to his hometown in Inner Mongolia.

His work was seasonal, and there wasn’t much for him to do during the winter.

That made him anxious. Should he find another job?

Judging by his savings, his family was actually unlikely to face serious financial difficulties for several years. But even so, after just a few months without work, he began to feel insecure, almost as if life would slip out of his control unless he continued earning money.

Another major source of anxiety was his child.

The child was in the final year of junior middle school, equivalent to ninth grade, and was already under considerable academic pressure.

The father hadn’t gone far in school when he was young. Although he later earned a junior college diploma through adult education, he still felt that many career opportunities had remained closed to him because of his educational background.

Naturally, he didn’t want his child to follow the same path.

The content creator suggested that education in the future might not always work the way it does today.

Why did our modern school system take shape during the industrial age?

To a large extent, it was because a modern society operating on a large scale needed a relatively standardized way to educate large numbers of people.

Children entered school according to their age, studied broadly similar subjects, took standardized or comparable examinations, and then moved through different stages of selection based on their qualifications.

The advantages of such a system are obvious. It can operate on a large scale, and it provides a relatively consistent way to judge what kind of training someone has received, even when it is impossible to know every individual personally.

But if AI continues to develop, could this way of selecting people change?

The content creator believed that individual differences might become increasingly important in the future.

What a child genuinely enjoys, what they are good at, and which subjects they are willing to explore deeply over many years might come to matter more than having everyone complete the same training according to the same pattern.

I don’t know whether academic qualifications will lose their current importance as quickly as he suggested over the next ten years. Nor do I know what schools will actually look like in the future.

But the conversation reminded me of something more fundamental.

Knowledge itself is becoming easier and easier to access.

A few hundred years ago, if an ordinary person wanted to learn a craft, they couldn’t simply open a webpage, find a book, or ask AI a question.

In many trades, they first had to become an apprentice under a master, a traditional practice known in Chinese as bàishī².

They would enter the master’s workshop, but that didn’t mean they would immediately begin learning the actual skills.

At first, they might be expected to do odd jobs, run errands, serve tea, and help with everyday tasks.

Only after spending a long time together and proving themselves trustworthy would the master gradually begin passing on their knowledge and experience.

Even then, traditional craftspeople often kept a few tricks of the trade to themselves.

Their knowledge was what allowed them to make a living.

Knowing a method that other people didn’t know gave someone a competitive advantage. There was little reason to teach another person absolutely everything without holding anything back.

In those days, gaining access to knowledge was difficult in itself.

Once someone entered a particular trade, they might remain in it for the rest of their life. Changing fields meant finding a new way in, becoming an apprentice again, and spending years accumulating experience from the beginning.

Much of that knowledge wasn’t written down in books, waiting for someone to look it up.

It existed in certain people’s minds. To gain access to it, you first had to enter their network of relationships.

The internet has already changed this enormously.

And AI has taken things a step further.

In the past, you at least needed to know which keywords to search for, which books might be useful, or which websites to visit.

Now, even if you have only a very vague idea, you can simply tell AI:

“I want to make something like this. Tell me how I should begin.”

This means that an ordinary person can now enter many completely unfamiliar fields at a remarkably low cost.

Don’t know how to code? AI can help you write the code.

Don’t know how to build a website? AI can help you create one.

Can’t translate something? AI can translate it for you.

Don’t know how to design something? AI can suggest an initial approach.

Even if someone knows almost nothing about a particular field, through an ongoing conversation with AI, they can quickly develop a basic understanding of its main ideas and problems. In the past, reaching that point might have required reading a great deal of material.

What has changed here may be more than productivity.

It is the scarcity of knowledge itself.

And if we look a little further into the future, what might happen if brain-computer interfaces eventually become advanced enough for people to store large amounts of information directly in their brains, almost as easily as saving files on a computer?

Imagine that learning a language, which currently takes years, could one day involve directly transferring hundreds of thousands of words, grammar rules, and examples of actual language use into the brain.

Or imagine that instead of spending years studying a subject, a person could acquire a large amount of its foundational information in a very short time.

If that ever became possible, the kind of learning that currently requires us to spend enormous amounts of time memorizing information might change fundamentally.

But that is also where another problem would appear.

Suppose I suddenly had ten thousand books stored in my brain.

Would that mean I possessed all the abilities represented by those ten thousand books?

Not necessarily.

If someone asked me, “Do you know this book?”

Of course I would.

My brain might even be able to retrieve it instantly.

But when I faced a real problem, would I know which of those books to use?

And even if I knew which book to look at, would I know which method in it was suitable for the situation in front of me?

If three theories contradicted one another, which one should I believe?

If a particular method had worked in the past, but circumstances had since changed, would I recognize that it was no longer suitable?

And if I faced a completely new problem, something nobody had encountered before, could I bring together knowledge from several very different fields and develop a new solution?

At that point, the amount of information I possessed would no longer be the deciding factor.

What would matter is whether I had developed my own way of organizing and making sense of that knowledge.

This is why I increasingly feel that the most important ability in the age of AI may not be how much information we have in our heads.

It may be whether we can judge when that information is useful, and whether the answers we receive can actually be trusted.

Writing is very similar.

Today, anyone can give AI a title and ask it to produce an article of about two thousand Chinese characters.

If they wanted to, generating a hundred articles in a single day wouldn’t be particularly difficult.

But then come the questions.

Does that article actually say what you wanted to say?

Do you agree with the opinions it expresses?

Are the facts it refers to true?

Has it quietly added a position that you never expressed in the first place?

If someone has no ideas or judgment of their own, and simply thinks, “Well, that sounds pretty convincing,” before repeatedly copying and publishing AI-generated content, they are not necessarily developing stronger writing skills.

They are simply accumulating more text produced by a machine.

As these tools become more powerful, I find it even more important to understand who I am, what I want to express, and what I believe is right.

Without that understanding, the more powerful our tools become, the easier it may be for us to hand over our own thinking.

Chris’s friend and his software project are really another example of the same problem.

His idea was a good one, and AI gave him the ability to turn it into something real for the first time. That is certainly something worth encouraging.

But if companies were eventually going to place large amounts of internal information into his software, he would also need to face a very different set of questions.

Is the AI-generated code secure?

Which third-party components does it use?

Does the data really remain on the local computer?

Could someone bypass its access controls?

Might a software update introduce new vulnerabilities?

And if something goes wrong, who will be able to understand the code well enough to find the cause?

These questions don’t simply disappear because the program happens to be running successfully.

If he doesn’t have the knowledge to make those judgments himself, then he needs to find someone who genuinely understands these issues and can review the software.

Because AI may be able to write tens of thousands of lines of code for you, but when a serious problem occurs, it cannot step forward and say:

“I’ll take responsibility.”

The person who approves the product, launches it, and allows customer data to be placed into it is still a human being.

So I’ve increasingly come to feel that the biggest change in the future may not be whether people still need to learn.

It may be what we actually need to learn.

If knowledge becomes easier and easier to obtain, perhaps education really won’t need to spend so much time on mechanical memorization and repeatedly passing information from one person to another.

But that doesn’t mean people can afford to have less and less understanding of their own.

Quite the opposite.

We may need to learn even more about logic, judgment, verification, asking questions, creating things, solving problems, and understanding ourselves.

A person needs to know what they are good at.

They also need to recognize what they don’t understand.

They need to know when to trust a tool and when to question it.

They need to recognize that an answer can look beautifully written and still be wrong.

And when they encounter an unfamiliar problem, they need to know how to find the most relevant knowledge among everything they have learned, then bring different pieces of it together in a useful way.

If every child really does have a personal AI tutor in the future, the most important question won’t simply be whether AI knows how to teach.

The child will also need to gradually understand where they learn quickly and where they keep making mistakes.

They will need to learn how to ask questions, how to check answers, and how to tell their AI tutor what they genuinely understand and what they are struggling with.

Only then can they use the tool to help create a learning path that better suits them.

AI may come to understand a person better and better.

But that person also needs to understand themselves better and better.

In the past, when knowledge was scarce, one person’s greatest advantage might have been:

“I know something you don’t.”

As knowledge becomes cheaper and more accessible, that advantage may gradually diminish.

Perhaps what will increasingly distinguish one person from another is a different ability.

Faced with the same information, some people merely possess it, while others know how to use it.

If the day ever comes when we really can place ten thousand books directly into our brains, I don’t think humanity’s most important question will be:

“How many books do I have in my head?”

It will still be this:

When the real world presents me with a problem I’ve never encountered before, will I know which knowledge to draw from those ten thousand books?

And will I know when even the answers in those books should no longer be followed without question?

---

Notes

1. Běipiāo (北漂)

Běipiāo is a Chinese term commonly used to describe people who move to Beijing from other parts of China in search of work, education, or better opportunities, often without having permanent roots in the city.

The word combines Běi (北), referring to Beijing, and piāo (漂), meaning to drift or float.

The expression carries more than the simple meaning of migration. It often suggests the uncertainty of building a life in a large city, the pressure of earning a living, separation from family, and the feeling of not quite belonging. Some people eventually settle in Beijing, while others return to their hometowns after many years.

In this essay, the term describes the man’s long experience of working in Beijing while living apart from his family.

2. Bàishī (拜师)

Bàishī refers to the traditional practice of formally becoming a student or apprentice under a master, especially in fields such as craftsmanship, martial arts, traditional medicine, and other skills passed down through personal instruction.

The expression literally means to acknowledge or pay respect to a master.

Historically, learning a trade often involved more than receiving technical instruction. An apprentice might spend years helping with everyday work, observing the master, and gradually earning trust before being taught more specialized knowledge.

The relationship could involve expectations of loyalty, respect, and personal obligation. Masters sometimes chose not to reveal all their techniques, partly because those skills were also the basis of their livelihood.

In this essay, bàishī illustrates a time when access to specialized knowledge often depended on personal relationships and long periods of apprenticeship, rather than simply being able to find information.

This essay is also available in other languages:

Chinese version: 当知识越来越容易获得，真正稀缺的也许是判断力

German version: Wenn Wissen immer leichter zugänglich wird, ist vielleicht Urteilsvermögen das, was wirklich knapp ist