1 Human understanding of the world is, in essence, a process of establishing connections

The process of human understanding of the world is, in essence, a process of continuously discovering differences, establishing classifications, and seeking connections between things.

For early humans, the external world first presented itself as a vast array of perceptual objects and phenomena yet to be understood: the sun, moon, and stars in the sky, the plants and animals on the earth, the changing seasons, weather variations, and humans themselves, all mingled together in a world that was not yet understood.

And the first step in human understanding of the world was to attempt to divide this complex whole: What are animals? What are plants? What are things of the heavens? What are things of the earth? What are phenomena that can be observed and measured? These seemingly simple questions actually represent a major shift in the human cognitive process:Humans began to attempt to distinguish things that were originally mingled together.

However, distinguishing things is not the ultimate goal of humanity. More precisely, it is a cognitive approach adopted by humans to understand the world. Because the real world contains a vast number of things with complex relationships, and the human brain cannot process all unorganized information simultaneously. If all things existed only in independent, disordered forms, it would be very difficult for humans to discover patterns within them, nor could they form a stable understanding.

Therefore, humans need to look for similarities and differences between things, grouping things with common characteristics together and assigning them common concepts—this is classification.

From a cognitive perspective, classification actually accomplishes an important transformation: humans no longer merely receive and store various pieces of information about the world, but instead, through abstraction and induction, transform this information into knowledge that can be understood and used.

For example, our understanding of the natural world today does not involve facing thousands of unrelated individual living things—cats, dogs, whales, eagles, roses, pine trees...—but rather placing them into different systems of categories:

Through this method of classification, humans no longer just memorize isolated things, but can further understand their position within a larger cognitive system. For example, when we learn about the animal “cat,” we do not just memorize the name “cat,” but also know that it belongs to larger categories such as animals and mammals.

More importantly, as classification systems gradually improve, humans can summarize common characteristics from a large number of concrete things and apply these experiences to new things. For example, when discovering a new mammal never seen before, although humans do not know all its details, they can make certain inferences about its likely characteristics based on past understanding of mammals—it is precisely through the process of continuously understanding concrete things and discovering commonalities that humans gradually form an understanding of the unknown world.

However, as understanding deepens, humans have found that merely knowing “what things exist” is still not enough to understand the world. Classification can help us answer “what characteristics a thing has,” but it cannot explain “why connections arise between things.”

Therefore, humans further began to explore the relationships between things and the laws behind these relationships. For example, the development of biology is not simply about discovering more species, but about continuously establishing deeper connections: evolutionary relationships between species; relationships between genes and biological traits; and relationships between organisms and their environment. These connections have gradually transformed life sciences from a “list of species” into a knowledge system capable of explaining life phenomena.

A similar process has occurred in other scientific fields. For example, the development of physics is not just about continuously recording more natural phenomena, but also about searching for unified laws among these phenomena and describing these relationships through mathematical models.

On the surface, humans are constantly accumulating information about the world; but more accurately, what humans are truly deepening is their understanding of things, relationships, and laws in the world, and this understanding is gradually formed in the process of continuously establishing connections.

2 From Information to Understanding: Why Do Humans Need Structured Knowledge?

1 From Information to Cognitive Structure: How Humans Understand Knowledge

Establishing connections is only the beginning of understanding the world. A further question is: once humans discover and understand these connections, how exactly do these understandings enter their own cognitive systems, and how are they preserved, retrieved, and reused over the long term?

If the process of human learning were merely about continuously collecting and preserving information, then learning should be a very simple matter: just keep reading, observing, and recording, save all the information encountered, and retrieve it when needed.

But reality is not like this. Many people, when learning a new field, experience a similar process: they are exposed to a large number of new concepts, terms, and methods in a short period, and can even memorize many of them, but after some time, they find they still cannot truly understand the field.

The problem is not necessarily a lack of information. Often, what people lack is not more information, but rather that this information has not yet been understood, interpreted, and associated to form knowledge that can be used to understand and solve problems.

Computers can save data to storage media according to addresses and read it accurately when needed; but the human brain does not work this way. Humans do not save information received from the outside world piece by piece in a completely independent manner; instead, they continuously process, classify, interpret, and abstract information. In this process, information that was originally just received is gradually transformed into knowledge that can be understood and applied, and this knowledge further forms higher-level cognitive structures through association. One important reason why information needs to be processed in this way is that the human cognitive system itself has processing limitations.

Research on working memory in psychology shows that the amount of information humans can hold and process simultaneously is very limited. Early research proposed “Miller's Law,” which suggested that human short-term memory capacity is about 7±2 chunks of information; subsequent research suggests that in more complex cognitive tasks, this capacity may be even lower.

Regardless of the specific capacity, these studies all point to one fact:Humans cannot rely on processing large amounts of isolated information simultaneously over the long term to understand a complex world. If faced with dozens of unrelated pieces of information, the brain will quickly experience a huge cognitive load, a limitation that is particularly evident when learning a complex field.

For example, someone new to computer networks might quickly memorize:

TCP
UDP
IP
DNS
HTTP
HTTPS
NAT
BGP
CDN
Firewall

These contents themselves are real and valid information, but if they only exist in the brain in isolation from one another, it remains difficult for a person to determine how to use this knowledge when faced with a practical problem. For someone new to networking, this may just be an ever-growing list of terms.

However, when this information is no longer just arranged by category, but instead forms interconnected structures based on those categories, what the brain processes is no longer a set of isolated terms, but rather several wholes with internal relationships:

This is “chunking” in cognitive psychology. Chunking is not simply about reducing the amount of information, but rather organizing and integrating information to combine multiple independent pieces of information into a higher-level cognitive unit, thereby reducing the brain's burden when processing complex information. When this information is understood, interpreted, and organized to form cognitive units with clear meanings, it begins to transform from mere fragments of information into knowledge that can be understood and used.

From this, we can see that the process of humans understanding the world is not a process of simply accumulating information, but a process of continuous processing and organization: information is understood, interpreted, and abstracted to form knowledge, and knowledge then gradually forms a cognitive structure by establishing connections.

Information, knowledge, and cognitive structure therefore constitute a continuous hierarchical relationship: information is the raw material for humans to understand the world, which can be recorded, transmitted, and preserved, but itself does not represent understanding; knowledge is formed on the basis of understanding, interpreting, and abstracting information, and can be used to explain phenomena, solve problems, and guide actions; cognitive structure further describes the overall organizational network formed after connections are established between knowledge, which determines how a person understands new information and transfers existing knowledge to new problems.

2 From Connecting Information to Understanding: How Cognitive Structures Generate Meaning

Through chunking, humans can compress a large amount of complex information into cognitive units that are easier to process. These cognitive units, after being understood, organized, and abstracted, can further form independent knowledge units. However, a single knowledge unit is still only a partial understanding of a certain object or phenomenon; only when connections are established between different knowledge units can a cognitive structure be further formed.

So, how exactly do knowledge units establish connections with each other? And how do they use these connections to jointly explain a more complex phenomenon?

Taking computer networks as an example, a person who is new to computer networks might know: DNS is used for domain name resolution; TCP is responsible for reliable transmission; HTTP is used to transmit web page content; CDN can improve access speed.

These basic descriptions of the concepts are correct in themselves. However, if DNS, TCP, HTTP, and CDN only exist in the brain as several independent concepts, then when faced with practical problems, it is still difficult to understand why these technologies appear together and how they work together to complete a network access.

Before being exposed to CDNs, a person may have already formed such a basic access model:

In this cognitive model, the user finds the target server through DNS and then communicates directly with the server. This model is not wrong; it can explain a large number of basic website access processes.

However, when learning further about CDNs, a new question arises: how should CDNs be integrated into the original access model? If one only memorizes “CDN is a website acceleration technology,” this concept still remains at the level of definition and has not truly entered the existing cognitive structure.

True understanding requires integrating CDN into the original model and readjusting the original access process:

At this point, CDN is no longer just a “tool to increase speed,” but becomes an integral part of the entire website access architecture. At the same time, the original concept of “target server” has also changed—Before: the user accesses the server, and the server returns content; Now: the user first accesses the CDN node, and the CDN directly returns cached content or further requests the origin server depending on the situation.

This is the change after a new concept enters the existing cognitive structure: it does not simply add a new knowledge point, but changes the relationship between existing knowledge units and reorganizes the original understanding model.

The “Schema” theory in psychology describes a similar cognitive process. Schema can be understood as a cognitive framework used by humans to organize experience and understand things, reflecting how concepts, experiences, and relationships are organized within the cognitive structure.A concept is truly understood not just because people memorize its definition, but because it has entered the existing cognitive structure and established stable connections with other concepts.

This also explains why many people, when learning a new field, read a lot of materials and memorize many concepts, yet still fail to truly master the field. This is because learning is not simply increasing the amount of information in memory, but constantly adjusting and improving the existing cognitive structure. Only when new concepts can establish connections with existing knowledge will they truly integrate into the cognitive structure and become knowledge that can be used for understanding, reasoning, and application.

Experts are often able to make judgments more quickly than ordinary people when faced with unfamiliar problems, also because they have formed a large number of stable conceptual connections through long-term learning and practice. When new problems arise, they can map them to their existing cognitive structures and analyze them using accumulated past experience.

Therefore, the process of human learning and understanding the world is not simply saving more information in the brain, but constantly establishing new connections on the basis of existing cognitive structures, so that new information gains a clear position, meaning, and application scenario. In other words, humans do not simply accumulate information, but constantly process information into knowledge, and form and improve cognitive structures through the connections between knowledge.

3 How Humans Actively Build Their Own Cognitive Structures

However, the cognitive structure is not fixed once formed. For simple problems, humans can gradually build and adjust their understanding through their own accumulation of experience; but when faced with complex fields, the number of concepts to be processed continues to increase, and the relationships between concepts become more and more complex, making it increasingly difficult to rely solely on the brain to naturally form and maintain cognitive structures.

Therefore, humans have gradually developed various methods of auxiliary cognition to help themselves observe, verify, expand, and express existing cognitive structures.

Although these methods take different forms, they essentially solve the same problem:How to make the cognitive structures hidden in the human brain clearer, more stable, and capable of continuous improvement.

Visual Structuring: Mind Maps and Concept Maps

When humans understand a complex field, a difficult problem is that even if some connections have already formed in the brain, these connections are often implicit and difficult to observe directly.

Therefore, humans have developed methods such as Mind Maps and Concept Maps to express the relationships between concepts through external graphics.

For example, in the previous section, we demonstrated how CDN enters the original website access cognitive structure through the comparison of two website access models. In fact, these two models can also be converted into mind maps.

Basic access model:

Access model after CDN is added:

Diagramming is by no means a simple “text-shoveling” or “layout makeover,” but rather making implicit cognitive structures explicit: it helps us see clearly—which knowledge units belong to the same module (categorization); what kind of relationships exist between which knowledge units (association); and how the introduction of a new knowledge unit reconstructs the original system like triggering dominoes (evolution).

Whether mind maps or concept maps, their essence is a “visual lens” of the brain's thinking process.

Semantic Structuring: The Feynman Technique

Mind maps and concept maps address “whether we can see the relationships between knowledge,” while the Feynman Technique further addresses “whether we truly understand these relationships.”

Often, people can remember a concept and identify its position in the knowledge structure, yet cannot explain why it is in that position, or why such a relationship exists between it and other concepts.

The Feynman Technique focuses precisely on this process. It not only helps learners discover whether connections have been established between knowledge, but also exposes gaps in understanding through active expression, driving the further refinement of their cognitive structure:

For example, a person might memorize: “CDN is a technology that improves website access speed.” This statement itself is not incorrect, but it is merely an isolated knowledge node.

If we ask further, “Why can a CDN improve access speed?” it requires understanding why user requests do not necessarily access the origin server directly, but can instead retrieve content through edge nodes deployed in different regions. If we continue to ask, “Why does DNS affect CDN node selection?” it requires understanding how the DNS resolution process participates in selecting the user request path. If we ask yet again, “Why does a CDN sometimes still need to return to the origin server?” it requires understanding the relationships among caching mechanisms, cache invalidation, and the origin server.

These questions are not actually testing a single concept; rather, they are testing whether the concept of CDN has established connections with other knowledge such as DNS, HTTP, caching, distributed nodes, and origin servers.

Inability to answer these questions indicates that although the concept of “CDN” has been memorized, it remains an isolated knowledge unit and has not truly integrated into the existing knowledge structure. Through active explanation, learners can discover what they truly understand, what remains merely memorized, and what connections are still missing in their existing cognitive structure.

Therefore, the Feynman Technique is not essentially a simple memory trick, but rather a method of checking and refining cognitive structures through verbal output.

Network Structuring: The Zettelkasten Method

As the volume of knowledge continues to grow, even with an established cognitive structure, people still face new challenges: new concepts constantly emerge, old knowledge continuously forms new connections, and the original structure requires ongoing adjustment. At this point, relying solely on one-time organization can no longer meet the demands of long-term accumulation.

The Zettelkasten method offers a different approach: it does not require knowledge to be fixed within a specific classification system, but instead allows the knowledge network to grow continuously through independent knowledge units and the connections between them.

For example, when learning about CDNs, one can create separate cards for different knowledge units—such as CDNs, DNS resolution, caching mechanisms, and edge computing—and form a knowledge network through their associations:

Unlike mind maps, this approach does not require knowledge to revolve around a single central theme; instead, it allows each knowledge unit to become a node that can exist independently and connect continuously. As learning deepens, new knowledge can be constantly integrated into the existing network, rather than requiring a redesign of the entire classification system every time.

This approach is actually closer to the characteristics of human cognitive structure: knowledge is not a simple directory, but a dynamic network composed of a vast array of concepts and relationships.

Expressive Structuring: The Pyramid Principle

Beyond building and maintaining cognitive structures, humans also need to convey their understanding to others. However, the cognitive structure in an individual's mind is usually complex; if output directly, it can easily devolve into a mass of scattered information.

Therefore, humans have also developed methods to help express complex ideas, such as the Pyramid Principle, which emphasizes: state the core point first, and then organize the supporting reasons according to logical relationships, thereby reorganizing complex information into an easy-to-understand structure.

For example, after a person understands CDNs, a vast amount of related knowledge may have already formed in their mind:

DNS
CDN 节点
缓存
源站
TCP
HTTP
访问距离
流量调度

Although there are connections between these contents, if explained directly to others, it can easily become a simple pile-up of technical terms. The listener can receive the information but will find it difficult to understand why these concepts appear at the same time and how they work together to explain the role of CDN.

When organizing expression according to the Pyramid Principle, one first needs to determine a core question: “Why do websites need CDN?”, and then organize supporting concepts and reasons around this question:

The output content in this way is no longer a list of knowledge, but a logical structure that can be understood. It does not change the cognitive structure itself, but reorganizes and presents the existing cognitive structure according to the purpose of expression, so that complex knowledge can be understood and transmitted in a clearer way.

From these methods, we can see that the various learning and thinking tools developed by humans are not essentially meant to simply save more information. The problem they jointly solve is actually how to make the connections between concepts more stable, clear, and reusable, and gradually transform from an individual's implicit cognitive structure into an observable and expressible structure.

4 From Cognitive Structure to Knowledge Structuring

However, the external expression of an individual's cognitive structure does not mean that knowledge has completed its structuring. Although methods such as mind maps and note-taking systems can help individuals record and express parts of their cognitive structure, when the scale of knowledge continues to expand, or when knowledge needs to be preserved, shared, and reused across individuals over the long term, relying solely on personalized recording methods still has limitations.

Therefore, humans need a more systematic method to explicitly represent the concepts in knowledge and the relations between them, so that the structure originally implicit in human cognition can be preserved in anunderstandable, reusable, and extensible form—this is the very problem that knowledge structuring aims to solve.

Unlike simple information organization, knowledge structuring does not focus on adding more content, but on explicitly representing the concepts, attributes, and relations between concepts in knowledge, so that some of the relations originally implicit in the human cognitive structure can be clearly represented.

For example, three independent knowledge units:

A

B

C

B

C

  A
 / \
B---C

But once relations are established between this information:

A

3 From Knowledge Organization to Knowledge Engineering: How Do Humans Build Structured Knowledge?

1 Knowledge Engineering: Making Knowledge a Representable Structure

/ \

B---Cthey are no longer just three isolated points, but form a whole with a stable internal structure. In this structure, each node not only has its own meaning, but also gains a more explicit position and meaning due to its relations with other nodes.Therefore, the focus of knowledge structuring is not simply to record existing information, but to further explicitly represent: which knowledge units constitute the knowledge system of a domain, and how these knowledge units are interrelated.

When knowledge needs to be preserved for a long time, shared, or even processed by computers, the understanding that exists only in personal cognition needs to be externalized and represented in a more explicit way.

To address this need, humans began to explore how to describe the

concepts and their relations in knowledge

in an explicit form. This exploration gradually formed the research direction of Knowledge Representation, and further developed into the research field of Knowledge Engineering—which focuses on how to acquire, represent, organize, and utilize knowledge, so that the knowledge and relation structures originally existing in human cognition can be preserved and processed in a structured form.

如果满足条件A，
那么可以推导结论B。

However, computers themselves do not possess the domain understanding that humans develop through long-term learning. Therefore, if we want computers to process complex problems, we need to convert human knowledge into a structured form that computers can represent, store, and process. This has made "how knowledge is acquired, represented, and utilized" an increasingly important issue in artificial intelligence research.

Against this background, knowledge engineering gradually developed. Early knowledge engineering mainly focused on how to acquire and represent expert experience and convert this experience into a form that computers could process. Among these, Expert Systems became the most representative application form of this stage.

• In expert systems, the judgment experience of human experts is usually represented as a series of rules:
• If condition A is met,
• then conclusion B can be derived.
• Through a large number of such rules, computers can simulate the knowledge and reasoning processes of experts in specific domains (such as medical diagnosis, financial risk assessment, industrial fault analysis, etc.).

However, as the scope of application continued to expand, people gradually discovered that knowledge in the real world does not consist only of simple rules. Knowledge in a domain usually contains a large number of different types of objects and complex relations between these objects. For example, in the field of computer networking:

CDN is a type of network service;

• CDN contains multiple distributed nodes;
• DNS affects the CDN nodes accessed by users;
• CDN nodes reduce the pressure on the origin server through caching.
• These contents are not just a few independent judgments, but a knowledge structure composed of multiple concepts and the relations between them.

Therefore, modern knowledge engineering has gradually evolved from "how to store rules" to "how to represent knowledge structures." It has begun to focus on:

Which objects constitute the knowledge structure;What attributes these objects have;

2 The Basic Units of Knowledge Representation: Entities, Attributes, and Relationships

What relations exist between different objects;

• How these relations together form an understanding of a certain domain.

• In other words, knowledge engineering attempts to express the knowledge network formed in the human cognitive process in an explicit structure, so that knowledge no longer exists only in personal experience and understanding, but can be preserved, shared, and further processed by computers.

• To achieve this structured representation, a fundamental question must first be solved:

What exactly are the basic building blocks of knowledge?

To represent knowledge in a structured form, the basic elements that constitute knowledge must first be clarified. From the perspective of knowledge representation, people usually organize knowledge around three basic components:

Entity: Represents an object in knowledge, i.e., a basic unit that can be independently described and establish relations with other knowledge elements. It can be a concrete object in reality, such as a person or a place, or an abstract object with an explicit meaning, such as a certain technology, software system, network component, etc.

Attribute: Describes the characteristics possessed by an object;

Relation: Describes the connections between different objects.

These three parts together form the foundation of most knowledge structures.

Entity: Objects in Knowledge

When humans understand the world, they first need to identify “what exists.”

For example:

对象：
CDN

属性：
类型：网络服务
功能：缓存内容、优化访问路径
特点：通过分布式节点降低用户访问延迟

Therefore, the first step of knowledge structuring is to identify various objects in the world and represent them as independent knowledge units.

Attribute: Describing what characteristics an object has

However, even having a large number of objects and attributes is still not enough to form complete knowledge. This is because what truly determines the depth of understanding is the nature of the connections that exist between objects.

For example:

DNS
 |
选择
 |
CDN节点
 |
缓存内容
 |
来自
 |
源站服务器

What is expressed here is not information about individual objects, but the relationships between them: DNS affects which CDN node a user accesses; CDN nodes can cache content from the origin server; and CDN nodes need to access the origin server in the event of a cache miss.

It is precisely these relationships that allow objects, which originally existed independently, to form a knowledge structure capable of explaining real-world phenomena.

Entities, attributes, and relationships in knowledge engineering are not three arbitrary abstract concepts, but correspond to the most fundamental cognitive ways humans understand the world: we first identify what objects exist in the world, then understand what characteristics these objects have, and finally explore how these objects are interconnected.

What knowledge engineering does is to represent this way of understanding, which originally exists in human cognition, in an explicit structural form. When a large number of entities, attributes, and relationships are organized together, they form a structured representation capable of describing knowledge in a specific domain.

3 From Knowledge Units to Knowledge Networks: How Knowledge Forms a System

As the scale of knowledge continues to expand, merely storing individual, independent units of knowledge is no longer sufficient. This is because a single knowledge unit can only describe localized information, whereas a truly complete understanding of a domain comes from the overall structure formed among a large number of knowledge units. Knowledge in the real world is not composed of a few isolated knowledge units, but contains a vast number of interconnected knowledge units and relationships. In this structure, the value of knowledge comes not only from the individual knowledge units themselves, but even more from the connections formed between different knowledge units.

Therefore, the further problem that knowledge engineering needs to solve is:As knowledge units continuously increase, how can we connect these scattered units through their relationships to form a knowledge network that can be understood, queried, and utilized?

Through these connections, computers can further understand: what relationships exist between different concepts; where a certain knowledge unit is positioned within the entire domain; and how a concept, together with other concepts, constitutes a complete knowledge system.

Knowledge Graph is a method for building structured knowledge networks developed on this very idea. By organizing a large number of knowledge units into a network composed of nodes and relationships, it enables originally scattered knowledge to be stored and utilized in a structured manner:

Here, different nodes represent different knowledge units, and the connections between nodes represent the relationships that exist between them. In this way, knowledge is no longer just passively stored information, but becomes a structure that can be further queried, analyzed, and expanded by computers.

The application scope of knowledge graphs is extremely broad and is not limited to technical fields. As long as a domain contains a large amount of knowledge that needs to be organized and associated, a structured representation can be established through knowledge graphs, such as in fields like interpersonal relationships, healthcare, scientific research, business analysis, and cultural knowledge:

Although knowledge graphs in different domains contain different types of entities and relationships, their core idea is consistent: to organize the knowledge units, attributes, and relationships in a domain into a knowledge network that can be understood and utilized.

4 AI Era: Structuring Challenges After the Increase in Knowledge Expression

1 From Knowledge Acquisition to Knowledge Generation: Changes Brought by Large Language Models

In the previous development of knowledge engineering, humans mainly adopted a method ofexplicitly constructing knowledge: first identifying the basic units in knowledge, then organizing them through entities, attributes, and relationships, and ultimately forming a knowledge network that can be processed by computers—knowledge graphs are a typical representative of this approach.

But this approach also has an obvious limitation:Knowledge needs to be actively discovered, organized, and defined by humans. As the scale of knowledge continues to expand, relying on manual efforts to continuously construct and maintain complete knowledge structures faces increasingly high costs. Therefore, artificial intelligence began to explore another path:Can computers stop relying entirely on human-defined knowledge structures and instead learn patterns and regularities directly from large amounts of data?

Around this goal, artificial intelligence has gradually evolved from relying on human-defined knowledge to using data to train models, allowing computers to extract patterns and regularities from large amounts of data through statistical learning and form implicit representations of the data. This means that the way knowledge is acquired and represented has begun to shift from “manually constructing explicit structures” to “machines learning from data.”

The emergence of Large Language Models (LLMs) represents an important stage in this direction of development. By learning language patterns and statistical associations from massive amounts of text and generating new content based on context, they have endowed computers with unprecedented capabilities for knowledge representation and content generation.

At the same time, the way humans acquire knowledge has also changed. Before the emergence of LLMs, people usually needed to actively search for materials, read and understand them, and organize existing information; in the era of LLMs, people can interact directly with models through natural language, allowing the models to quickly generate complete answers on a given topic.

This means that the process of human information acquisition and knowledge representation has begun to shift fromactively searching for and organizing existing materials, to gradually interacting with intelligent systems capable of generating content.

But new questions have also emerged:Does the ability to generate clearly structured and fully expressed answers mean that computers have formed cognitive structures similar to those of humans?

2 Structuring Challenges After the Increase in Knowledge Expression

LLMs have lowered the cost of knowledge representation, enabling the rapid generation of large amounts of structurally complete and linguistically fluent content. However, an increase in knowledge representation does not mean the automatic formation of knowledge understanding. This is because the answers generated by LLMs are usually generalized expressions aimed at a broad audience—they can cover a large number of concepts and organize content according to a certain logic, but this way of organization does not necessarily align with the way each individual's existing cognitive structure is organized.

For those who already have a relevant knowledge base, AI-generated content can be quickly connected to their existing cognitive structures, helping them supplement information, discover connections, and further expand their understanding. However, for those lacking a foundation, even if the answer itself is complete enough, the vast number of knowledge units and relationships it contains may still fail to establish effective connections with their existing cognition. This is because, in the absence of a necessary knowledge base, new knowledge units often lack the concepts and relationships with which they can connect, making it difficult to truly integrate into the original cognitive structure.

For example, when someone new to computer networking asks an AI about CDNs, the AI can generate a complete explanation containing knowledge representations such as the definition of a CDN, its workflow, caching mechanisms, and edge nodes.

But if the user does not previously understand how DNS affects access paths, how HTTP requests reach servers, why caching can reduce pressure on origin servers, or why distributed nodes can improve access efficiency, the relationships between these concepts remain difficult to truly establish. The problem is not whether the AI can generate a complete representation of knowledge, but whether these representations can be integrated by the users themselves into their existing cognitive structures.

This issue does not only affect individual learning; when knowledge representation becomes increasingly easy to generate while true understanding, verification, and structuring processes do not increase in tandem, the entire knowledge ecosystem may also change.

In the past, a large amount of knowledge on the Internet mainly came from human experience, professional creation, and long-term accumulation. Although limited in quantity, this content usually contained the authors' understanding, judgment, and organization of the field.

With the large-scale emergence of AI-generated content, a new content cycle is appearing on the Internet:

AI生成内容
      ↓
进入互联网
      ↓
被搜索、引用和传播
      ↓
成为未来模型训练数据

If a large amount of content remains primarily a recombination of existing information, lacking real-world experience, professional verification, and further structured organization, then the amount of information on the Internet may continuously increase, but the knowledge structures within it that can truly help humans form cognitive structures will not necessarily grow in tandem.

3 From Knowledge Acquirers to Knowledge Organizers

Before the advent of artificial intelligence, the process of human knowledge acquisition was usually accompanied by a large amount of information gathering and organizing. Due to the high cost of acquiring information, people had to actively search for materials, read and learn, and summarize experiences, gradually forming an understanding of a certain field in the process.At this stage, the process of knowledge acquisition itself was also an important process for humans to build cognitive structures.

But in the era of LLMs, this process is changing—when AI can quickly complete information retrieval, content organization, and knowledge representation, humans no longer need to invest a lot of time in simple information collection and organization, and the efficiency of acquiring knowledge is greatly improved.

However, the increased efficiency of knowledge acquisition has also changed the past process of knowledge formation. The understanding, association, and accumulation that naturally accompanied knowledge acquisition in the past are no longer guaranteed to occur today: people can quickly obtain a complete representation of knowledge on a topic from AI, but whether this content can be truly understood by oneself and connected with one's existing knowledge has shifted from being a ”naturally occurring outcome of the learning process” in the past to an ”actively completed process” today:

Therefore, in the AI era, the problem humans need to solve when facing knowledge has also become: how to understand this AI-generated content, how to discover the connections between them, and how to integrate them into existing cognitive systems—AI can help humans quickly obtain relevant information, summarize existing knowledge, discover potential connections, and provide explanations from different perspectives, but how to transform these knowledge expressions into one's own knowledge and further form a cognitive structure still requires active human participation.

A truly valuable knowledge structure is not just the accumulation of the quantity of information, but more importantly: which concepts constitute a field; what relationships exist between different concepts; which laws can explain phenomena; which experiences can be transferred to new problems... These factors determine whether a person truly understands and masters a certain field.

And these elements that make up the cognitive structure do not form naturally with the increase in the quantity of knowledge expressions. Whether in a traditional knowledge environment or in today's world of ever-increasing AI-generated content, knowledge must undergo understanding, organization, and association to transform from scattered information into a structured system—and this is precisely the core issue that knowledge engineering focuses on.

5 Returning to Personal Knowledge Engineering

In previous articles, when I had ChatGPT and Qwen work together to analyze the content of blog posts, hoping to discover the long-term stable concepts and relationships between different articles, I unexpectedly derived the basic units of knowledge engineering (see article:Personal Knowledge Engineering (I): From Article Index to Knowledge Structure—A Practice of AI-Collaborative Exploration of Personal Knowledge Systems）。

Through this period of exploration, I have gradually discovered that knowledge engineering is not just a technical method for sorting and organizing content; behind it, it actually corresponds to the fundamental way humans understand the world: humans do not simply memorize a large number of facts, but rather establish connections between concepts in the process of continuous learning and practice, forming an overall understanding of a certain field.

This also leads to the questions discussed in this article: Why does knowledge need to be structured? Why do humans need to understand the world through concepts, relationships, and cognitive structures? And in today's world where large language models can quickly generate a large amount of content, why has knowledge structuring actually become more important?

Therefore, this article is actually an exploration from practice to theory in the process of personal knowledge engineering practice: the first article focused on how to use AI to assist individuals in discovering connections between content and gradually establishing a knowledge structure; while this article further explores why these connections exist, as well as the knowledge structure and cognitive foundation behind them.

In this process, I have also further realized that knowledge structuring is not simply organizing existing content, but rather reorganizing scattered knowledge units according to a certain way of understanding, so that clear relationships are established between different concepts, and forming a knowledge system that can be understood, reused, and expanded. There is no single answer to this way of organization: for the same set of knowledge, different people may form different structures due to differences in experience, goals, and cognitive styles.

This means that knowledge structuring is actually also the externalization process of an individual's way of understanding: how a person organizes knowledge largely reflects how they understand the relationships between concepts, and the final result of this structuring also reflects the depth of their understanding of a certain field.

Through this article, I have also further clarified the long-term development direction of the blog: the information, practical experience, and reflections accumulated by an individual should not just be scattered records of content, but should leverage the concepts of knowledge engineering to gradually transform valuable content into knowledge, and combine it with one's own way of understanding to form a knowledge structure with personal characteristics that can continuously develop through the connections between concepts.

In the future, the content in the blog should not just be an ever-growing collection of articles, but should gradually form a knowledge network that can be continuously maintained, constantly expanded, and reflects the accumulation of personal long-term practice and reflection.

This is also the direction I will continue to explore in my personal knowledge engineering series.