[Memory and Records] Chapter 4 | Turning Records into Knowledge

Memory and Records—Why Do Humans Keep Records?

Chapter 3 | We Mistake Data for Knowledge

After reading the three chapters so far, some of you may have had a question arise.

I understand that memory and records are fundamentally different. I understand that records require context. I understand that data does not automatically become knowledge. So, what should I change? There is another bridge between understanding and acting. In this chapter, we will consider how to cross that bridge.

Shuhari and Fading Knowledge

In Japanese martial arts and traditional performing arts, there is a concept called "Shuhari".

"Shu" is the stage of faithfully following the forms taught by a master. "Ha" is the stage of applying those forms according to the situation after fully understanding them. "Ri" is the stage of transcending the forms and reaching one's own unique state of mastery.

These three are considered the universal path in the acquisition of skills.

This concept offers sharp insights into the problem of records and memory.

The "Shu" stage can be recorded. It can be written in procedure manuals. It can be broken down into manuals. The procedure of "mixing these materials in this order at this temperature" can be verbalized and preserved. However, once you enter the "Ha" stage, recording begins to become difficult. Why did you change these conditions? What kind of feeling led to that judgment? It can be verbalized, but it makes no sense unless compared with the person's own experience. And the wisdom of an expert who has reached the "Ri" stage is no longer suited for records. The sensations ingrained in the body, the intuition to read a situation, the judgments born from the accumulation of experience—these reside within memory and are outside of records.

In other words, Shuhari shows a structure where as proficiency deepens, knowledge moves away from records and closer to memory. This overlaps with the "transition from knowledge to wisdom" in the DIKW hierarchy. As one enters the stage of wisdom, it is no longer recorded from the outside, but resides as internal memory.

Here lies a structural problem that the manufacturing industry has faced for many years: the retirement of skilled workers.

A craftsman operating a lathe judges the state of the machine by the subtle vibration sound. A welder determines quality by the color and flow of the molten metal. A chemical researcher decides the next step by changes in the color and viscosity of the reaction liquid—no matter how much digitalization progresses, this wisdom disappears at the same time as retirement. In terms of Shuhari, knowledge that has grown to the "Ri" stage is lost from the organization along with one person's memory.

The wave of mass retirements in Japan, known as the 2025 problem, is accelerating this issue. Not only on the manufacturing floor, but in research and development, sales, and service—the "Ri" wisdom held by veterans is being lost without being recorded in every workplace. Organizations continue to lose a portion of their knowledge every year.

So, is this problem unsolvable? Perhaps it is impossible to solve completely. It is likely impossible to fully record the wisdom at the "Ri" level. However, by carefully recording the "Shu" and "Ha" stages, it is possible to shorten the time it takes for a successor to reach "Ri." By keeping a record of the thought process behind why an expert made a certain judgment, a successor can start from that starting point without having to go through trial and error from scratch.

Records cannot pass on wisdom itself. However, they can illuminate the path to wisdom.

The Illusion of "Recording"

The problem is that there are not many workplaces that are facing records from such a perspective yet.

As a test, try opening an Excel file you created six months ago. Can you explain right now why the numbers lined up there are those values? Can you trace when, who, and based on what judgment they were entered?

In many cases, it should be difficult.

This is not a matter of ability or personality. Entering numerical values and keeping records are completely different acts. The former is 'saving data.' The latter is 'giving meaning to data and preserving it.' This difference is the very gap between 'data' and 'information' in the DIKW hierarchy we saw in Chapter 3.

Many of us feel that we have 'kept a record' while saving data. But in reality, we might just be piling up data in the form of records. In terms of Shu-Ha-Ri, it is a state where even the fragments of 'Shu' (adherence) are not accurately preserved. When an expert retires with the wisdom of 'Ri' (transcendence), even those fragments cannot be found in the Excel files left behind.

So what is missing? The answer is one word: context.

Context, the lifeblood of records

Think of a library bookshelf. What is lined up there is not just a stack of paper. There are titles, authors, publication years, and classification numbers. Each book exists as information with context: 'when, who, what, and why it was written.' That is precisely why you can find the book you are looking for among tens of thousands using a catalog. Even a book from decades ago can be read and understood today.

On the other hand, the Excel files many of us create at work are closer to a pile of books without titles. The numbers are there. But we don't know when, who, in what situation, or why those values became what they are. A state where files like 'Final Version,' 'Final Version 2,' and 'Final Version_Revised' are lined up is the same as a bookshelf without spines or a catalog.

In the past, 'memory' supplemented this context. Memories like 'That data was created by Mr. Tanaka during that meeting...' connected the files to their meaning. However, when Mr. Tanaka retires, that context disappears. Memory leaves with the person. This is not an individual problem, but a problem of record design.

The data 'yield 82%' says nothing on its own. But if you add a line saying, 'Yield 82%—lower than last time. Possibly affected by a change in raw material lot; will check next time,' then when someone in the future reads it, the situation and judgment at that time are reproduced. This one line turns data into information. The jump in the DIKW hierarchy we saw in Chapter 3 begins with this one line.

Adding context to a record is an act similar to writing a letter to a future reader. And that reader might be your future self, a successor colleague, or perhaps generative AI. To any reader, a record without context is a letter where you can understand what is written, but not why it was written.

Design changes behavior

So why can't many people leave context behind? The reason might be that it is troublesome. But a more fundamental cause is that there is no mechanism in place.

Records are not something you do with sheer willpower. They are something you do with design.

Here, we notice a paradox that the modern age holds.

In the past, the number of people who created records was limited. Scribes carving cuneiform, monks writing manuscripts, merchants keeping ledgers—the people responsible for records were a part of society and received specialized training for that role. Libraries had librarians who professionally collected, classified, preserved, and guided records. It was an era where creating records and designing records were entrusted to the same people.

However, in the modern age, we have entered an era where anyone can create data. Digitization has brought about the democratization of records. But at the same time, the specialized philosophy of record design has not been democratized. Anyone can create data now. But the people trained to think about record design have not kept up. In terms of Shu-Ha-Ri, the current situation is that everyone continues to create records freely without even the 'Shu' stage being designed.

As I wrote in Chapter 2, data scientists are the equivalent of modern librarians. They play the role of structuring data and delivering it as knowledge. However, not every workplace has a data scientist. In many workplaces, the person who creates the data and the person who designs the record remain separated, or there is no one designing it at all.

That is why the question is whether an individual can become their 'own librarian.'

Including the date and purpose in a file name is equivalent to assigning a classification number in a catalog. Adding a line of reasoning next to a numerical value is equivalent to an author giving context to a book. Managing versions with Git is equivalent to a library storing revised editions and first editions separately. We have the tools to practice at an individual level what librarians once handled as an organization. The problem is not the tools, but whether you have the philosophy of design.

And this philosophy of design does not require difficult technical knowledge. You can start with the 'Shu' of Shu-Ha-Ri—that is, deciding on basic rules and following them. Naming conventions, how to leave context, version management. Just by designing these three things for yourself once, the quality of your records will change significantly.

As a researcher, what has changed

I have had an experience where I realized the power of record design as an organization.

For many years, all data generated in materials development was managed in uniquely formatted Excel files. Each product, department, and person created Excel files in a 'format easy for them to use.' The notation for raw materials differed from person to person, and the names of manufacturing processes were inconsistent. Even though the data existed, it was almost impossible to compare across departments or integrate and analyze it. Everyone thought they were recording things carefully, but in reality, they were continuing to create a collection of disconnected records as an organization.

It was truly a state where fragments of the 'Shu' (adherence) in Shu-Ha-Ri were scattered in different places based on individual interpretations.

At one point, a movement arose to create a database for the entire organization. We used PostgreSQL to standardize the raw materials database and unify the records of common manufacturing processes. We eliminated variations in notation, standardized naming conventions, and designed the system so that data would be accumulated in the same structure regardless of who entered it. This work was by no means flashy. It was tedious, time-consuming, and involved continuous adjustments to the details.

One of the goals was to utilize it for materials informatics. If the dataset grew and the structure was unified, machine learning methods like Bayesian optimization could be used. Data, not just experience and intuition, would start to suggest the next candidate conditions for experiments. This goal was achieved through the creation of the database.

However, there was an unexpected change.

With the creation of a unified database, different development teams were able to refer to the same data. Until then, there had been no opportunity to know what the neighboring department was developing or what materials they were using. That became visible through the common database. Realizations such as 'That department is also using this raw material we use' and 'This manufacturing process is similar to ours' emerged, and information exchange and collaboration across departments began naturally.

The design of records changed communication.

This overlaps with the story of the library we saw in Chapter 2. By having a common catalog, different users can access the same knowledge. The 'Shu' knowledge that was closed within individual memories begins to turn into 'explicit knowledge' that can be shared by the organization. What was once trapped in records close to individual memory, such as Excel, was opened up to the organization by migrating to a designed database.

The day generative AI becomes a librarian

Organizing the design of records is not just for your own sake today. In the near future, well-maintained records will be the subject read by a new librarian called generative AI.

A technology called RAG (Retrieval-Augmented Generation) is a mechanism where generative AI searches and refers to relevant records from databases or document collections before creating an answer. In other words, if an organization's records are well-maintained, there is a possibility that generative AI will read the records like a library librarian and answer questions like 'If you ask that person, you'll know.'

However, there is an unchanging premise here as well. If the original records are not designed, AI will not function either. Even if AI reads a list of numbers in an Excel file without context, no meaningful answer will be produced. In a state where even the 'Shu' of Shu-Ha-Ri is not recorded, AI cannot inherit anything. As I wrote in Chapter 3, the quality of the records determines the upper limit of the AI's intelligence.

In the past, knowledge was inherited through oral tradition. Next, it became possible to inherit it through paper and libraries. Now, we are entering an era where AI will help with inheritance if there are designed records. However, one thing remains unchanged no matter what changes. If there is no context in the records, inheritance will fail in any era.

Supplementing the limits of memory with records

The wisdom of an expert who has reached the 'Ri' (transcendence) of Shu-Ha-Ri cannot be completely recorded. It resides within human memory and has parts that transcend language. However, by leaving the 'Shu' and 'Ha' (break) stages as carefully designed records, an organization can inherit the path to wisdom.

Through the experience of unifying organizational records with PostgreSQL and seeing different development teams begin to collaborate, this is what I felt. Record design is not about tools. It is about the philosophy of correctly separating the roles of memory and records.

What memory should handle—emotions, intuition, judgment, and creativity—cannot be replaced by records. But data, conditions, background, and results—these are things that records should handle, and they will inevitably deteriorate if left to memory. From the moment you become conscious of this division of roles, your approach to records changes.

And changing the design of records is a choice made one line at a time today. Add a line of context next to a numerical value. Place a single memo in a folder. Decide on a naming convention just once. All of these are small, plain, and do not show results immediately. But that accumulation changes your time six months from now, changes the depth of the organization's knowledge a year from now, and eventually changes the very state of 'you have to ask that person to know.'

Recording the 'Shu' of Shu-Ha-Ri is also about building a foundation for the next generation to advance to 'Ha' and then to 'Ri.' When you stop trusting memory too much and start designing records, the landscape of your work begins to change quietly.

To the next chapter

However, there is a significant barrier between organizing personal records and organizing records for an entire organization.

You can change your own habits if you decide to do so. But trying to change an entire organization suddenly becomes difficult. You set rules, but no one follows them. Knowledge disappears entirely when someone resigns. You have no idea what your predecessor was doing—why does this problem, which exists in every workplace, occur? And how can we turn individual knowledge into organizational records?

In the next chapter, we will consider the collapse of records in organizations and how to design them.

Continue here👇

Start from Chapter 1 here👇

#CreativeAwards2026 #BusinessCategory