We once explored an idea in an earlier case:

For a system to maintain itself, it must continually reconfigure the relationship between its own resources and the environment around it.

But in reality, systems often choose another path — one that costs less.

They inherit answers that have already been tested by the previous generation.

And when the environment continues to move, that inheritance, which once reduced the cost of figuring things out, can gradually become a boundary around what comes next.

We Think We Are Judging

We rarely question the decisions we are making.

More often, we simply find an answer that seems reasonable.

Go to school.

Graduate.

Get a job.

Buy a house.

Retire.

Even something as ordinary as “retirement at 65” is rarely questioned:

Why 65?

It seems so normal.

So normal that we no longer need to know where it came from.

The previous generation did it this way.

So the answer gets passed on.

And when an answer is passed down for long enough, it can gradually change from “a method that once worked” into “a rule that was always supposed to exist.”

That is where the problem begins.

A Story About an Artificial Organ

In another lecture given in 1984, Michio Kushi told a story about a friend from his younger years.

Many years later, that friend had become a leader of Japan’s artificial-organ industry association. When they met again, Kushi spent two days learning about the development of artificial-organ technology and the industry surrounding it.

What he encountered was already a highly developed system.

What the technology could do.

Which organs could be replaced.

How much it would cost.

Whether there was a market.

How the industry could reduce costs.

How banking, insurance, medicine and pharmaceuticals could come together to form a new market.

Even the next ten years could be planned.

There is something interesting here.

The story gives us two different ways of looking at the same situation.

One asks:

We already have the ability to solve this problem.

The other asks:

Why did we arrive at a point where we needed this solution in the first place?

These do not necessarily represent opposing positions.

They are two different questions that can be asked separately.

One starts from the solution that already exists.

The other goes back to the process that created the problem.

Getting Answers From Results

Human beings are exceptionally good at the first approach.

We look at the past.

We identify successes and failures.

We organize the causes.

We find patterns.

Eventually, we arrive at an answer.

That answer can be written into books.

It can become education.

It can become institutions.

It can become corporate procedures.

It can become common sense for the next generation.

This is an extraordinary ability.

Because it compresses the cost of past trial and error.

Someone walked this road yesterday.

The person walking it today does not have to start from zero.

That is how civilization accumulates.

The problem is not this ability.

The problem begins when:

A result is turned into an answer, and the answer gradually separates from the environment in which it was created.

The longer an answer is remembered, the easier it becomes to lose the circumstances that gave birth to it.

We remember:

“This worked.”

But gradually forget:

“Why did it work then?”

Or Do We Start Again?

The other path is completely different.

It does not begin with an answer.

It begins by looking at the present.

Observe.

Form a hypothesis.

Act.

Receive feedback.

Judge again.

Act again.

It is slower.

And more expensive.

Because every change in the environment may require us to pay the cost of trial and error again.

But it has something that a purely result-based approach struggles to provide:

It remains connected to the environment of the present.

So perhaps the real divide was never:

Having an answer vs. having no answer.

It is:

Taking an answer from the results of the past.

Or

Producing an answer through the process of the present.

The first allows civilization to accumulate.

The second allows civilization to continue adapting.

And perhaps what we actually need is not to choose between them.

It is to know:

When can we copy?

And when must we judge again?

When Experience Starts Replacing Judgment

A successful method is often the easiest thing for a system to turn into a limitation.

Because success leaves a memory.

Memory becomes a process.

Processes become institutions.

Institutions are then taught to the next generation.

So:

Success → Replication → Standardization → Inheritance.

Eventually, something that once required judgment becomes something that no longer requires judgment.

This is why experience is both valuable and dangerous.

Experience can save us from taking unnecessary detours.

But it can also make us stop looking at the road.

Algorithms Make This Faster

In the past, humans passed answers to the next generation.

Now, we also pass answers to algorithms.

What you watch.

How long you stay.

What you click.

What you buy.

What makes you stay.

What makes you leave.

A system builds predictions from past behavior.

Then sends those predictions back into your world.

So the loop becomes:

Past behavior
→ Pattern
→ Prediction
→ Feedback
→ New behavior
→ Further learning.

An algorithm does not even need to tell you:

“You should believe this.”

It only needs to become increasingly accurate at knowing:

“What are you most likely to accept?”

Freedom has not necessarily disappeared.

Choice still exists.

But another question begins to emerge:

If the options we see have already been filtered by our past behavior, are we still judging?

Or are we simply choosing the answer that feels most familiar?

AI Takes It Further

When this way of building patterns from the past is no longer used only to predict human choices, but also begins to participate in the generation and inheritance of knowledge, the question moves to another level.

If AI learns from data left behind by humans, that is natural.

Human beings have always accumulated knowledge through inheritance.

But what happens when the next generation of AI increasingly learns from content produced by the previous generation of AI?

What is it learning?

The world?

Or:

The previous generation’s description of the world?

The two are not exactly the same.

Because the data humans leave behind is already the result of selection.

We record some things.

We ignore others.

We study certain questions.

We do not study others.

Successful answers are preserved.

Failed answers disappear.

So the data itself is already the result of history being filtered.

And when machines learn from those results again,

and the next generation of machines learns from the outputs of the previous generation,

the real question is no longer simply:

Is there enough data?

It becomes:

Where does new judgment come from?

The Business World Has Faced the Same Problem

This is why putting Nike, Kodak and Lee Kum Kee together is more interesting than looking at any one of them alone.

They are not three stories of “success and failure.”

They are three different observation windows.

Nike: Can a Successful Answer Become a Limitation?

Nike continues to place sport, product innovation, its brand and its relationship with consumers at the center of its strategy.

But its own recent annual reporting also acknowledges that consumer preferences, as well as trends in sport and design, change — requiring adjustments to its product portfolio, channels, new products and market strategies.

So the question is not:

Was Nike’s previous approach right or wrong?

It is:

How much of what worked in the past should still be copied in exactly the same form today?

Kodak: Seeing a New Answer Does Not Mean the Judgment Is Complete

Kodak actually invented the world’s first digital camera in 1975.

It later introduced digital cameras, digital imaging products and related systems as well.

So the story that “Kodak did not know the digital age was coming” is an oversimplification.

The more interesting question is:

Is knowing that a new answer exists the same as allowing the entire system to judge again?

Not necessarily.

Because a new technology is only part of an answer.

The harder task is to reconsider:

If the world has changed, what problem were we actually trying to solve in the first place?

Lee Kum Kee: Inheritance Does Not Necessarily Mean Replication

Lee Kum Kee began in 1888 and has now crossed three centuries.

The second generation began exporting to the United States.

The third generation redefined its business policies and expansion strategy.

The fourth generation helped drive modernization.

Today, the company offers more than 300 sauces and condiments, distributed across more than 100 countries and regions.

Here, another possibility appears:

Inheritance itself is not the problem.

The real question is:

What are we inheriting?

The form?

Or the core?

If only the form is inherited, it can easily become replication.

If the core is inherited, while the form is allowed to change with the environment, it can become the foundation for the next judgment.

So Where Is the Real Boundary?

Replication itself is not wrong.

In fact, without replication, human civilization could not accumulate.

The questions are:

What should be replicated?

More importantly:

When should replication stop?

If the environment has not changed,

replication is efficiency.

If the environment begins to change,

replication must be subjected to judgment again.

And perhaps the most dangerous moment is not when an answer has already become wrong.

It is when:

The answer still produces some results.

Because as long as it still works,

the system has little immediate reason to rethink it.

Until the results begin to drift.

Until the original success rate begins to decline.

Until yesterday’s experience begins creating today’s costs.

Only then do we discover:

Perhaps the problem was never that the answer was wrong.

Perhaps:

The answer no longer belongs to the same environment.

A Fluid World Cannot Rely Only on Frozen Answers

A result can be preserved.

An answer can be inherited.

A process can be replicated.

But the environment in which that judgment was originally made

cannot be completely inherited along with it.

This may be one of the easiest things for any inheritance system to overlook.

Humans inherit knowledge.

Businesses inherit experience.

Algorithms inherit patterns of behavior.

AI inherits data and structures produced by previous models.

But the environment keeps moving.

And this creates a strange contradiction:

The better we become at preserving the past,

the easier it becomes to forget the conditions under which that past was valid.

So What Is Actually Worth Passing On?

Perhaps it is not the answer.

Nor a method that is supposed to remain correct forever.

Not even a fixed formula for success.

Perhaps it is:

The ability to generate an answer again.

Knowing what can be inherited directly.

Knowing what must be observed again.

Knowing when experience can reduce costs.

And knowing when experience has started to become a bias.

This may be where judgment truly matters.

Because judgment is not a rejection of the past.

It is allowing the past to enter the present,

without allowing the past to automatically decide the present.

We often think of learning as:

Knowing more.

But perhaps something more important is:

Knowing when we can no longer simply trust what we already know.

Humans need to learn.

Businesses need to learn.

AI needs to learn.

But if all learning does is improve our ability to reproduce answers from the past,

then learning itself can become another limitation.

Because a system can become increasingly precise at predicting the past,

without necessarily becoming better at facing the unknown.

So perhaps real freedom

has never been the absence of answers.

It is this:

When an answer appears,

we still have the ability to ask again—

“Does it still hold?”

If what we inherit is an answer,

while the environment never stops moving,

then who will produce the next answer?

Case 64 | 答案可以被繼承，判斷卻必須重新發生

我們曾在一篇推演中討論過：一個系統要維持自身，必須不斷重新配置自己與環境之間的資源關係。

但現實中的系統，也常常選擇另一條更省成本的路——繼承上一代已經驗證過的答案。

而當環境繼續移動時，這份原本用來降低成本的繼承，也可能逐漸變成限制下一步的邊界。

我們很少懷疑自己正在做的決定。

更多時候，我們只是找到了一個看起來合理的答案。

讀書、畢業、工作、買房、退休。

甚至連「65歲退休」這件事，都很少有人真正問過：

為什麼是65歲？

它看起來太正常了。

正常到我們甚至不需要知道它從哪裡來。

上一代就是這樣走的。

於是答案開始被傳承。

而當一個答案被傳承得足夠久，它就很容易從「曾經有效的方法」，變成「本來就應該如此的規則」。

問題也從這裡開始。

我們以為自己在判斷

1984年，久司道夫在一場談論 judgment 的演講中，把人的判斷區分成不同層次；他的另一場同年的演講，則談到「delusion」與人如何受到既有認知的影響。

但其中一個更值得我們今天重新看的問題，也許不是：

哪一種判斷才是最高層次？

而是：

我們以為自己在判斷，還是在使用一個已經形成的答案？

一個人可以非常聰明。

可以讀很多書。

可以掌握大量資訊。

甚至可以用非常漂亮的邏輯，證明自己的選擇是合理的。

但「合理」和「判斷」並不是同一件事。

合理，只代表它符合某一套已知的規則。

判斷，卻必須允許另一個可能：

當初讓這套規則成立的環境條件，現在是否仍然存在？

一個人工器官的故事

久司道夫在1984年的另一場演講中，講到一位年輕時的好友。

多年後，這位朋友已成為日本人工內臟產業協會的領導者。兩人重逢後，久司花了兩天了解人工器官技術與產業的發展。

他聽到的是一個已經非常完整的系統。

技術可以做到什麼。

哪些器官可以被替換。

成本是多少。

市場有沒有需求。

產業如何降低成本。

銀行、保險、醫療、藥品等不同力量如何一起形成新的市場。

甚至連未來十年的發展都可以被規劃。

這裡有一個很有意思的地方。

從這段故事中，我們可以看到兩個不同的觀察入口。

一個是：

我們已經有能力解決這個問題。

另一個則是：

我們為什麼會走到需要這個解決方案的地方？

這不一定是兩個人的立場對立，而是兩個可以被分開追問的問題。

一個從已經形成的解決方案出發。

一個回到問題形成的過程。

從結果取得答案

人類其實非常擅長做第一種事情。

我們會觀察過去。

找出成功與失敗。

整理原因。

形成規律。

最後得到一個答案。

它可以被寫進書裡。

可以變成教育。

可以變成制度。

可以變成企業流程。

也可以變成下一代人的常識。

這是一種非常強大的能力。

因為它把過去付出的試錯成本壓縮了。

昨天有人走過的路，今天的人不必再從頭走一次。

所以文明才能累積。

問題不在於這種能力。

問題在於：

結果一旦被整理成答案，它就開始脫離當初產生它的環境。

一個答案記得越久，它越容易失去自己的出生背景。

我們記得：

「這樣做有效。」

卻慢慢忘記：

「當時為什麼有效？」

還是從零開始？

另一條路完全不同。

不是先拿答案。

而是重新看現在。

觀察。

提出假設。

行動。

得到回饋。

再判斷。

再行動。

它比較慢。

也更昂貴。

因為每一次環境改變，都可能要求我們重新付出試錯成本。

但它有一個結果論很難提供的東西：

它仍然與當下的環境連在一起。

所以真正的兩極，可能從來不是：

有答案 vs 沒有答案。

而是：

從過去的結果取得答案。

與

在現在的過程中產生答案。

前者讓文明可以累積。

後者讓文明可以繼續適應。

而我們真正需要的，也許從來不是選其中一個。

而是知道：

什麼時候可以複製，什麼時候必須重新判斷。

當經驗開始取代判斷

一個成功的方法，最容易變成一個系統的限制。

因為成功會留下記憶。

記憶形成流程。

流程形成制度。

制度再被下一代學習。

於是：

成功 → 複製 → 標準化 → 傳承。

最後，原本需要判斷的事情，變成不需要判斷的事情。

這也是為什麼「經驗」既珍貴又危險。

經驗可以讓我們少走很多彎路。

但它也可能讓我們不再看路。

演算法讓這件事情變得更快

以前，人類把答案交給下一代。

現在，我們也把答案交給演算法。

你看過什麼。

停留多久。

點過什麼。

買過什麼。

什麼讓你留下。

什麼讓你離開。

系統從過去的行為中建立預測。

然後把它再次送回你的世界。

於是形成：

過去的行為
→ 模式
→ 預測
→ 回饋
→ 新的行為
→ 再次學習。

演算法甚至不需要告訴你：

「你應該相信什麼。」

它只需要越來越準確地知道：

你大概率會接受什麼。

於是自由並沒有消失。

選擇也仍然存在。

但有一個問題開始出現：

如果我們看到的選項，本身已經被過去的行為篩選過，我們還是在判斷，還是在選擇自己最熟悉的答案？

AI接過來了

當這種從過去建立模式的方式，不再只用來預測人的選擇，而是開始參與知識的生成與傳承時，問題就進入了另一個層次。

如果AI從人類留下來的資料中學習，這很自然。

人類本來就是靠傳承累積知識。

但如果下一代AI又大量使用上一代AI產生的內容呢？

那麼它學到的究竟是：

世界？

還是：

上一代對世界的描述？

兩者並不完全相同。

因為人類留下來的資料，本身就已經是一次選擇。

我們記錄某些事情。

忽略另外一些事情。

研究某些問題。

不研究另外一些問題。

成功的答案被留下。

失敗的答案被淘汰。

於是資料本身，就是歷史篩選後的結果。

而當機器再次從這些結果中學習，

再由下一代機器學習上一代機器的輸出，

真正的問題就不只是：

資料夠不夠多？

而是：

新的判斷從哪裡產生？

商業世界早就遇過同一個問題

這也是為什麼 Nike、Kodak 與李錦記放在一起，比單獨看任何一家更有意思。

它們不是三個「成功與失敗」的故事。

它們是三個不同的觀察窗口。

Nike：成功的答案會不會變成限制？

Nike目前仍把運動、產品創新、品牌與消費者關係放在核心策略中；但它自己的最新年報也明確指出，消費者偏好、運動與設計趨勢會改變，因此需要調整產品組合、渠道、新產品與市場策略。

所以真正值得問的不是：

Nike以前的方法對不對？

而是：

過去有效的方法，今天還應該被原樣複製多少？

Kodak：看見新答案，不等於完成判斷

Kodak其實早在1975年就發明了世界上第一部數位相機。它後來也推出數位相機、數位影像產品與相關系統。

所以「Kodak不知道數位時代來了」其實是一個過度簡化的故事。

真正值得看的問題反而是：

知道一個新答案存在，和讓整個系統重新判斷，是同一件事嗎？

不一定。

因為新技術本身只是答案的一部分。

真正困難的，是重新理解：

如果世界變了，我們原來是在解決什麼問題？

李錦記：傳承不一定等於複製

李錦記從1888年開始，跨越三個世紀。

第二代開始出口美國；第三代時重新制定經營政策與擴張策略；第四代加入後推動現代化。今天公司提供超過300種醬料與調味品，分銷至100多個國家和地區。

這裡反而出現另一種可能：

傳承本身不是問題。

真正的問題是：

傳承的是形式，還是核心？

如果傳承的只有形式，它就容易停留在複製。

如果傳承的是核心，並允許形式隨環境調整，它就可能成為下一次判斷的基礎。

所以，真正的分界線在哪裡？

複製本身沒有錯。

事實上，沒有複製，人類甚至無法累積文明。

問題在於：

複製什麼？

更重要的是：

什麼時候停止複製？

如果環境沒有變化，

複製就是效率。

如果環境開始變化，

複製就必須重新接受判斷。

而最危險的時刻，也許不是答案已經錯了。

而是：

答案還能得到一些結果。

因為只要它還能運作，

系統就沒有迫切理由重新思考。

直到結果開始逐漸偏離。

直到原本的成功率下降。

直到昨天的經驗開始產生今天的成本。

這時候我們才發現：

問題也許從來不是答案錯了。

而是：

答案已經不再屬於同一個環境。

流動的世界，不能只靠凝固的答案

一個結果可以被保存。

一個答案可以被傳承。

一個流程可以被複製。

但當初做出那個判斷時的環境，

不能完整地被一起傳承。

這也許是所有傳承系統最容易忽略的地方。

人類傳承知識。

企業傳承經驗。

演算法傳承行為模式。

AI傳承資料與模型產生的結構。

但環境一直在流動。

於是我們面對一個很奇怪的矛盾：

我們越擅長保存過去，就越容易忘記過去是在什麼條件下成立的。

那麼，什麼才是真正值得傳承的？

也許不是答案。

也不是某一套永遠正確的方法。

甚至不是某一種固定的成功模式。

而是：

重新產生答案的能力。

知道什麼可以直接繼承。

知道什麼必須重新觀察。

知道什麼時候經驗可以節省成本。

也知道什麼時候經驗已經開始成為偏見。

這可能才是「判斷」真正重要的地方。

因為判斷並不是拒絕過去。

它是允許過去進入現在，

但不讓過去自動決定現在。

我們常常把學習理解成：

知道更多。

但也許更重要的是：

知道什麼時候不能再直接相信已經知道的東西。

人類需要學習。

企業需要學習。

AI也需要學習。

但如果所有學習都只是提高「複製過去答案」的能力，

那麼學習本身也可能成為另一種限制。

因為一個系統可以越來越精準地預測過去，

卻不一定因此更有能力面對未知。

所以，也許真正的自由，

從來不是沒有答案。

而是：

當答案出現時，我們仍然有能力重新問一次——

「它現在還成立嗎？」

如果我們傳承的是答案，

而環境卻從未停止流動，

那麼下一個答案，

又是誰產生的？