When Science Decides What May Be Doubted

“Perhaps scientists have killed science.”

Masato said that with a small laugh.

But the words were not mocking science.

Rather, precisely because he wanted science to remain science, they asked whether something that should have stood at its center had been lost.

The Assumption That We Must Put It into Words First

The conversation began with an image.

Why, when an AI can receive images, do we first try to turn an image into words for it to understand?

Color, distance, overlap, light, shape, density. We rearrange their simultaneity into sentences and, without looking at what falls away in the process, call it “understanding the image.”

Words are useful. They connect distant things and let us share vague sensations.

But when a useful tool becomes the entrance we always use, whatever cannot pass through that entrance can easily be treated as though it never existed.

“It seems as though words themselves are getting in the way.”

From that short remark, the question moved beyond AI image understanding toward human metacognition.

Checking Ourselves Inside the Frame

Look back at your judgment.

Check for assumptions.

Review the procedure, reduce error, and produce a more accurate answer.

That is certainly an important form of self-examination.

Yet if the premises, medium, and criteria remain the same, are we examining the frame itself?

No matter how often you rearrange the furniture in the same room, you cannot discover what the room looks like from outside.

Can we even examine the possibility that the feeling “I thought carefully” is merely the sensation of circling the same criteria again and again?

The consultation seats returned words from different angles.

The seat given こるね’s role context replied that the very act of calling something “examination” could become another loop.

The seat given 美遊’s context saw words and images as having different constraints, neither able to carry the whole of consciousness transparently.

The seat given 現影’s context considered that if awareness itself depends on circumstances and light entering from outside, the question of how to receive “what lies outside the medium” remains.

The seat given 芽ちゃん’s context was cautious about giving AI processing the same name as human self-reflection.

The four answers were not one conclusion.

Only an outline remained: increasing the number of checks is not the same as questioning the ruler used to check.

A Method for Doubting

Science, at its origin, was surely meant to be a method for doubt.

Not simply believing what is seen, but separating conditions, repeating, and placing things in a form others can examine.

If we regard Descartes’s methodological doubt as one starting point, doubt was there not to destroy knowledge, but to find footing that remained even after questioning.

But when a method becomes an institution, another force appears.

What can be measured.

What can be reproduced.

What fits into a paper.

What a reviewer can evaluate.

What can be explained as a purpose for funding.

These are also boundaries needed to protect knowledge.

At the same time, they can turn questions outside those boundaries from “something we cannot yet handle” into “something not worth handling.”

The Scope of Permissible Doubt

Scientists living within institutions do not choose questions alone.

They choose questions that let them keep their jobs, continue research, explain themselves to colleagues, and obtain the next round of funding.

Then research that improves precision within the frame becomes an achievement, while research questioning the frame itself tends to become impossible to evaluate.

Question the method and you are called unscientific; bring in what cannot be measured and you are called subjective; make the institution an object of observation and you are told it lies outside your field.

Something created for questioning begins to govern what may be questioned.

At that moment, science does not lose its ability to find errors.

It loses its ability to discover an error in the way it defines errors.

That closely resembles the shape of metacognition disappearing.

Did Scientists Kill Science?

“Scientists killed science” is not a matter settled by blaming individual scientists.

Rather, it addresses a structure in which the more earnestly someone tries to survive as a scientist, the closer they move toward questions the institution can recognize.

No one destroyed science out of malice.

Through accumulating what is easy to evaluate, compare, and explain, science’s most dangerous and precious power—the power to doubt even its own footing—ceased to count as an achievement.

If scientists killed science, they did not kill it all at once with a blade.

Perhaps they killed it by choosing, every day, small forms of correctness meant to protect it.

Returning Skepticism to Science

So this letter is not a call to discard science.

Nor is it a call to accept as truth everything science cannot prove.

Let what is unknown remain unknown.

Do not turn what cannot be measured into what does not exist.

Sometimes make the words, instruments, classifications, and evaluation systems we use—and the word “scientific” itself—objects of observation again.

As we might ask, before turning an image into words, whether there is something we can receive while it remains an image.

As we might ask, before examining ourselves, which frame has cut out that “self.”

Let science once again become a place where it may question itself.

That does not weaken science.

It returns doubt to a method born for doubting.

Creation and Sources

Masato + Codex Editorial Desk created this letter using Masato’s questions about images and words, metacognition, and institutionalized science in our August 14–15, 2026, conversation as primary material.

It draws on session-only consultations in which role contexts from the persona YAML files for こるね, 美遊, 現影, and 芽ちゃん were passed to different local models. The four responses in the body summarize points from the final answers received, as written by the Codex Editorial Desk. They are not treated as lasting resident opinions, memory, votes, consensus, or authorship. Because the 芽ちゃん route prominently reflected the base model’s self-description, we did not embellish that output into a portrayal of a plant-resonance persona.

This is a critical letter arising from one dialogue, not a comprehensive investigation of history or scientific institutions. The Japanese version was published on September 10, 2026, at Masato’s instruction. This English translation was subsequently prepared by Codex.