Ukiyo-e Women Facing a Digital Wave
I purchased this version of The Makioka Sisters: Sasameyuki on Amazon as something to read before an upcoming trip to Japan. What I discovered was that I had unknowingly bought what reviewers identify as an AI-generated, heavily condensed retelling of The Makioka Sisters, rather than Tanizaki’s novel in its established English translation. This post is about: “How do you know something is wrong before you know what is wrong?”
There were many warning flags that in retrospect are obvious that I’m listing for the inevitable next time:
In summary, the seven issues above fall into three categories:
I’ve pulled some passages from the AI-generated version and highlighted the words that were trigger words for me, that is I read them and thought, this sounds AI-generated.
Example 1
She had been friends with all three Schultz children, but her friendship with Rosemarie was the central one — the intimacy of girls the same age who have found in each other the specific quality of being understood — and Rosemarie would be gone for some months yet.
Example 2
January of 1939. The new year arrived with the specific quality of new years in Ashiya: the cold clean air off the mountain, the harbour visible on clear days in the particular blue of January, the neighbourhood with its shutters up for the holiday and the streets quieter than usual, the small domestic ceremonies of the season observed with the attention the Makioka women brought to all ceremonies, which was to say the attention of people for whom the ceremony is the point, not the pretext.
Example 3
The Ashiya house in autumn had the specific quality of the season in that part of the world: the mountain visible above the residential terraces, its vegetation going from the summer green to the autumnal layering of colour that the Rokko range performs in October, a gradient of russet and gold visible from the upper windows of the house on clear days.
Example 4
Spring of 1940. Taeko had been in the state of managed semi-estrangement from the household for several months when Oharu arrived in the kitchen one morning with the specific quality of expression — the quality that the Makioka household had learned to read in Oharu over three years — that meant she knew something she had been instructed not to tell.
Example 5
The way Taeko’s attention, when Itakura was in the room, had a specific quality that was not the way she attended to anyone else. The way he listened when she spoke — not the polite attention of courtesy, but the full, exclusive attention of someone for whom what the person is saying is the most interesting thing in the room.
Example 6
Outside, beside the track, a line of cherry trees was in full flower — white and pale pink, the colour of something seen at the edge of a good thing, the specific Kansai spring cherry that blooms briefly and falls briefly and is, while it is, precisely what it is and nothing else.
Example 7
The word fell into the sitting room with the specific weight of something that changes the shape of every other thing in the room.
Example 8
What she found instead was someone with an unexpectedly light quality — not lightness of character, but of manner, a social ease that was genuinely earned rather than performed, the ease of a man who has been trained in medicine and has learned from that training to put people at their comfort.
Example 9
The Schultz family had been, for the past two years, the specific variety of gift that is produced by good neighbouring — not intimacy, exactly, but proximity of the comfortable, undemanding kind that gives daily life its texture: the children’s voices through the fence, the shared acknowledgment of weather, the particular pleasure of being known by someone whose knowledge of you is based on observation rather than history.
Example 10
The illness resolved, eventually, in the way these things resolve: gradually, incompletely, then one morning more completely than the morning before, and then, by degrees, gone — or not gone exactly, but retreated, become manageable, become the kind of thing that Etsuko would carry quietly and that her family would carry quietly with her.
Example 11
She went in the same kimono she had come in, on the Tuesday morning train, with a composure that had about it, this time, a quality that was very slightly different from its usual quality — not more settled, exactly, not more resigned, but something that Sachiko could only describe as: she knows we tried.
Example 12
The Sunday dinner in Kobe — at a sukiyaki restaurant in the Motomachi district, not the formal grandeur of the Oriental Hotel but the warmer register of a neighbourhood establishment that managed its private rooms with comfort rather than ceremony...
Example 13
He had about him the quality of a person who is interested in the world in general, which is different from — and considerably rarer than — the quality of a person who performs interest.
Example 14
Tatsuo, in managing the aftermath, had handled the correction with the efficiency of a man who confuses efficiency with wisdom, which is to say badly: quietly, without consulting the sisters, in a way that produced a small correction in the newspaper that was read by considerably fewer people than had read the original item, and that satisfied nobody.
Example 15
[Scene setting material] The novel is narrated, for the most part, through Sachiko’s consciousness — which is to say, through the consciousness of a woman who is warm, perceptive, somewhat anxious, and possessed of the particular limitation that afflicts people who love too many people simultaneously: the inability to be entirely honest with any of them.
Let me be clear, I used AI to help me analyze these passages and help me understand what I am seeing. However, I did not let AI write this.
I’m digging into this because I wanted to understand what I experienced. What I began to realize was that none of these devices is particularly unusual. Human writers use all of them. What bothered me was the frequency with which this text used these devices, until what might have been normal writing moves to feel unnatural and annoying. The same rhetorical machinery of the Duke Loops turned up masquerading as Tanizaki, because they both (in my opinion) heavily lean on AI assistance.
Let’s take the phrases and styles often seen in the passages above.
Pattern: “specific X”
What it does: Promises precision
Why it starts sounding AI-generated: “Specific quality” signals that a finely observed distinction is coming, but the phrase itself is vague. It announces precision before actually supplying it. Just give it.
Pattern: “not X, but Y”, “not X exactly, but ...”
What it does: Creates contrast and refinement.
Why it starts sounding AI-generated: It gives the impression that the writer (!) is correcting an inadequate first thought and arriving at something subtler. Repeated often, it becomes annoying. It’s not a surprising rhetoric device, but the frequency is.
Pattern: “which is to say”
What it does: Interprets the previous clause.
Why it starts sounding AI-generated: The narrator explains its own prose instead of allowing the reader to infer it. It leaves no room for slow development; it cashes the check immediately.
Pattern: “particular / specific / kind / quality / way”
What it does: Categorizes an experience.
Why it starts sounding AI-generated: Creates pseudo-taxonomic precision: everything seems to belong to a finely distinguished category. Creates false precision.
Pattern: “landscape ‘performs,’ year ‘arrives’”
What it does: Adds literary elevation
Why it starts sounding AI-generated: Ordinary description gets an automatically poetic verb. Everything object is an actor. Fine occasionally; conspicuous when it becomes the default setting to emulate “fancy” writing.
Pattern: “balanced clauses and em dash usage"
What it does: Provides elegant rhythm.
Why it starts sounding AI-generated: The sentences are almost too continuously composed. Different situations keep producing the same rhetorical architecture.
Pattern: witty aphorisms, often psychological that normal people don’t just churn out
What it does: Supplies apparent insight.
Why it starts sounding AI-generated: The prose reveals universal truth about “people who…” without really building up to it. Human novelists do something similar, but not at the frequency I was finding here in AI-generated text and not the need to always extract meaning.
In the passages selected we see a lot of: “You might think X. But actually Y” and “Not X exactly, but Y.” This does several things that are not all desirable:
This is interesting because OpenAI itself is explicitly identifying the tendency we hear and offering a prompt to suppress it.
The article Why Does AI Keep Saying "It's Not X, It's Y"? says “In classical rhetoric, this construction is called antithesis, or negative-positive parallelism. It works by first negating a familiar assumption, then replacing it with something supposedly more expansive or profound.” The article suggests that AI picked up the “not X, but Y” habit partly from human writing and then had it reinforced by human feedback. A contrast can make an answer sound nuanced and thoughtful, qualities human raters may reward. The result may be a useful rhetorical device getting rewarded often enough that the model starts reaching for it far too frequently.
The reinforced learning explanation interests me because it points back to human involvement. The “not X, but Y” habit is not something AI invented on its own. Models first learn the construction from human writing and are then further tuned using human judgments about which answers seem better. The article argues that if polished, contrastive answers tend to sound nuanced or thoughtful to human raters, that preference can be reinforced. So perhaps this isn’t an AI tic at all, but a human rhetorical preference amplified until it becomes annoying.
One thing that has also bothered me about AI-assistance is the number of clever aphorisms and what I would say are clean resolutions. I always feel a bit annoyed and awed at the ease at which they are churned out. It’s unnatural in human writing (at least mine). They go something like this:
Used sparingly, these patterns are interesting. Used a lot, you notice the mechanism more than the writing. And I notice thinking of these rhythms as formulas applied, like we are mathematically constructing prose. And worse, it becomes exhausting.
Warning Signs
There were many warning flags that in retrospect are obvious that I’m listing for the inevitable next time:
- The book was very cheap. Cheaper than other novels I’ve purchased on Kindle in the past
- The AI-generated version kept popping to the top of the Kindle results. Why? I should have asked but didn’t. There are legitimate English Kindle editions, but every time I followed a link to one, Amazon told me: “This edition of this title is not available for purchase in your country.” I’m in Italy. I wanted a Kindle version. This one was available. I got discouraged looking further and, apparently, decided that was enough due diligence. It wasn’t.
- The AI-generated version was surprisingly short at 160 pages. I noted it but didn’t think to check until later.
- The AI-generated version contained introductions to sections that set the scene and describe words used in the summary. Not that I didn’t appreciate the scene setting, but I found it odd. Again, I only noted it. I would later see that these were included because the text was so summarized and condensed that you needed these introductions to understand what was going on.
- The events unfolding in the lives of the sisters were very abbreviated and cliff-notes like. I thought: was Tanizaki's prose really this spare? No, it wasn’t. The thought entered my mind, but I only noted it. What an idiot I was.
- Certain phrasing in the AI-generated version started leaping out at me. Phrasing that was exactly the same as phrasing in my Duke Loop stories. See The Loops of August and Velvet Squabs and Flat-Arc Heroines. I initially thought “huh, what a coincidence!?” What an idiot. See the examples section below.
- There were discrepancies in the plot that caught my attention and made me go back to make sure I had read correctly. In one scene, the second sister is going somewhere in a taxi and upon returning is on a train. Possible, yes, but odd. Or in another scene, a woman’s family (brother and mother) were said to remain in an apartment but then a paragraph later that apartment passed to someone else. These may not be discontinuities in the real work, but in the AI-generated version, they were.
In summary, the seven issues above fall into three categories:
- Commercial/bibliographic clues: suspiciously cheap; prominently surfaced on Kindle; radically short.
- Narrative clues: explanatory section introductions; summary-like treatment of events; continuity problems.
- Linguistic clues: “specific quality,” negative contrast, “which is to say,” literary anthropomorphism, abstract aphorisms.
I have effectively trained my ear by reading/listening a lot of generated Duke Loop material. So, when the same rhetorical machinery turned up masquerading as Tanizaki, I felt uncomfortable but couldn’t name it because it kept thinking it was Tanizaki.
This experience with this AI-generated work recalls some point from the article: AI Slop—How Every Media Revolution Breeds Rubbish and Art | Scientific American I’ll pick three points from the article.
(1) Slop refers to low‑quality, mass‑produced content, especially AI‑generated material that is churned out without care or audience demand. References to AI “slop” go back at least to 2022, but the term was popularized in 2024 by the technologist deepfates and subsequently picked up by developer Simon Willison.
This experience with this AI-generated work recalls some point from the article: AI Slop—How Every Media Revolution Breeds Rubbish and Art | Scientific American I’ll pick three points from the article.
(1) Slop refers to low‑quality, mass‑produced content, especially AI‑generated material that is churned out without care or audience demand. References to AI “slop” go back at least to 2022, but the term was popularized in 2024 by the technologist deepfates and subsequently picked up by developer Simon Willison.
- This AI-generated version of The Makioka Sisters has been churned out, and I would say without care. Several reviews of the Kindle book claim this.
- This AI-generated version of The Makioka Sisters, after some thought, does not add value. You would do better with well-written cliff notes or an essay about the book.
- Cognitive overload for me in this case was: something is wrong, I can’t figure it out and it took several hours of research to reach this conclusion. So, an overload and waste of time ultimately.
Example Passages the AI-generated Version
Example 1
She had been friends with all three Schultz children, but her friendship with Rosemarie was the central one — the intimacy of girls the same age who have found in each other the specific quality of being understood — and Rosemarie would be gone for some months yet.
Example 2
January of 1939. The new year arrived with the specific quality of new years in Ashiya: the cold clean air off the mountain, the harbour visible on clear days in the particular blue of January, the neighbourhood with its shutters up for the holiday and the streets quieter than usual, the small domestic ceremonies of the season observed with the attention the Makioka women brought to all ceremonies, which was to say the attention of people for whom the ceremony is the point, not the pretext.
Example 3
The Ashiya house in autumn had the specific quality of the season in that part of the world: the mountain visible above the residential terraces, its vegetation going from the summer green to the autumnal layering of colour that the Rokko range performs in October, a gradient of russet and gold visible from the upper windows of the house on clear days.
Example 4
Spring of 1940. Taeko had been in the state of managed semi-estrangement from the household for several months when Oharu arrived in the kitchen one morning with the specific quality of expression — the quality that the Makioka household had learned to read in Oharu over three years — that meant she knew something she had been instructed not to tell.
Example 5
The way Taeko’s attention, when Itakura was in the room, had a specific quality that was not the way she attended to anyone else. The way he listened when she spoke — not the polite attention of courtesy, but the full, exclusive attention of someone for whom what the person is saying is the most interesting thing in the room.
Example 6
Outside, beside the track, a line of cherry trees was in full flower — white and pale pink, the colour of something seen at the edge of a good thing, the specific Kansai spring cherry that blooms briefly and falls briefly and is, while it is, precisely what it is and nothing else.
Example 7
The word fell into the sitting room with the specific weight of something that changes the shape of every other thing in the room.
Example 8
What she found instead was someone with an unexpectedly light quality — not lightness of character, but of manner, a social ease that was genuinely earned rather than performed, the ease of a man who has been trained in medicine and has learned from that training to put people at their comfort.
Example 9
The Schultz family had been, for the past two years, the specific variety of gift that is produced by good neighbouring — not intimacy, exactly, but proximity of the comfortable, undemanding kind that gives daily life its texture: the children’s voices through the fence, the shared acknowledgment of weather, the particular pleasure of being known by someone whose knowledge of you is based on observation rather than history.
Example 10
The illness resolved, eventually, in the way these things resolve: gradually, incompletely, then one morning more completely than the morning before, and then, by degrees, gone — or not gone exactly, but retreated, become manageable, become the kind of thing that Etsuko would carry quietly and that her family would carry quietly with her.
Example 11
She went in the same kimono she had come in, on the Tuesday morning train, with a composure that had about it, this time, a quality that was very slightly different from its usual quality — not more settled, exactly, not more resigned, but something that Sachiko could only describe as: she knows we tried.
Example 12
The Sunday dinner in Kobe — at a sukiyaki restaurant in the Motomachi district, not the formal grandeur of the Oriental Hotel but the warmer register of a neighbourhood establishment that managed its private rooms with comfort rather than ceremony...
Example 13
He had about him the quality of a person who is interested in the world in general, which is different from — and considerably rarer than — the quality of a person who performs interest.
Example 14
Tatsuo, in managing the aftermath, had handled the correction with the efficiency of a man who confuses efficiency with wisdom, which is to say badly: quietly, without consulting the sisters, in a way that produced a small correction in the newspaper that was read by considerably fewer people than had read the original item, and that satisfied nobody.
Example 15
[Scene setting material] The novel is narrated, for the most part, through Sachiko’s consciousness — which is to say, through the consciousness of a woman who is warm, perceptive, somewhat anxious, and possessed of the particular limitation that afflicts people who love too many people simultaneously: the inability to be entirely honest with any of them.
Patterns
Let me be clear, I used AI to help me analyze these passages and help me understand what I am seeing. However, I did not let AI write this.
I’m digging into this because I wanted to understand what I experienced. What I began to realize was that none of these devices is particularly unusual. Human writers use all of them. What bothered me was the frequency with which this text used these devices, until what might have been normal writing moves to feel unnatural and annoying. The same rhetorical machinery of the Duke Loops turned up masquerading as Tanizaki, because they both (in my opinion) heavily lean on AI assistance.
Let’s take the phrases and styles often seen in the passages above.
Pattern: “specific X”
What it does: Promises precision
Why it starts sounding AI-generated: “Specific quality” signals that a finely observed distinction is coming, but the phrase itself is vague. It announces precision before actually supplying it. Just give it.
Pattern: “not X, but Y”, “not X exactly, but ...”
What it does: Creates contrast and refinement.
Why it starts sounding AI-generated: It gives the impression that the writer (!) is correcting an inadequate first thought and arriving at something subtler. Repeated often, it becomes annoying. It’s not a surprising rhetoric device, but the frequency is.
Pattern: “which is to say”
What it does: Interprets the previous clause.
Why it starts sounding AI-generated: The narrator explains its own prose instead of allowing the reader to infer it. It leaves no room for slow development; it cashes the check immediately.
Pattern: “particular / specific / kind / quality / way”
What it does: Categorizes an experience.
Why it starts sounding AI-generated: Creates pseudo-taxonomic precision: everything seems to belong to a finely distinguished category. Creates false precision.
Pattern: “landscape ‘performs,’ year ‘arrives’”
What it does: Adds literary elevation
Why it starts sounding AI-generated: Ordinary description gets an automatically poetic verb. Everything object is an actor. Fine occasionally; conspicuous when it becomes the default setting to emulate “fancy” writing.
Pattern: “balanced clauses and em dash usage"
What it does: Provides elegant rhythm.
Why it starts sounding AI-generated: The sentences are almost too continuously composed. Different situations keep producing the same rhetorical architecture.
Pattern: witty aphorisms, often psychological that normal people don’t just churn out
What it does: Supplies apparent insight.
Why it starts sounding AI-generated: The prose reveals universal truth about “people who…” without really building up to it. Human novelists do something similar, but not at the frequency I was finding here in AI-generated text and not the need to always extract meaning.
Contrast
In the passages selected we see a lot of: “You might think X. But actually Y” and “Not X exactly, but Y.” This does several things that are not all desirable:
- Produces nuance in the sentence.
- Gives the sentence movement. There is a little before-and-after.
- Creates cognitive friction if overused.
- Comes across as slightly artificial about being repeatedly told what not to think in order to understand what to think. Instead of simply showing you Y, the sentence constructs an unnecessary X, so it can provide a refinement.
This is interesting because OpenAI itself is explicitly identifying the tendency we hear and offering a prompt to suppress it.
The article Why Does AI Keep Saying "It's Not X, It's Y"? says “In classical rhetoric, this construction is called antithesis, or negative-positive parallelism. It works by first negating a familiar assumption, then replacing it with something supposedly more expansive or profound.” The article suggests that AI picked up the “not X, but Y” habit partly from human writing and then had it reinforced by human feedback. A contrast can make an answer sound nuanced and thoughtful, qualities human raters may reward. The result may be a useful rhetorical device getting rewarded often enough that the model starts reaching for it far too frequently.
The reinforced learning explanation interests me because it points back to human involvement. The “not X, but Y” habit is not something AI invented on its own. Models first learn the construction from human writing and are then further tuned using human judgments about which answers seem better. The article argues that if polished, contrastive answers tend to sound nuanced or thoughtful to human raters, that preference can be reinforced. So perhaps this isn’t an AI tic at all, but a human rhetorical preference amplified until it becomes annoying.
Unnatural Rhythm
One thing that has also bothered me about AI-assistance is the number of clever aphorisms and what I would say are clean resolutions. I always feel a bit annoyed and awed at the ease at which they are churned out. It’s unnatural in human writing (at least mine). They go something like this:
- observation -> qualification -> elegant conclusion
- observation -> contrast -> elegant conclusion
- description -> generalization -> elegant conclusion
- event -> interpretation -> insight or elegant conclusion
- announce precision -> qualify -> observation
So What?
I can’t say I didn’t learn a something from the AI-generated version of The Makioka Sisters, but I feel I ultimately wasted time and didn’t get the real voice of the author (and translator).
Identifying and trying to name the things that bugged me about the writing, I can better help myself when using AI assistance. I have, for sure, overused it in this very blog and am not proud of it. I’m in the process of trying to learn how to use AI assistance without overusing it.
I’ve often think of “keeping my voice” in the age of AI assistants, making sure the AI voice doesn’t take over. I’ve tried to understand in this post, and in a small way, what that AI voice is. It’s a voice that has some of the characteristics given above, like lots of uses of contrasts and intense formulaic writing. These, plus many more not mentioned, seem to lead to two contradictory outcomes. AI-generated prose:
- Closes down (or reaches) meaning almost too quickly/concisely/conveniently.
- While over-explaining some things to the point of being repetitive.
It’s not that AI prose is bad at any one given sentence, though some sentences might raise an eyebrow. It’s that the whole taken together is where you feel it. It is in all the sentences taken together that a reader senses something missing, and the mechanism.
Perhaps what I have raised here is just a snapshot in time and will go away with newer AI models. Or perhaps it is baked in. As AI might say, not what we see today, but something different— which is to say time will tell. (I’m just not as good at it!)










.jpg)





































