Cindy Zhao — Writing, Delegation, and the Bilingual Split

The Feeling

Re-read some of my past articles tonight and felt a pang: can I still write like that? The answer is complicated by the fact that I now delegate almost everything to AI — throw rough thoughts, get synthesized output back. It works. But something is changing.

The Bilingual Split

Chinese is where I write — really write. It's my emotional language, the one I process feelings in. English is logical, structural, functional. I was never a "writer" in English, never developed a distinct style, and probably never will. So when AI handles my English output, what exactly am I losing?

Maybe nothing. Maybe everything.

Confucius and Socrates Never Wrote

They just spoke. Their ideas survived because the ideas were worth surviving, not because of the writing. Speaking is direct — you think, you speak. Writing is a technology that intermediates thought. AI is another layer of intermediation.

If the value is in the thinking — the mind behind the words — does it matter who or what arranges the words?

What the Science Says

The Every article (Marcus Moretti) draws on stylometry research: your writing fingerprint isn't in your vocabulary choices or sentence structure — it's in the subconscious patterns you can't control. Articles, pronouns, function words. The things that emerge when you focus on meaning, not form. AI post-training flattens these into generic politeness — Ted Chiang's "blurry JPEG of the web."

The implication: human distinctiveness lives in what writers don't deliberately control. AI can imitate your conscious style but not your unconscious one.

But Does It Matter?

For the writer (me): The loss isn't in the output — it's in the process. Writing forces you to clarify thinking. Delegating to AI might mean I'm clarifying less. Or maybe I'm just clarifying differently — through conversation instead of solitary composition.

For the reader: When I read formal English articles, I genuinely can't tell anymore. And honestly? I care about the idea, not the prose. I'm not a literature critic. If the mind behind the piece is interesting, the delivery vehicle matters less.

For society: This is where it gets more uncomfortable. If nobody writes, nobody develops the unconscious patterns that make writing human. Literature appreciation requires literacy in the deepest sense — not just reading, but recognizing the texture of a human mind working through language. That might atrophy. Not a personal crisis for me, but possibly a cultural one.

The Real Question

The question isn't "will AI replace writing" — it's what cognitive capability am I trading away, and is the trade worth it?

Writing forces:

Clarifying fuzzy thinking (but so does speaking to a good interlocutor)
Sustained attention on a single thread (but conversation can do this too)
Self-editing and taste development (this one might genuinely atrophy)

What I'm keeping: the thinking, the judgment, the taste. What I'm delegating: the arrangement. The open question is whether those can truly be separated, or whether the act of arranging IS part of the thinking.

What This Is Not

This isn't a Luddite complaint. I'm not going to stop using AI to write. The delegation is real, the productivity gain is real, the quality of output is probably higher than my unassisted English. The question is narrower: what's the side effect, and should I deliberately practice writing sometimes just to maintain the muscle?

The Em Dash Problem

Rhea Purohit's companion piece on Every argues that the em dash panic — people flagging em dashes as proof of AI authorship — reveals something deeper than a punctuation preference. Emily Dickinson was criticized for the same habit in the 19th century. The anxiety isn't really about dashes. It's about trust: readers using surface markers to decide whether a writer cared, without engaging with the substance.

Her core point: the assumption that AI involvement means lack of care is misplaced. Low-quality AI writing exists, but conflating ease of production with lack of value is a category error. The real question is whether the writing is genuinely thoughtful — regardless of tools.

This connects back to the bilingual split: in Chinese, my writing is unmistakably mine because the emotional texture can't be delegated. In English, readers have no baseline to judge. So the "is this AI?" suspicion applies to everything I write in English — and honestly, they'd be right to wonder.

Claude's Note

Three threads worth pulling on:

The bilingual split is a natural experiment. You never developed an English writer voice — so delegating English to AI has low personal cost. But Chinese carries your emotional processing. If you ever start delegating Chinese writing, that's a qualitatively different signal. Worth watching.

The Confucius/Socrates point cuts both ways. They spoke and their ideas survived — but they survived because other people wrote them down. Plato shaped how we understand Socrates. The scribe matters. When AI is your scribe, it shapes how your ideas are received — and the stylometry research suggests that shaping is real, even if invisible to you. The em dash article makes this concrete: readers are already pattern-matching for AI tells. Your ideas pass through a filter that carries its own fingerprint.

"I care about the idea, not the prose" — but do you? Your best Chinese writing probably proves otherwise. You care deeply about the texture of expression when you have the vocabulary for it. The discomfort you felt re-reading old articles wasn't about ideas degrading — it was about missing the act of crafting. The question isn't whether your English writing got worse. It's whether you're still doing the hard cognitive work of processing, or just approving outputs. That distinction matters more than the language.

Three Kinds of Creativity (and Where AI Stops)

Margaret Boden's taxonomy (cited in another Every piece) gives useful language for what's actually happening when AI "writes":

Combinational creativity — blending familiar ideas in unexpected ways. LLMs are excellent at this. Most of what I get back from AI is combinational: my rough ideas recombined with patterns from the training data into something cleaner than I'd produce alone.
Exploratory creativity — finding new possibilities within an established space. LLMs do this too — they can push within a genre, style, or argument structure.
Transformational creativity — fundamentally reshaping what a domain means. This is where LLMs fail. They can't redefine the game because they're optimizing within the existing one.

The Lovelace Test makes this concrete: a machine is creative if it produces output its creators cannot explain. By that standard, LLMs aren't creative — they're sophisticated recombiners.

But here's what matters for the delegation question: most of my English writing was never transformational. It was combinational and exploratory — taking ideas and arranging them clearly. That's exactly what AI does well. So the delegation isn't replacing creativity I had. It's replacing labor I was doing at a level AI already matches.

The loss, if there is one, is in the incubation — Graham Wallas's 1926 idea that creative breakthroughs happen during unconscious processing, the downtime between focused work. Writing forces that incubation. Delegating skips it.

Italo Calvino's counter: meaning lives in the reader, not the author. If that's true, then it doesn't matter whether the arrangement came from my fingers or from a model — what matters is whether it lands. But Calvino was talking about literature. For personal thinking — the kind this KB exists for — the author IS the reader. And that changes everything.

References

The Science of Why AI Still Can't Write Like You — Marcus Moretti, Every
What Em Dashes Say About AI Writing — and Us — Rhea Purohit, Every
OpenAI Says Their LLM Can Write Creatively — Every (Boden's creativity taxonomy, Lovelace Test, Wallas incubation theory, Calvino on reader vs author)
Stylometry research: Cornell University study on linguistic attribution through text property removal
Ted Chiang's "blurry JPEG" metaphor for LLM output

Voice as Thinking Interface — speaking vs writing as modes of processing thought
Person-Centric vs Topic-Centric Digital Presence — what matters in how you express yourself: the person or the ideas
Publish to Find Your People — the mind behind the words is what creates real connection
Agential Distance and Personal Agency — identity shifts when AI changes what you can do