Why We Hate AI But Can't Stop Using It

Last autumn, I came across a video on a history blogger's page. An old photograph from 1937 had been animated by AI—figures twisted like rubber, mouths twitching, eyes pointing in opposite directions. The top comment: "Please make it stop."

That scene brought back the rush I felt two years earlier, the first time I generated an image with Midjourney. Back then, "AI" seemed like a promise from the future, wrapped in a halo. Two years later, the halo is gone. Open any social feed and you'll trip over AI‑generated plastic—over‑smoothed faces, six‑fingered hands, web novels that repeat themselves mid‑paragraph.

But do we really hate AI?

The loudest voices do. Yet many more people—civil servants drafting official memos, social‑media operators on Xiaohongshu, web novelists grinding out 4,000 words a day—use it quietly, like any other tool. No fanfare, no guilt.

Xiaohongshu Copywriting Fell First

I knew a freelancer who made pocket money writing content‑farm posts for luxury brands. Before 2023, he'd get a product link, spend an afternoon polishing a few hundred words—selling points, emotional hooks, emoji placements, the closing "Go for it, girls." At 300 yuan per post, twenty orders a month kept him afloat.

In early 2024, Doubao (a Chinese AI assistant) began promoting its Xiaohongshu copywriting feature. He pasted the brief into the app, and three seconds later out came a draft. He tweaked a few phrases, added two images, delivered it. The client couldn't tell the difference.

"What does a customer actually need?" he said. "Three things: what this product is, why I should want it, and where to buy it. Everything else is packaging. And packaging just needs to look like someone is managing this account properly."

That's it. Xiaohongshu copy was never literature. It was a certificate of legitimacy—proof that the account was active and the product wasn't fly‑by‑night. Hand‑writing and machine‑writing served the same function; the hand‑written version just took longer.

Eventually the gig vanished—not because AI replaced him directly, but because a cheaper intern now does the same job in half an hour. He felt wronged, yet couldn't pinpoint why. The readers never noticed, and they didn't care.

Soulless Images, Measured Data

Critics slam companies for using AI‑generated promotional images. "Fake," "plastic," "obviously AI"—the outrage in comments is real.

But the data tells another story. An AI‑generated ad often gets similar click‑through rates to one drawn by a human. Sometimes, precisely because it's dramatic or absurd enough, it spreads further. Operators know this, so they keep using it. Managers know it too, so they keep demanding it.

"Managers don't care about quality; they just insist everyone use AI." That scene became common after 2025. It's less about technology and more about managerial anxiety—showing that "we're using the most advanced stuff."

That anxiety trickles down into workplace theatre. Copy teams spend hours tweaking prompts, competing over who can craft better "incantations" rather than who understands language better. Designers run AI outputs through twenty rounds of revision and often settle on the first version. Everyone follows orders, but nobody asks what actually works.

A friend in branding put it this way: "Before, when you paid 2,000 yuan for an illustrator, you tread lightly. After two revisions, you'd feel embarrassed to ask for more. Now, since it costs nothing, you can revise twenty times without anyone blinking. The final result is about the same, but the process is much more exhausting."

AI hasn't made work easier. It removed the cost barrier to endless revisions, trapping everyone in longer, more draining workflows.

Documents Nobody Reads

A civil‑service friend shared a scene.

A higher‑up directive arrives. He uses AI to generate a first draft. AI helps polish the wording. He uses AI to generate speaking points for a meeting. At the meeting, he rambles for five minutes based on those points. His secretary uses AI to summarise the minutes. They file it.

"Did any human being actually read that document from start to finish?" I asked.

He thought for a moment: "Probably not."

Absurd as it sounds, there's a premise: those documents were never read by anyone anyway. Before AI, they were written, revised, recited, and summarised by human hands—equally unread. The only difference was that people stayed up three nights to do it.

AI's role here isn't creating garbage. It's making an already empty process more efficient—so efficient that everyone now sees the emptiness. That's the real problem.

"At least before, I could fool myself that I was 'writing something'," he said. "Now I can't even fool myself."

Meanwhile, AI solves real problems elsewhere. A friend in tourism used AI to build a mini map app—tap a scenic spot and get official tips. Previously, that would have meant outsourcing for tens of thousands of yuan; now it took one afternoon. No one calls this AI application "plastic" or "perfunctory"—it works, and users actually use it.

What's the difference? The first case addresses the necessity of existence; the second solves an actual need.

The Web Novel Hierarchy

Among web‑novel author groups, there's an unspoken tier system:

Tier 1: Write a 3,000‑word outline and let AI polish it to 4,000. The author owns the plot and pace; AI handles description and transitions. Readers find it "smooth but forgettable"—that's acceptable.

Tier 2: Write a 1,000‑word outline and let AI inflate it to 4,000. Plot slows, repetition increases, but it stays within safety limits. Veteran readers sense the padding, but with auto‑scroll and audio features, they don't stop consuming.

Tier 3: Dump a synopsis into AI and generate whole chapters. These texts aren't really read. They exist because some readers just want background noise—commuting, chores, falling asleep—a steady stream of voice‑over with stock plots (CEO romance, cultivation, sweet fluff), the more generic the better.

"The market shifted," an editor told me. "Free reading plus audio—text is no longer something you read; it's a duration to be consumed."

Readers aren't stupid. They know AI text when they see it. Most don't mind—except when they've paid for it and got AI‑generated content. That's a consumer‑rights issue, not an aesthetic one.

Meanwhile, writers who still write by hand are quietly squeezed. Their edge was never "better than AI" – it was "more sincere than AI." But sincerity is hard to price in a system that sells reading by the hour.

The Invisible Line Between Generations

The loudest critics are creators, designers, writers—anyone who lives by craft.

The most tolerant—those who don't even debate "good or bad"—are people over thirty, with spending power and decision‑making authority.

I saw this at a dinner. A sixty‑year‑old relative opened his phone and showed me an AI‑generated "moving famous painting" he'd found on Douyin. He thought it was novel, even touching. "I never imagined this was possible. Now you can play with it on your phone."

He didn't notice the plastic brushstrokes. What impressed him was the contrast: a job that used to take days now takes three seconds to get a rough idea. That was enough.

That generation's aesthetic sense formed in a different era—hand‑drawn blueprints, typewriters, telegrams. For them, an AI image with six fingers is still "more comfortable" than going back and forth with a designer for three days.

In their value system, "having something" ranks above "having something good." That's not aesthetic decline—it's lived experience.

What Are We Really Complaining About?

Look back at the most criticised AI applications—animating old photos, writing web novels, churning out promotional images. They share one trait: they invaded domains once dominated by handcraft, yet they haven't produced anything better.

They've only done three things: faster, cheaper, more controllable.

And "faster, cheaper, more controllable" happens to be the full logic behind most business decisions. Aesthetics, sincerity, craftsmanship—these don't appear on any balance sheet.

So it's not AI itself that angers us. It's that AI brutally exposes a fact: much of what we called "important creative work" is, in the eyes of the market and the system, just "good enough to wing it."

Xiaohongshu copy is wing‑it. Promotional images are wing‑it. Official notices are wing‑it. Web‑novel daily updates—in the free tier—are also wing‑it.

When AI takes over those "wing‑it" tasks, the people who once took them seriously suddenly realise: what they cherished has no higher value. That's unsettling.

If AI Were an Intern

Think of AI as an intern. The picture becomes clear.

This intern is hard‑working, cheap, and never complains. It does most tasks to a passing grade—sixty out of a hundred. Occasionally it hits seventy‑five, sometimes flops to forty. But it never rests, never asks for a raise, and never questions "why are we doing this?"

The boss loves this intern. Senior employees hate him—especially those who've been making a living with eighty‑point skills while delivering sixty‑point work.

But the problem is, this intern isn't going away. It will only get better. When it can consistently score seventy‑five, the sixty‑point craftsmen will be squeezed out. When it hits eighty‑five, it'll be the turn of the seventy‑point crowd. And so on.

This isn't AI's fault. It's the law of the market: any work that can be standardised, repeated, and quantified will eventually be replaced by a cheaper mode of production. Handcraft won't disappear—it will retreat to its original place: a luxury for the few, and for things that are inherently non‑standard.

What Can We Still Keep?

I keep coming back to one question: what can AI never learn?

Not logic, not information synthesis, not linguistic fluency—it already does those better than most people.

It can't learn irrational impulse—the urge to spend three months painting something that will never sell, to write a novel for a reader who doesn't exist, to revise a single sentence twenty times knowing no one will notice.

These acts make no economic sense. Yet they mark the last boundary between "human" and "tool."

The utility‑ification of AI is irreversible. It will become increasingly invisible, increasingly taken for granted. One day we won't discuss "whether to use AI" – just as we don't discuss "whether to use electricity."

What will still be discussed then are the things AI cannot replace. They may be expensive, rare, accessible only to a few. But they will redefine what deserves to be called "creation."

Until then, we need to accept a brutal truth: not everything we create deserves that name. Much of what we thought of as "work" was, in the eyes of the market and the system, always just "good enough to wing it." AI merely made that secret public.

Seven in the evening, on my way home. A push notification lights up my phone. That history blogger has posted another AI‑animated old photo. The comments are still raging. I stare at the distorted faces and suddenly remember that afternoon two years ago, the first time I generated an image with AI—staring at the screen, convinced the world was about to change.

The world did change. But not into any of the futures we'd imagined. It simply made visible certain rules that had always operated in the shadows.