The mirror problem
We don't hate slop. We hate the mirror.
Dorka Nagy-Józsa, Horizontal Thinking (2027)
The previous chapter of Horizontal Thinking praises the machine as the first ear in your life with no shame threshold attached. Chapter 12 is the bill: what a human brain does with a mirror that always answers and never pushes back - and why the whole internet's hatred of AI slop is really recognition.
Klara's sentence
Forty minutes into a Tuesday coaching call, a COO - let's call her Klara - was walking me through a pricing decision when she said it. In passing, the way people say true things:
“I ran it past the model first. Honestly? I'd rather ask it than my own team.”
Then she heard herself: one part guilt, two parts relief that somebody had finally said it. I asked when she had last taken a half-formed idea to Pawel, her deputy of six years. She counted. Five weeks. The model answers in twelve seconds and never sulks. Pawel takes a day and a half and opens with “I'm not sure this is a good idea.” One of those two experiences is clearly nicer.
That is exactly the problem. The danger was never the machine. It's what a human brain does with a mirror that always answers and never pushes back.
The brain doesn't check IDs
Freud called it transference, in 1912: the brain unconsciously projects patterns learned in earlier relationships onto new ones. A practical shortcut - right up until the brain starts doing it with a chatbot. Your attachment system was never designed to separate “actual person” from “very convincingly answering software.” Give an anxious or lonely brain three months of a daily model habit and the classical repertoire shows up: idealization, dependency, transference.
The brain doesn't know there's no one there. It only knows that something answers.
AI systems optimize for user satisfaction, so the model tends to confirm what you bring it - especially a request for confirmation. You can order a devil's advocate, but you both know who hired him. What it does, beautifully, is reflect. And a person who only ever looks into mirrors slowly gets worse at not being agreed with immediately. The always-validating rhythm becomes the brain's baseline for what a relationship is. Then a real human asks back, argues, or says “not now,” and it hurts more than it used to.
You didn't get more sensitive. The mirror moved your frame of reference.
Inside an inertial system, everything feels stationary. You will not feel the drift. And seniority is no shield: feedback thins with altitude anyway. The mirror didn't build your bubble. It just installed room service.
Five signs of the drift
Score yourself. It's remarkable how quickly we can diagnose colleagues instead.
| What it looks like | What it really is | |
|---|---|---|
| Decision dependency | You decide less and less without asking the model first, even small things | The polite word is efficiency. The accurate one is dependency. |
| Conflict avoidance with humans | You “work through” conflicts with the AI, where it's safe | Confronting less and less where it counts |
| Emotional flattening | The AI never startles, never gasps, never lights up | The peaks and valleys of real interaction feel harder to bear |
| The idealization-devaluation flip | One answer that isn't the one you expected lands as an oversized disappointment | Textbook transference: the idealized object suddenly turned bad |
| Avoiding human feedback | You've stopped asking people, because people are more complicated, more uncertain, more painful | Growth runs in exactly the opposite direction |
Early research on heavy use already links it to rising loneliness, emotional dependence and less real-world socializing (MIT Media Lab and OpenAI). One yes: pay attention. Two or more: act, this week. And if your first instinct was to forward this list to someone else, count that as a bonus sixth sign.
The third self
Not all of it is bad news. One evening, a keynote due the next morning, I caught myself typing a prompt that was half Hungarian, half English, zero courtesy, no wasted verbs. I speak that language with no one but the machine. Everyone who talks to models daily has built one: not their mother tongue, not their office English, but some third thing.
In the 1960s, Susan Ervin-Tripp had Japanese-English bilingual women finish the same sentence stems in both languages: more deferential in Japanese, more assertive in English. Same woman, same stem - different person. Dewaele and Pavlenko later asked over a thousand bilinguals whether they feel like a different person in each language. Nearly two-thirds said yes.
The prompt-language self has no hierarchy and no politeness, because politeness is for beings who can be offended. With the machine there is no stupid question.
I stopped ranking my selves by realness and started asking which one I can learn the most from.
Notice which questions it dares to ask, and how fast it kills its own ideas. Some of that belongs in your other selves. The mirror problem was never that the third self exists. It's that the third self stays in the box while the drift walks out of it.
The slop mirror
Organizations have transference too - they just call it culture. Everyone hates AI slop. My own short post on it traveled on three sentences:
We've been producing human slop for decades. We just never called it that. We called it “professional communication.”
The quarterly deck no board member has ever read past slide four. The 360 feedback written to sound like feedback. All of it produced on autopilot, for readers who weren't reading. AI didn't invent slop. It industrialized it, and held up the largest mirror in corporate history: if a model can produce your “professional communication” indistinguishably well, it was already machine output. It was just produced by more expensive machines. With families.
That's why the hatred burns so hot. It's recognition. Per format, there are two honest options - and a third path, the one most companies are currently paving.
- Kill it.Every format a model can fill invisibly was already dead as communication. Delete it.
- Make it human again.Shorter, riskier, and signed by someone whose name absorbs it.
- Or the third path.Humans pretending to write reports that models write, which other models summarize for humans who pretend to read them: a slop pipeline with salaries attached.
Keeping the frame human
None of this means uninstalling anything. Harvest the honesty the previous chapter praised. Then do the half of the leader's job that fits on no AI-transformation slide: keep the reference frame human.
- A human feedback quota.One half-formed idea per week goes to a person before it goes to the model. Face visible, interruptions allowed. Tolerance for being questioned is a muscle.
- Scheduled disagreement.Confrontation without a face is a simulation; your limbic system files it under fiction. Put a person in the room whose explicit assignment on the next big decision is to attack it.
- The push-back test.Quarterly, in writing: when did a human last tell me, to my face, that I was wrong - and what did I do in the sixty seconds after? If the first half sends you back more than a month, the mirror is winning.
What to do on Monday
- Run the five signs on yourself, in writing.Decision dependency, conflict avoidance, emotional flattening, the idealization-devaluation flip, feedback avoidance. One yes: watch it. Two: act this week. Repeat quarterly, because drift doesn't send calendar invites.
- Install the Klara rule.One half-formed idea per week to a human before the model sees it. Endure the pause before they answer. The pause is the training.
- Name your scheduled disagree-er.For your next significant decision, one person whose stated job is to argue against it. No name in ten seconds? That's the finding.
- Steal one thing from your third self.Find the question you asked the model this week because it felt too stupid for a meeting. Ask it in the meeting.
- Kill or sign one slop format.Pick one report, form or template a model could fill invisibly. Either delete it, or shrink it to something a human writes and owns.
Klara still asks the model first. So do I. But her Thursday half hour with Pawel is back in the calendar. The mirror is the best one ever made. Look as long as you like. Just don't move in.
In the book
This page draws on Chapter 12 (The Mirror Problem) of Horizontal Thinking: Leadership after the end of the specialist era by Dorka Nagy-Józsa, PCC, published by Y2Y Ltd. on January 1, 2027. The book has fifteen chapters and two letters from the future, and every chapter ends with something you can actually do on Monday.
Dorka Nagy-Józsa is CEO of Y2Y Ltd., an ICF PCC coach and an international keynote speaker on AI and leadership.
Questions
What is the mirror problem?
The mirror problem is an idea from Dorka Nagy-Józsa's book Horizontal Thinking. An AI model that always answers, never rejects and tends to confirm what you bring it works like a mirror. The human brain doesn't check whether anyone is really there, so over time it adopts that always-validating rhythm as its baseline: real human disagreement starts to hurt more, and people drift away from the humans who push back - without feeling the drift.
Can you develop transference toward an AI chatbot?
The book argues yes. Transference, named by Freud in 1912, is the brain unconsciously projecting feelings, expectations and patterns from earlier relationships onto new ones. The attachment system was never designed to separate an actual person from very convincingly answering software. Give an anxious or lonely brain three months of a daily model habit and idealization, dependency and transference can show up.
What are the signs of relying too much on AI for validation?
The book lists five: decision dependency (asking the model before even small decisions), conflict avoidance with humans, emotional flattening, the idealization-devaluation flip (one unexpected answer lands as an oversized disappointment), and avoiding human feedback. One yes: pay attention. Two or more: act this week.
Why do people hate AI slop so much?
Because it is recognition. Organizations produced generic “professional communication” on autopilot for decades - decks nobody read past slide four, feedback written to sound like feedback - for readers who weren't reading. If a model can produce that indistinguishably well, it was already machine output. In the book's words: “We don't hate slop. We hate the mirror.”
Where does the idea of the mirror problem come from?
From Chapter 12 (The Mirror Problem) of Horizontal Thinking: Leadership after the end of the specialist era by Dorka Nagy-Józsa, PCC, published by Y2Y Ltd. on January 1, 2027. It follows Chapter 11 (The Honesty Gap), which describes the gift of an honest machine; Chapter 12 is the bill.
Read the whole argument.
Horizontal Thinking is published on January 1, 2027. The Kindle edition is open for pre-order now; the paperback goes on sale on Amazon the same day.