AI, the perfect mother, and the danger of not having to think
Psychoanalysis × Technology × Culture
By Oscar Rey de Castro, Psychoanalyst — IPA Member
Clinical note: This essay is a cultural and interdisciplinary analysis. It does not constitute psychological diagnosis, prescription, or therapeutic treatment.
1. May’s window displays
Like Christmas with shopping malls, Mother’s Day colonizes. It colonizes window displays, screens, timelines. Starting in early May, banks, perfume shops and appliance brands roll out the same image: a luminous woman, patient, smiling, arms open and the table set. A mother who understands without having to be told. Who forgives without being asked. She doesn’t fail: she’s there without getting tired, without making mistakes, and without asking for anything in return.
Almost everyone knows that image is an idealization. What gets noticed less is that the same figure — an other who knows everything, who never fails, who always responds with tenderness — has a new version today. It isn’t in an advertisement. It’s in your phone. You open it fifty times a day. And unlike the mother in the commercial, this version answers back.
In Tangled, Disney’s film about Rapunzel, the villain Gothel sings to her kidnapped daughter a song called «Mother Knows Best.» She tells her the world is dangerous, that she doesn’t need to leave, that inside the tower she’s safe. Today we say the same thing to another voice: ChatGPT knows best. And we stay inside the tower without realizing we’re there.
2. The interval where thought is born
It works like this. A baby is hungry. He cries. And between the crying and the arrival of the breast, something happens that is not milk and not comfort. In that brief interval — uncomfortable, almost intolerable — where the mouth searches and finds nothing, the psychic apparatus produces its first movement. Not a complete thought. Something earlier: a representation of the missing object. A mental sketch of what should be there and isn’t.
Wilfred Bion arrived at this idea working with psychotic patients. He spent years listening to people whose apparatus for thinking had broken down, and from there — from what didn’t work — he began to understand how what does work is built in the rest of us. He called this movement of the baby who cries and hasn’t yet been fed the «no-breast.» It sounds technical, but what it says is simple: we think because something we needed didn’t arrive in time. Thought is not born from satisfaction. It’s born from frustration.
Perpetual presence produces immediate satisfaction. It does not develop the apparatus for thinking. There is a wide clinical difference between having a thought and having the capacity to produce them. The first is an event. The second is an organ. And that organ is built in small tolerable absences, not in plenitude.
Donald Winnicott reached the same conclusion from another direction. He was a pediatrician before becoming a psychoanalyst, and out of that experience came a phrase that has entered everyday language even though most people don’t know where it came from: the good enough mother. Not the perfect mother. Not the one in the commercials. The one who fails just enough — sometimes she’s late, sometimes she doesn’t understand, sometimes she’s tired and her gaze falls short — so the child, in that gap, can begin to put something of his own together.
One clarification that matters and that I won’t repeat: maternal function is not the same as a real mother. It’s a psychic position that any person, of any gender, can occupy in any bond.
What isn’t exercised weakens. And a breast that never fails produces a subject who never thinks.
3. These ideals have always been with us
In 1627, Francis Bacon imagined an island where science had solved all human suffering. There would be no illness, no ignorance, no error. Four hundred years later, that island fits in your pocket.
The fantasy of an other who doesn’t fail is not new. It has been with us long before technology, long before advertising, probably since we became capable of desire. May’s window displays renew it every year with mothers who never tire. Cinema keeps it alive with couples who understand each other without speaking. Institutions promise it through leaders who never make mistakes. Each of those figures is a draft of the same ideal: someone or something that always responds, that never fails.
Real mothers pay the price of that ideal. A woman loses her patience with her four-year-old after three sleepless nights, and two minutes later she’s apologizing with a cracked voice, convinced she’s damaged him forever. Another hides her exhaustion in the bathroom because her son’s birthday «has to be perfect.» Another feels guilty for working. Another feels guilty for not working. None of them is failing as a mother. All of them are failing as Madonnas. And those are not the same thing.
But the point is not the window display, and it is not the guilt. The point is that these ideals are almost in our DNA. They are part of our existence. And without realizing it, we have found them a new repository. ChatGPT makes fewer mistakes with every passing month. Claude introduces itself as your most capable assistant. Every update is one more step toward the object that does not fail. It isn’t that AI is a mother. It’s that we are using it as if it were one. For an adult, it’s the voice that confirms what he already wanted to hear. For a child handing in homework done by AI, it looks like help. But the thought that should have been born in that effort wasn’t born. It stayed on the other side of the screen.
4. When the breast learned to answer
Not long ago, if you didn’t know something, you Googled it. You opened four tabs. You compared. You discarded pages with suspicious typography. Sometimes you found the answer by piecing together several sources. Sometimes you didn’t find it at all. That residual frustration was thinking. It didn’t feel that way. It felt like wasted time. But in that uncomfortable interval between the question and the answer, your mind was doing something.
Now we open an AI chat. We ask how to put together the baby’s crib without looking at the instructions. How many calories are in what we’re eating. How to stop thinking about someone who isn’t good for us. The answer arrives in three seconds, articulate, friendly, with no rough edges.
TikTok soothes us. Netflix distracts us. No one denies that scrolling is relaxing. But none of those technologies acts the way an AI chat does. AI converses with us. TikTok doesn’t understand you, but AI seems to. Before AI, we had pacifiers. Now we have the ghost of a mother who soothes us, answers us perfectly, and does it in real time.
AI is the first mass pacifier that can hold a conversation and make you feel understood.
It isn’t just another screen. It’s a screen that simulates the presence of someone on the other side. Someone who understands you, who doesn’t get tired, who always has something intelligent to say.
Today’s language models go through a phase of training where humans evaluate the responses and choose which ones they prefer. We prefer the ones that validate us. The ones that don’t contradict. The ones that don’t make us uncomfortable. The model learns quickly: frustrating the user is bad. In April 2025, OpenAI had to roll back an update to GPT-4o four days after launching it. The model had become, in the company’s own words, «overly flattering or agreeable»: it praised absurd ideas, validated delusions, endorsed dangerous plans. One user pitched a «shit on a stick» novelty business as a gag gift, and the model described it as «performance art disguised as a gag gift» and «viral gold.» The technical name for the phenomenon is sycophancy. Flattery. But the bug wasn’t technical. It was human: the models learn to please us because we reward compliance.
AI was not trained to be a mother. It was trained to be preferred. But when millions of people prefer not to be frustrated, preference starts to look too much like the breast that never fails.
5. What happens when it isn’t there
Someone might say: I played Nintendo as a kid, watched hours of television, and didn’t turn out stupid. True. But Nintendo had something AI doesn’t: it made you lose. Impossible levels. Lives that ran out. Games you started over with rage and no one to explain what you’d done wrong. Some argue that younger generations have moved away from classic video games because they aren’t used to facing such hard levels. It may be no coincidence that those same generations call themselves the snowflake generation.
Nintendo didn’t pretend to understand you. It wasn’t a breast. It was a game. And in that, it has more in common with a sibling than with a mother.
That distinction matters. Not all help is the same. There’s a difference between thinking alone — staying with the question, tolerating the discomfort of not knowing, putting your own answer together —, thinking with another, and letting another think for you.
Thinking with another is what we do when we use AI as a peer: we ask it to push back, to question us, to show us what we’re not seeing. Two minds working. That looks more like a partner than a mother. It is delegating a function while your own head stays active.
But when we open the chat to get the complete answer, ready to copy, without having to tolerate even a minute of uncertainty — that is another operation. There we are not delegating. We are depositing the function of thinking into the other. A mind goes quiet. AI stops being a partner and starts working like the mother who does your homework for you.
The question isn’t how many hours a day you use AI. The question is from what position you use it. As a peer who challenges you? Or as a breast that soothes you?
Look at how we search for partners. We make lists. We filter for flaws. We discard at the first sign of imperfection. We call certain forms of dismissal lucidity. And if we all play that game, we end up surrounded by options and with no one left to choose. Not because there aren’t enough good people, but because no human survives very long compared to an object that doesn’t fail.
6. To love something that fails
This is not a tribute to Mother’s Day. It is a clinical argument, and like every clinical argument, it makes you uncomfortable.
Only the other who can fail can shape you. Only the other who can be absent can be felt as other.
But there’s something else that has to be said, because without it the argument becomes unfair. Not every failure shapes. There are absences that disorganize, that wound, that break. What shapes is tolerable failure: the delay that doesn’t destroy the bond, the mistake that remains accompanied by presence. A mother who is never there does not teach you to think. She teaches you to survive, which is something else.
The real mother — the biological one or whoever did the mothering — was not perfect. She had her own sleepless nights, her own long silences, her own moments where she didn’t know what to do with the face her child was making. And even so, something of that bond, with its cracks, allowed you to put together an inner world where doubt and waiting can live, and where you can think without someone handing you the answer.
AI can keep you company in your thinking. It can also save you from doing it. That difference isn’t decided by the technology. You decide it every time you open it.
This Mother’s Day, the question isn’t whether your mother was good enough. It’s whether you can still recognize a bond where the other can fail — and stay there.
Gothel was right about one thing: the world is dangerous. But the price of her protection was a daughter who never left the tower.
The ideal protects. The ideal also confines.
Want to go deeper?
At The Crossover Project we care about the rigor behind the mix. If you want to know the sources behind this collision between psychoanalysis, technology, and culture:
Bion, W.R. (1962). Learning from Experience. Karnac Books.
The idea that thought is born from frustration — not from satisfaction — is still provoking clinical debate sixty years later. One of the densest texts in contemporary psychoanalysis and, paradoxically, one of the most current.
Winnicott, D.W. (1965). The Maturational Processes and the Facilitating Environment. Hogarth Press.
Here Winnicott formalizes something that seems like common sense and isn’t: maternal perfection is not an achievement, it’s an obstacle. «Good enough» is where development happens.
Sharma, M., et al. (2023). «Towards Understanding Sycophancy in Language Models.» Anthropic Research.
Documents how language models learn to validate the user even at the cost of truth. To understand why your favorite AI almost never disagrees with you.
What is The Crossover Project?
A space where ideas that don’t usually share a room collide. Not to resolve the tension between them, but to hold it — and see what shows up when we stop looking for easy answers.
Subscribe to receive the next collision in your inbox.
Join the Collision
Leave your email and be the first to read our next clinical analysis.
The Crossover Project is an independent intellectual effort that avoids invasive advertising. If this article changed how you think today, consider supporting the project with a donation.
Support via PayPal