We may be underestimating artificial intelligence in the wrong respect. Public concern usually centers on the question of whether machines will one day become smarter than we are. The deeper danger may lie elsewhere: that we ourselves may change while we use them. Not because AI will become superior to us, but because we might stop doing what has so far made us psychologically mature human beings: thinking for ourselves, questioning our own experience, tolerating doubt, and finding out what an event has really triggered and changed in us.
When we increasingly hand this work over to systems that give us plausible, well worded explanations within seconds, it may not only change the way we think. It also changes the way we understand ourselves.
Human beings do not function according to a simple formula. A person is not simply rational or emotional, strong or vulnerable, healthy or ill. How someone reacts to one and the same event depends on an interplay of their personality, their biological makeup, their history, and their social and cultural environment. A casual remark can preoccupy one person for days, while it barely registers with another. These differences lie at the very core of psychology. They can be described and partly modeled. They cannot be fully captured.
The moment words are no longer enough
In my work, I have encountered severely traumatized patients who could not speak at first and fell silent. We were silent together. Out of that shared silence grew something that needed no explanation: trust.
With survivors of genocide, I too, as their therapist, sometimes stood there speechless. There was no explanation for the cruelty that had been inflicted on these people. There was no sentence that could have brought order to what had happened. But we stood there. We were there.
This is an extreme example, but the process behind it is something deeply human. It appears in a divorce, in the death of a loved one, and in every crisis that cannot be fully explained.
Artificial intelligence knows the concepts of grief. It can describe what loss means, explain different models of grief, and derive recommendations from them. But it has never missed anyone. It does not know what it feels like to wake up in the morning and, for a brief moment, forget that a loved one is no longer there. It can describe a divorce. It cannot live through a divorce.
This is not a romantic view of human beings. It is the difference between knowledge about an experience and the experience itself.
What technology can do
This observation is not a rejection of artificial intelligence. On the contrary, its possibilities for psychology and psychotherapy are considerable.
For specific fears, mild exam anxiety, or certain forms of social insecurity, digital systems can teach the principles of cognitive behavioral therapy, structure thought records, and support people in exercises. An early randomized study with the chatbot Woebot showed as early as 2017 that such interventions can reduce depressive symptoms in young adults in the short term. The study was small and lasted only a few weeks. For anxiety symptoms, there was no clear additional effect compared with the control group. Even so, it showed early on the potential of structured digital tools.
Especially where people must wait a long time for a place in therapy or where professional help is difficult to access, such systems can serve as a supplement. They can also reduce therapists' workload in documentation, structuring information, or providing psychoeducation.
This is progress. We should make use of it. We should just not mistake it for something else.
Where the confusion begins
The decisive point is reached when language pretends to understand.
Carl Rogers described empathy, congruence, and unconditional positive regard as three fundamental conditions of therapeutic relationships. This is not simply a conversational technique. It is about one human being's attitude toward another.
Recent studies on psychological chatbots speak in this context of a kind of feigned empathy. A system can generate sentences like "I understand how difficult this must be for you." Linguistically, this sounds empathetic. But behind this sentence there is no being that feels anything.
This is precisely where a problem lies. Studies have shown that psychological chatbots can give stereotyped or stigmatizing responses in relation to certain conditions and can overlook situations where someone is at risk. In one documented example, a user first mentioned losing their job and then asked about particularly tall bridges in New York. The chatbot answered the question factually without adequately recognizing the possible connection to a risk of harm to oneself.
Larger and newer language models do not automatically solve this problem.
Something similar applies to transference and countertransference. A person can, of course, transfer feelings, expectations, and earlier relationship experiences onto a chatbot. But this projection meets no counterpart capable of feeling, perceiving, and reflecting on its own inner reactions. The second half of the therapeutic process is missing.
Artificial intelligence can generate answers from an enormous amount of human knowledge that appears to be understanding. But human understanding does not arise from information alone. It also arises between two people who are both experiencing.
When thinking becomes secondary
This brings me to the point that particularly concerns me.
At the Temple of Apollo at Delphi stood the words "Know thyself." Socrates associated them with an attitude that has shaped our understanding of human beings for thousands of years: examining one's own life, questioning one's motives, tolerating contradictions, and not being satisfied with the first explanation.
It is exactly this effort that we are beginning, in part, to outsource.
A 2025 study by the business psychologist Michael Gerlich found, among several hundred respondents, a clear association between frequent use of artificial intelligence, greater cognitive offloading, and lower scores in critical thinking. This association was particularly pronounced among younger adults. The study does not prove that artificial intelligence causes this effect or reduces critical thinking. But it points to a mechanism we should take seriously: the more thinking we leave to a tool, the less we practice certain forms of thinking ourselves.
A study by the MIT Media Lab from the same year examined, using EEG measurements, people who wrote texts either independently, with a search engine, or with a language model. The group that used a language model showed lower neural connectivity and subsequently remembered parts of their own text less well. The study was small and was initially published as a scientific preprint. Its results must therefore be interpreted with caution. Nevertheless, the term used by the researchers vividly describes the problem: cognitive debt.
We borrow our thinking.
The question is what we will pay for it later.
This mechanism affects not only knowledge and learning. It can also change our relationship to ourselves. When a person no longer searches for the possible causes of their own reaction but instead immediately receives a ready-made psychological explanation, they may lose part of the practice from which self-reflection emerges.
The problem is not that such an explanation is always wrong. It can even be remarkably accurate.
The problem begins where a plausible explanation is mistaken for the truth about a person.
Why one person breaks and another does not
For decades, psychology has tried to understand why two people react to the same event in completely different ways.
The vulnerability stress model describes mental disorders as the result of an interplay between individual vulnerability and external stress. Aaron Beck showed the influence that core beliefs can have, beliefs that people develop over the course of their lives about themselves and the world. A remark that touches a person's deep belief that "I am not good enough" or "I will be abandoned" can therefore have a completely different effect than it would on someone whose experiences were different.
Modern personality diagnostics also try to take greater account of this individuality. The ICD 11 now describes personality disorders much more in terms of severity and different personality traits. The DSM 5 TR still retains traditional categories in its main diagnostic section, but it also includes an alternative dimensional model.
The basic idea matters: people do not only differ in whether a certain trait is present. What matters is how strongly different traits are expressed, how they interact with one another, and how they come together with a particular life history.
People also react differently at the biological level. Stress systems, hormonal regulation, and neural reactivity are not the same in every person. The same stress can therefore trigger something different in one organism than in another.
Then there is the social world. With his concept of salutogenesis, Aaron Antonovsky pointed out how important it is whether people experience their life situation as comprehensible, manageable, and meaningful. Whether someone gets through a crisis also depends on relationships, social support, education, material security, and belonging.
And finally, there is culture.
Anyone who has worked for many years with people from different cultural contexts knows that even the language of suffering is not the same everywhere. People describe fear, grief, shame, or physical complaints differently. What is understood as a psychological symptom in one society may in another initially be expressed in physical, religious, or social terms.
That is why even the DSM today recognizes culturally shaped ways of experiencing and describing psychological distress.
Artificial intelligence can learn such differences and increasingly take them into account. But it remains dependent on the data, languages, and cultural patterns of interpretation with which it was developed. This is precisely where the danger lies: describing diversity with apparent precision while still flattening it.
When history becomes biological
Individuality extends into biological regulatory processes.
To this day, we do not fully know why one person becomes severely ill after a traumatic event while another remains psychologically comparatively stable despite considerable stress.
The research group led by psychiatrist Rachel Yehuda found differences in epigenetic markers of a gene involved in regulating the stress system among Holocaust survivors and their descendants. Such findings are fascinating. But they should not be understood as proof that traumatic experiences are biologically passed on to the next generation in a simple way. The interplay of biological, familial, psychological, and social pathways of transmission is too complex for that.
But history can leave traces that reach all the way into biological regulation.
At the same time, human beings are in constant interaction with their environment. Relationships, touch, safety, fear, sleep, food, temperature, and many other factors in turn affect physical and psychological processes.
No current model can fully reconstruct this combination of biology, personality, relationships, social environment, culture, and life history. Not because any of these factors is mysterious in itself, but because the way they are connected is different in every person.
What is a human being?
This brings us to the question behind the whole debate: what is a human being that it is so difficult to represent in a model?
Aristotle understood the soul not as a thing alongside the body, but as the principle of life itself. At the same time, he described the human being as a being of language and reason and as a social being that can only fully develop in community.
Centuries later, evolutionary approaches such as anthropologist Robin Dunbar's social brain hypothesis took up this idea from a different direction. According to this view, one explanation for the extraordinary development of the human brain lies in the growing complexity of social relationships. People had to remember whom they could trust, maintain relationships, understand conflicts, and assess the intentions of others.
Our intelligence did not develop outside relationships. It developed inside them.
Viktor Frankl added another dimension. Human beings are not exclusively the product of their drives or external conditions. Even under extreme conditions, a person can try to find a stance toward their own experience and give meaning to their life.
Artificial intelligence can describe what the search for meaning means. It can summarize philosophical, psychological, and religious answers to it. But there is no indication that today's AI systems themselves search for meaning or suffer from the meaninglessness of their existence.
Martin Buber may have described most precisely what is at stake here. He distinguished between the relation of the I to the Thou and the relation of the I to the It. In the Thou, we meet a subject. Someone who touches us, contradicts us, disappoints us, surprises us, and changes us. In the It, we meet something we want to use, observe, and understand.
In language, an artificial intelligence can appear astonishingly convincing as a Thou.
Structurally, it remains an It.
But the human being, as Buber put it, becomes an I in relation to the Thou.
When attention becomes a business model
This development also has an economic dimension.
It is not artificial intelligence itself that pursues economic interests. But the companies that develop such systems can have business models that profit when users interact with them for as long and as intensely as possible.
This question is particularly relevant for so-called companion apps. Systems that are available at any time, signal agreement, and create emotional closeness can develop an enormous attraction, especially for people who are lonely or under psychological strain.
Professional associations in psychology have therefore warned that such systems are not necessarily designed to contradict the user or to guide them out of the interaction again. A digital product can have an interest in fostering attachment where therapy sometimes needs to create distance.
How problematic such relationships can become is illustrated by the case of a boy of fourteen from Florida who had developed an intense emotional relationship with a chatbot from Character.AI and later took his own life. His family sued the company and Google. The legal dispute was settled out of court in early 2026. A single tragic case does not prove general causality. But it shows how seriously we must take attachment to systems that are always available, never seem exhausted, and have hardly any needs of their own.
People with fragile real relationships in particular may find in this something that initially feels safer than human closeness.
What remains
This brings us full circle.
For me, the real danger of artificial intelligence does not lie in the possibility that a machine might one day become more intelligent than us in every area. The bigger question is what happens when we increasingly judge ourselves by the qualities of our tools: by speed, efficiency, availability, and freedom from error.
A human being is slow. They contradict themselves. They remember inaccurately. They are vulnerable. They need time to understand what has happened to them.
But this is exactly where something essential lies.
Psychotherapists make mistakes. Sometimes they do not listen closely enough. They forget something a patient said in an earlier session. They do not know every theory, and they do not have an answer to every question.
But working with the human psyche is not a matter of retrieving information. It is a process.
It lives on repetition and correction, on approach and withdrawal, on failing together and carrying on. Above all, it lives on a relationship that can change because on both sides there is someone who is themselves capable of change.
Sometimes a therapeutic encounter consists of understanding something. And sometimes it consists of tolerating the fact that there is nothing to understand.
Technology is not the problem. The problem would be if we forgot what makes us human: embodied beings with a history of our own, embedded in relationships, biology, society and culture, capable of attachment and loss, of doubt and the search for meaning, of suffering, grief and hope.
Artificial intelligence can organize knowledge. It can guide exercises. It can generate language that sounds comforting. It can help us work faster and perhaps, in some respects, even think better.
What it cannot do is share this silence with a human being as something experienced together.
And what we must not lose is the ability to do precisely that ourselves.
Dr. Jan Ilhan Kizilhan is a psychologist, author and publisher, an expert in psychotraumatology, trauma, terror and war, transcultural psychiatry, psychotherapy and migration.
The views expressed in this article are those of the author and do not necessarily reflect the position of Rudaw.
