Are we becoming overly reliant on AI, and is it wise to share our inner world with a language model? This piece looks at where AI is a genuinely useful tool, where it is merely comforting, and where it may quietly get in the way.
Artificial intelligence has become one of the fastest adopted technologies in history. Millions of people now ask ChatGPT, Perplexity and other tools questions they once reserved for friends, mentors, physicians, teachers or therapists. They seek advice, emotional support, relationship guidance, and explanations for almost any question that arises in everyday life.
Many report feeling understood. Some even describe AI as more patient, more validating and less judgmental than the people around them.
The short answer, before the long one: AI can supply information and a convincing simulation of empathy. It cannot supply a relationship. Psychotherapy works largely through what happens between two conscious people, which is why human therapists remain relevant even as language models continue to improve.
Within the euphoria that surrounds AI, we are tempted to think that nothing like this has been seen before, and to treat it as an invention that will change our lives.
Oh well… curb your enthusiasm, for a minute.
Change our lives? Well, yes. But doesn’t that apply to every invention?
And never seen before? I beg to differ.
The Search for Answers Is as Old as Civilisation
In our quest for knowledge, we humans have always been searching for answers. This phenomenon is as old as time itself and did not begin, as we might be tempted to think, with large language models. While those models draw on computational power almost beyond human comprehension, the idea of having an unlimited source of wisdom at our disposal is as old as civilisation.
Allow me to remind you.
There has always been an aspiration to create an external source of reliable, comprehensible knowledge. Gnothi seauton (“Know Thyself”), the idea of getting to know oneself, was inscribed above the entrance of the Oracle of Delphi: the sacred place where kings, generals, state councils and private citizens travelled miles to seek divine guidance on political, military and personal matters. They all consulted Pythia, the priestess of Apollo, believing that Apollo himself spoke through her. Although not a machine, the Oracle of Delphi was arguably humanity’s first knowledge interface. For reference, this was the 5th century BC.
And that was only the beginning as we know it.
The idea of finding answers beyond ourselves created the talking heads tradition: a widespread literary and folkloric motif found across antiquity and the Middle Ages. The tradition refers to any severed, sculpted, magical or artificial head capable of speech, prophecy or dispensing wisdom, which creates an entire category of intelligent or prophetic objects, one of the closest mythological ancestors of AI.
Talking heads made room for the Brazen Head: a medieval legend about an artificially constructed head made of brass. Unlike mythological talking heads, the Brazen Head was significant because it was manufactured by human ingenuity, often through a combination of mechanics, alchemy, astrology and natural philosophy. It represented an attempt to build an intelligence rather than discover or inherit one.
Image: illustration on the title page of the 1630 reprint of Robert Greene’s Friar Bacon and Friar Bungay. The brazen head is in the centre. Source: icysedgwick.com
And so the story continues.
Fast forward to the twentieth century, where two brilliant minds, Douglas Adams and Isaac Asimov, independently pursued the automaton that would be both a source of wisdom and a provider of answers to any question a human mind can ask. Both arrived at a peculiar conclusion: that bare information is, in fact, quite useless and meaningless.
This is clearly depicted in Douglas Adams’ The Hitchhiker’s Guide to the Galaxy, in the Deep Thought computer built to answer the greatest question of life, the universe and everything (is that not how we tend to use language models today?). To everyone’s surprise, after seven and a half million years of computing, Deep Thought produced a simple answer: “42”.
This was as unexpected as possible. Although probably mathematically and statistically correct, the answer provided no meaning or value to anyone, “so another, even bigger, computer had to be built to find out what the actual question was. And this computer, which was called the Earth, was so large that it was frequently mistaken for a planet, especially by the strange apelike beings who roamed its surface, totally unaware that they were simply part of a gigantic computer programme.”
Adams, Douglas. The Ultimate Hitchhiker’s Guide. Wings Books, 1996.
So there we were again, despite all the technological advances, back to the talking heads. Conceptually there was no real difference between them and the computers: all yapping continuously, yet providing no answer of any actual use.
This did not stop with Deep Thought, or with the automaton.
Despite the peculiar prediction, humans continue to seek shortcuts to finding answers. The story continues into today’s frustration of trying to hold a debate with ChatGPT, which keeps adjusting and agreeing with everything we say. It is comforting to meet with agreeableness, even artificial agreeableness, and easy to forget that in that comfort we have missed the thing we came looking for: encouragement to think, to consider different perspectives, and, in the end, the answer.
The comfort has lulled us into forgetting that “if everybody is thinking alike, then somebody isn’t thinking.” (George S. Patton)
Whom Does the Grail Serve? The Question We Forget to Ask
Now that we can see how old the idea of an ultimate source of infinite wisdom is, the Holy Grail if you will, it becomes harder to argue that there is anything genuinely new about contemporary language models built to provide instant answers.
There is one feature that does appear as a novelty. Thanks to automation and immense computing power, AI is very successful at appearing humanoid. Its communication style can be adjusted to our preferences, so that it becomes, at times, difficult to tell whether we are talking to a person or a model. The fact that we have grown accustomed to communicating through text rather than old-fashioned verbal conversation only feeds the illusion. This is one of the reasons the question of whether chat-based therapy is effective is not as simple as it first appears.
While it becomes more cryptic to decipher who or what we are engaging with, it is worth stating plainly: AI is not human, and its language is based on statistical probability.
A language model is a statistical system that learns patterns in large collections of text by estimating the probability of one token (a word, in this context) occurring after a sequence of previous tokens. Although it can generate complex and meaningful language, it operates through learned statistical patterns rather than human-like understanding or conscious reasoning.
Let us stop here for a moment and consider the unexpected fate of the talking heads, the predecessors of contemporary language models.
It is rarely mentioned that Thomas Aquinas destroyed the automaton because it was so talkative that it continually interrupted his train of thought.
Interrupting his train of thought.
Legend has it that the medieval talking heads were created only by compelling demons and that, in the end, the ultimate answer was always missed, either from forgetfulness or from exhaustion. Through constant yapping, the receiver would miss the only thing the talking head had been made for. This exhausted and disarmed the receiver and, paradoxically, defeated the head’s single purpose.
That leads us back to the absurdity Douglas Adams glimpsed in Deep Thought, which gave a statistically correct answer that was, ironically, completely useless, and which led only to the simulacrum of building yet another computer commonly mistaken for a planet.
Whom does the Grail serve, really? In the legend, the Grail could have saved the barren kingdom had the question been asked. Whom does the Grail serve? It remained the only question unasked.
It is also the question we fail to ask, lulled by the comfort of instant answers. Which brings the matter into its contemporary form: if AI can explain trauma, identify attachment styles, generate coping strategies and respond with apparent empathy twenty-four hours a day, why would anyone continue paying a human therapist?
What Artificial Intelligence Is, and What It Is Not
To understand the conundrum, we need to understand what artificial intelligence is and, more importantly, what it is not.
Artificial intelligence is not a mind, and certainly not a human mind. It does not possess consciousness, intentions, subjective experience, emotions, curiosity or self-awareness.
Most conversational AI systems today are, as mentioned, large language models. Rather than thinking in the human sense, they learn statistical relationships across enormous amounts of text and predict which words are most likely to follow one another, drawing on patterns acquired during training.
The result can feel astonishingly intelligent, because human language carries knowledge, emotion, reasoning and culture. By mastering linguistic patterns, these models become remarkably effective at generating responses that resemble thoughtful conversation.
But resemblance does not extend further. It is we, the humans, who make the abstraction.
Put more plainly: a language model does not understand suffering because it has suffered. It recognises patterns associated with descriptions of suffering, and that is all. That distinction is fundamental.
Image: this is what ChatGPT actually looks like, inside a Microsoft data centre. Source: CNET
Information, Simulation and Relationship Are Not the Same Thing
One of the greatest sources of confusion in conversations about artificial intelligence is that we often treat three fundamentally different phenomena as if they were interchangeable: information, simulation and relationship.
Information is the transmission of knowledge. It answers questions, explains concepts, organises ideas and helps us understand the world. AI excels at this. It can summarise decades of psychological research, explain attachment theory, compare therapeutic approaches, or help someone articulate an emotional experience with remarkable speed and clarity.
Simulation goes a step further. It does not merely provide information; it creates the experience of interacting with another mind. Through language, tone and responsiveness, AI can convincingly imitate empathy, curiosity, validation and reflective dialogue. The experience can feel deeply personal, because our brains are naturally wired to infer intention and consciousness from coherent social interaction. A convincing simulation can therefore evoke genuine emotional responses, even when we know intellectually that no subjective experience exists on the other side.
A relationship is something else entirely. It exists between two subjects who each possess their own inner world, and it is shaped by mutual influence, vulnerability, responsibility, unpredictability and the possibility of genuine encounter. In psychotherapy, healing does not arise solely from receiving accurate interpretations or compassionate responses. It emerges from experiencing oneself in the presence of another conscious human being whose perceptions, emotions and choices are real. The therapist is not simply generating appropriate language; they are participating in an unfolding interpersonal process for which they carry ethical and relational responsibility.
Confusing these three levels leads to unrealistic expectations. Information can educate. Simulation can comfort and support reflection. Neither is equivalent to a therapeutic relationship.
That a conversation feels meaningful does not necessarily mean a relationship has taken place. Recognising this distinction does not diminish the value of AI; it clarifies what each can contribute. AI may become an extraordinary source of information and an increasingly sophisticated conversational partner, but psychotherapy remains rooted in something that cannot be reduced to language alone: the encounter between two human minds.
Why Human Psychotherapy Still Matters
Whether we welcome AI or fear it has become largely irrelevant. It is already here and it will continue to develop. Attempting to reject it entirely is neither realistic nor especially useful.
The task is adaptation.
Like every transformative technology before it, artificial intelligence asks us to become more intentional about how we use it.
Can we preserve our capacity for reflection when answers become effortless?
Can we tolerate uncertainty when certainty is available on demand?
Can we maintain authentic relationships when simulated ones become increasingly persuasive?
These questions concern us far more than advances in computing power do. They are also being asked elsewhere on this blog, in When AI Entered the Therapy Room and in Where Is the Hive? Reflections on AI, Community, and What It Means to Stay Human.
AI systems show considerable capacity to support psychological reflection by organising thoughts, identifying cognitive and behavioural patterns, explaining psychological concepts and providing psychoeducation. For clients and clinicians alike, AI may function as a useful adjunct for self-reflection, case conceptualisation and therapeutic support. The same caveats that apply to the safe use of mental health apps apply here too.
Psychotherapy, however, extends beyond the provision of information and insight. If it were only about information, we would not be raising these questions in the first place.
A central distinction between AI-assisted support and human psychotherapy lies in the process of psychological transformation. Many people possess substantial awareness of their difficulties yet remain unable to achieve meaningful change, which remains the central problem in psychotherapy. This happens because psychological suffering is often maintained not only by conscious beliefs but also by implicit emotional patterns, attachment experiences, relational expectations and embodied responses.
The therapeutic relationship
The therapeutic relationship is one of the primary mechanisms through which psychotherapy facilitates change. A human therapist provides a consistent, attuned and emotionally responsive relationship in which clients can have corrective interpersonal experiences. Through empathy, appropriate boundaries and sustained presence, therapists help modify maladaptive relational patterns and support the development of greater emotional regulation and psychological security.
Co-regulation and reading between the lines
Another essential component of psychotherapy is interpersonal and physiological co-regulation. Human therapists use nonverbal communication, including tone of voice, facial expression, timing and presence, to support emotional regulation in ways AI cannot replicate. Therapists also read between the lines, observing subtle affective shifts, contradictions, avoidance patterns and implicit communication that may not be accessible through text-based interaction.
So while AI offers real value in supporting reflection, education and structured psychological input, it does not replace the relational and experiential dimensions of therapy. The enduring value of human therapy lies not merely in providing insight but in facilitating change through a relationship that engages emotional, interpersonal and embodied processes. This holds across modalities, whether the work is cognitive behavioural, psychodynamic or integrative.
Agency and the Search for Balance
So how do we cope? The healthiest position is neither enthusiasm nor rejection. It is discernment.
AI is exceptionally useful when it expands learning, stimulates reflection, organises thinking, supports creativity and increases access to knowledge. It can help people formulate questions they did not yet know how to ask. It can encourage curiosity and provide language for experiences that previously felt confusing or inexpressible.
It becomes a problem when it quietly replaces the capacities that allow human beings to mature: tolerating ambiguity, engaging in genuine dialogue, accepting interpersonal risk, and developing meaning through lived experience rather than instant explanation.
Technology should increase human agency, not diminish it. If we begin outsourcing not only our memory but also our reflection, judgment, emotional processing and relationships, convenience comes at the cost of psychological development.
The measure of AI’s value is therefore not how much it can do for us. It is whether, after using it, we are more capable of thinking, choosing, relating and taking responsibility for our own lives.
That remains the defining task of psychotherapy. And it remains, at least for now, profoundly human.
What Happened to the Brazen Head?
Perhaps Thomas Aquinas was right to destroy the yapping brazen head. It was his attempt to keep his train of thought active, to preserve his imagination and to protect his capacity to think.
Fast forward to today. Let us learn from that. If we allow AI to take the lead, we let our trains of thought be interrupted, slowly losing the ability to reason: the one ability AI does not have.
What, then, is the value of psychotherapy? It is worth remembering that clients rarely come to therapy searching for information. More often than not, they know about their state as much as we therapists do. They come for connection and relation, built within the therapeutic space, and it is in that space that they seek, and find, the path to the change they have been looking for all along.
That is unattainable without human connection. This is one of the older questions in the field, and it long predates the current debate, as I explored in The Philosophical Roots of Psychotherapy.
Is it possible to compute and statistically derive wisdom? Or are these tools and models only databases connected to sources of information, limited by the materials they are built from (semiconductors and rare elements), capable of providing an almost unlimited collection of data yet devoid of real reasoning and true wisdom?
And what is it that we, when looking for answers, are really searching for? Mere data, information stored on servers in data centres, or the meaning beyond that data?
Without conclusion, practical reasoning and real connection to attach true meaning to it all, data is fairly meaningless and almost useless. Douglas Adams had already shown that, and made us laugh in the process.
It is the way knowledge and information affect us, our lives and our context that creates meaning beyond mere data, and allows the space to grow and expand.
Now allow me to share a final thought.
When I was introduced to this topic I thought, quite naively, that it would be an easy one to write. My train of thought went something like: “Oh, it’s just AI, and it is such a common topic, so composing an article is going to be a matter of organising what is mostly common knowledge.”
I thought I would get through it without breaking a sweat.
Oh my, how wrong I was.
Ironically, the topic of unlimited answers has raised more questions than any other I have encountered so far. The one that still lingers, and the one I invite you to sit with, is this:
What is it that we look for in therapy? What is it that we need? What is it that we are hoping to find?
If it were mere answers, the information, I would not be here writing this, and you would not be there reading it. So I have come to think that perhaps it is not the answers we are searching for, or ultimately need, but the process of seeking and finding them ourselves, because that process, and not the information gained, is what leads to change.
It is not what we find out, but what happens within us in the process of seeking and finding, that matters.
It was never about the answers. It has always been about transformation: something that has happened, and will continue to happen, within the safety of the therapeutic space.
Let us not forget, then, that we humans are made of stardust, and that the interaction we create through our relationships cannot be computed or recreated.
It is something each one of us carries within.
If you are thinking about working with a human therapist, whether alongside the tools you already use or instead of them, you can explore practitioners on It’s Complicated, or use the matching service to be paired with someone suited to what you are looking for.
Ognjena Irsevic is a psychological counsellor and psychotherapist in training with over seven years of experience working with people with personality disorders and their partners and families. She integrates Compassionate Inquiry with Reality Therapy, CBT and DBT, and her work focuses on helping clients rebuild self-respect and confidence in the aftermath of difficult relationships. She practises online from the Opatija Riviera in Croatia. Find Ognjena on It’s Complicated.
