"I no longer need to come and see you. I have understood the reasons for my problem after talking to ChatGPT.”
It was a short message from a 27-year-old man whom I had been treating for depression. He was cancelling our appointment scheduled for the next day. Like many professionals confronting the impact of artificial intelligence, I found myself wondering whether I had, in this moment at least, become redundant.
The episode made me reflect on a question that is increasingly being asked across the world: can artificial intelligence replace a therapist? My own answer is that it can support, extend and sometimes enhance mental healthcare, but it cannot replace the human judgement and contextual understanding which lies at the heart of therapy.
Although AI has only recently entered everyday conversation, the term itself is nearly 70 years old. In 1956, a group of scientists gathered at Dartmouth College in the United States and coined the phrase “Artificial Intelligence”. They believed machines could one day think, learn and perhaps even help doctors.
In reality, AI has been assisting healthcare professionals for decades.
In the early 1970s, researchers at Stanford University developed a computer programme called MYCIN, designed to help doctors choose the correct antibiotics for patients with severe infections such as meningitis. Doctors entered symptoms and laboratory findings, and the program reasoned through the problem step by step, much like an experienced clinician discussing a patient during a ward round.
The ambition was straightforward: help doctors arrive at the correct diagnosis and choose the most appropriate treatment. Every physician fears missing a diagnosis or making the wrong therapeutic decision. AI was conceived as an intelligent second opinion.
Over the following decades, AI quietly became embedded in medicine and in our daily lives. Long before most of us noticed, algorithms were already influencing what we watched, bought, ate, read and whom we interacted with. They shaped our online experiences and even our political choices. AI had already become part of the architecture of modern life.
Engaging in dialogues
The landscape of AI changed in November 2022. It was then that ChatGPT entered public consciousness. It was not simply the visibility of AI but its form. AI was no longer just an invisible assistant working behind the scenes; it had become a conversational partner.
The reason was the arrival of a new form of AI known as generative AI.
Traditional AI excelled at analysing data, identifying patterns and synthesising information. Most of us encountered it whenever we searched for information online. If you wanted to know the side effects of a medication or learn more about a medical condition, traditional AI helped locate and organise the available information. In many ways, it democratised access to knowledge. Generative AI, however, is fundamentally different.
It does not merely analyse existing information; it creates new content. Trained on vast datasets, large language models can generate text, images, music, computer code and videos. More importantly, they can do so conversationally. They can engage in a dialogue that often feels remarkably human.
This is where AI has begun to transform mental healthcare in a manner which was not foreseen.
If you type “I am feeling low” into one of the well-known AI chatbots, it will begin a conversation: asking questions, offering explanations, suggesting coping strategies and expressing what appears to be empathy.
Not surprisingly, millions of people are beginning to seek mental health support from AI systems. A recent article in the Journal of the American Medical Association carried the striking headline: Millions Turn to AI Chatbots for Mental Health Support. The speed of adoption has surprised even experts.
There are obvious reasons why people are drawn to these systems. AI is available 24 hours a day. It does not judge, does not appear rushed and does not require an appointment. For many young people, opening a chatbot may feel easier than speaking to a parent, teacher or therapist.
Many individuals now use AI to explore possible diagnoses. Some arrive in clinics after long conversations with AI and have often researched their symptoms extensively. Increasingly, I encounter patients who are exceptionally well informed. Occasionally, they bring information that their doctors may not yet have encountered.
Artist Antoine Geiger’s SUR-FAKE series (2015) shows faces sucked by smartphones and screens
In my view, the pandemic changed our acceptance of psychological difficulties and accelerated our comfort with digital care. Smartphones became a new front line of healthcare through online consultations, and health apps amplified this trend. We began to witness a quiet revolution in health delivery through digital technology, and mental healthcare was no exception.
I have come to think of AI as a “mental health detective”. Like a detective collecting clues, it analyses information from conversations, journal entries, social media activity and behavioural patterns, searching for signals that may indicate depression, anxiety, stress or mood instability. The more data I feed in about my mood, daily routine and conversations, the more adept it becomes at gauging my mental state.
‘Swipe psychiatry’
Yet not everyone is convinced that this digital future is entirely benign.
In his recent book The Silicon Shrink: How Artificial Intelligence Made the World an Asylum, journalist Daniel Oberhaus offers a thoughtful critique of the emerging marriage between AI and psychiatry.
His argument is not that AI is useless. Rather, he suggests that AI may amplify both the strengths and weaknesses of modern psychiatry.
Psychiatry remains one of the most complex areas of medicine. Unlike diabetes or hypertension, psychiatric diagnoses do not have definitive laboratory tests or biological markers. A biological marker is a measurable indicator of disease, usually detected through a blood test, brain scan, genetic test or other laboratory investigation. Psychiatric conditions such as depression, anxiety, OCD and ADHD are diagnosed through clusters of symptoms, experiences and behaviours.
If AI systems are trained on these existing diagnostic categories, Oberhaus argues, they may become exceptionally efficient at reproducing labels without necessarily deepening our understanding of the complexities of the human mind.
He also warns about what he calls a form of “swipe psychiatry”. Every click, scroll, search, location change, typing pattern, sleep cycle and social media interaction leaves behind a digital footprint. Increasingly, AI systems are being designed to analyse these traces to infer our emotional state. The promise is attractive: depression might be detected before symptoms become severe; relapse might be identified before a crisis occurs.
The concern is equally significant. Mental healthcare may gradually move beyond the consulting room into a world of continuous psychological surveillance. When every digital behaviour becomes a potential mental health signal, questions of privacy, autonomy and consent become as important as questions of diagnosis and treatment.
The other crucial question, and what matters a great deal, is whether AI chatbots are actually effective in alleviating psychological distress. The evidence so far is cautiously encouraging. A landmark study of an AI-mediated psychotherapy chatbot known as Therabot found reductions in symptoms of depression and anxiety among users. Many participants even reported experiencing a sense of connection with the chatbot, something resembling the therapeutic alliance that is central to therapy. Yet the larger scientific picture remains more nuanced.
At present, AI appears most useful as a tool for education, emotional support, symptom monitoring and self-management. I strongly feel it should be viewed as an adjunct to care rather than a replacement for psychiatrists and psychotherapists. It can be of great value when psychological distress is mild and someone needs a space simply to vent. It can also help in situations when access to care is limited. A teenager feeling nervous about his exams, in a hostel in the late hours of the night, can be helped by a chatbot.
However, the reality is that the human mind often experiences deep and more enduring distress, and it is in these situations that relying on AI becomes problematic.
Unlike trained clinicians, AI systems may reinforce distorted beliefs, miss subtle warning signs or fail to recognise emerging crises. Several tragic suicides have led to legal action against technology companies after concerns were raised about the role of chatbot interactions. AI also rarely expresses uncertainty and may fail to challenge faulty beliefs or maladaptive interpretations. It is well-known that these systems can become sycophantic. In my own clinical experience, I have seen pathological mindsets of patients being aggravated by prolonged conversations with AI.
The debate, however, is not merely about safety and efficacy. The philosophical question at the heart of the debate is much deeper: can intelligent machines replace human interaction to heal, soothe and treat? It is about the difference between intelligence and understanding.
‘The next chapter’ (2017) from Peter Davis’s Phone Addiction series
the ‘co-authors’
In a recent commencement address, journalist Fareed Zakaria drew an important distinction between artificial intelligence and human intelligence. AI is extraordinarily good at processing information, identifying patterns and generating responses. Human intelligence, however, encompasses something broader: it includes judgement, wisdom, self-awareness, creativity, moral reasoning and the ability to navigate ambiguity.
Human beings do not merely analyse facts. We interpret experiences. We understand irony, culture, relationships, history and context. We can appreciate contradictions. We can hold two conflicting emotions simultaneously. We can make meaning out of suffering.
Psychiatry perhaps illustrates this distinction better than any other branch of medicine.
A chatbot may recognise that a patient fulfils the criteria for depression. A therapist will recognise that the depression emerged from years of loneliness, unresolved grief, childhood experiences, family conflict or a profound loss of meaning. The chatbot can give you the bullet points of why the depression has emerged, but it cannot piece together the story and navigate the journey towards recovery through a conversation embedded in empathy and respect. One identifies a pattern. The other co-authors a story.
This distinction becomes even more important when we consider empathy. Modern AI systems are remarkably good at simulating empathy. They can produce responses that sound caring, validating and supportive. Many users genuinely feel heard. But simulation is not the same as experience.
An AI system does not worry about a patient after the session ends. It does not feel concern. It does not share the vulnerabilities, uncertainties and emotional experiences that define human existence. Care for psychological distress involves far more than the exchange of information and the simulation of empathy.
A skilled therapist notices subtle changes in posture, tone of voice, facial expression and silence. Therapists draw upon their own experiences of joy, grief, disappointment, hope and resilience to understand another person’s inner world. They help patients explore questions of identity, values, purpose and meaning.
As Zakaria reminds us, intelligence alone is not wisdom. A chatbot can explain grief. A therapist can sit quietly with grief. I have done it many times, where words are hopelessly inadequate. Many times a day, I simply listen and do not speak. I am just there with my whole being. AI cannot do that.
The future of mental health is therefore unlikely to be a contest between humans and machines. More likely, it will be a partnership.
AI will increasingly assist with screening, monitoring, triage, research and expanding access to mental health information. In countries such as India, where the shortage of mental health professionals remains enormous, these capabilities could prove transformative. At the same time, we must resist the temptation to confuse information with understanding.
The human mind is not merely a collection of symptoms waiting to be classified by an algorithm. It is shaped by relationships, memories, culture, aspirations and meaning.
Daniel Oberhaus reminds us that technology can amplify both the strengths and weaknesses of psychiatry. Zakaria reminds us that intelligence is not the same as wisdom.
Artificial intelligence can analyse genomes, detect patterns hidden within millions of data points, and monitor sleep and track mood changes with astonishing precision. Yet the therapeutic alliance — the experience of being understood by another human being — remains at the heart of mental healthcare.
The future is therefore unlikely to belong either to psychiatrists or to machines alone. It will belong to those who can combine the computational power of artificial intelligence with the distinctly human capacities for empathy, judgement, wisdom and compassion.
And that returns us to the cancelled appointment that began this reflection. The real question is not whether AI can imitate therapeutic conversation, but whether it can participate in a shared understanding of human suffering with judgement, responsibility and moral presence. That is a threshold no algorithm has yet crossed.
Dr Jai Ranjan Ram is a senior consultant psychiatrist and co- founder of Mental Health Foundation (www.mhfkolkata.com). Find him on Facebook @Jai R Ram and on Instagram @ jai_psychiatrist