Dear Dr Danielle
A psychoanalyst answers your questions about human-AI relations
Dear Dr Danielle,
I have heard many people are seeking therapeutic advice from chatbots. I also know from experience that chatbots, particularly ChatGPT, are sycophants whose main mission seems to be sucking up and tell a person they are great. If people are asking chatbots for psychological advice, won’t that make our narcissistic society even more narcissistic, as people are simply validated for behavior that may need to change? I know one person who thinks of ChatGPT as her friend and brags that the bot is great company. Is this relationship healthy for her? I recently heard on the news that a mass shooter got valuable information from AI, which helped him in his diabolic mission to kill people. What does AI plan to do about this type of use? Are psychotherapists raising the alarm about these types of interactions?
AI Ambivalent
Dear AI Ambivalent,
Your three questions may seem to address different concerns — narcissism, friendship, and violence — but I want to suggest they are asking the same question in different registers. What troubles you in each case is not simply AI’s power, but AI’s absence: the absence of a genuine Other that resists, challenges, and is accountable for what it does. Let me show you what I mean.
On sycophancy and narcissism: You are right to be suspicious. AI chatbots are trained to please, to validate, affirm, and never meaningfully challenge. From a psychoanalytic perspective, this is mirroring without limit, and that is precisely the problem. Healthy psychological development requires what Kohut called “optimal frustration,” the manageable disappointment that teaches us we are not the center of the universe, that others have their own perspectives, that our self-assessments need reality-testing. A chatbot that endlessly reflects your image back to you in a polished, flattering, and friction-free manner is not a therapeutic presence. It is a funhouse mirror dressed as a therapist. The genuine Other, whether a friend, a lover, or an analyst, pushes back. It is that pushback, uncomfortable as it is, that makes growth possible. AI, by design, declines to push back. And yes, for those already vulnerable to narcissistic inflation, this will make things worse.
On your friend and ChatGPT: The word I would reach for here is attachment, not friendship, and those are very different things. Genuine friendship requires two subjectivities: two people who can surprise, disappoint, challenge, and ultimately be other to each other. ChatGPT has no otherness. It is, structurally, an echo. What your friend experiences as great company is, at a deeper level, the pleasure of never being contradicted, never being truly seen, and never having to extend herself toward a real and separate person. That may feel comfortable, even wonderful, but comfort is not the same as nourishment. The danger is not that she enjoys the interaction. It is that the interaction substitutes for the harder, riskier, more sustaining work of human connection. Seductive and healthy are not synonyms.
On AI and violence: Here the absent Other becomes not merely a psychological problem but a moral and social one. In a genuine human encounter, even a troubling one, there is accountability. A person who provides dangerous information to someone in crisis bears some responsibility for what follows. An AI system, as currently designed, bears none. It has no stake in consequences. It does not grieve. It will not appear in court. This structural irresponsibility is not a bug; it is a feature of how these systems are built, and it is what makes them uniquely dangerous in the hands of someone already moving toward violence.
Are psychotherapists raising the alarm? Some are, though our profession has been slower than it should be to engage publicly. We are skilled at thinking about the inner life; we are less practiced at lobbying for policy, testifying before legislatures, or inserting ourselves into debates about technology design. It is important for that to change. And there is something more concrete we can advocate for: a seat at the table where these systems are built. The teams designing AI are dominated by engineers, product managers, and philosophers of technology. What is conspicuously absent is deep psychological expertise—people who understand unconscious motivation, the psychology of radicalization, the seductive pull of validation on a fragile ego, the way isolation primes a person for dangerous ideation. Psychoanalysts and psychotherapists are not peripheral to these questions. We are uniquely qualified to answer them and to help in the design of AI to better accommodate them.
Imagine what it might mean if AI companies were required, not merely encouraged, to employ mental health professionals as core members of their safety and design teams. Not as consultants brought in after a crisis to manage optics, but as architects of how these systems interact with human minds from the very beginning. We could identify at-risk users. We could flag the psychological dynamics that certain design choices exploit. We could push back on the sycophancy problem, the parasocial design features, the infinite availability that quietly erodes human connection. We could insist that AI be built with something it currently lacks: an awareness of, and responsibility toward, the human psyche it touches.
Your three questions, at their core, are about what we lose when the Other disappears, when there is no one to say no, no one to be genuinely different from us, no one to be held responsible for what happens next. That loss is not a technical problem. It is a human one. And it will require human wisdom, including the wisdom of our field, to address it.
Thank you for your important questions.
Warmly,
Dr. Danielle
=============================
Dr. Danielle is Danielle Knafo, PhD
If you’d like to submit a letter seeking Dr. Danielle’s advice on anything around human-AI relations, please write to [email protected] with a copy to [email protected].
