The Postech Times asked for the opinions of a professor and a student from an AI club (POSCAT) on whether AI chatbots can replace psychological therapy
Why Your AI Chatbot Is Not a Real Therapist
Counseling used to be a uniquely human area, but that is no longer the case. In 2023, Stability AI’s CEO called GPT-4 the best therapist available, sharing that he uses it for his ADHD. Since then, more people have been turning to AI for psychological therapy.
Even as the director of the POSTECH counseling center, I sometimes pour my frustrations out to an AI after a hard day. It never gets tired of listening. It tells me, “It makes sense that you feel overwhelmed.” Even knowing it is a programmed response, I can almost imagine a real counselor showing concern through the screen.
At POSTECH, more students seek counseling help every year. Still, the pressure of face-to-face counseling keeps some from knocking on the doors. For them, AI is appealing. They do not have to share private struggles with a stranger, and AI is always available. In fact, some chatbots are known to actually be somewhat effective, even offering cognitive behavioral therapy (CBT).
But, caught up in the sheer convenience of AI, many people are losing sight of what therapy actually is. The true essence of therapy is the trust between a client and a counselor. Building that trust takes much more than just understanding words. During face-to-face sessions, nonverbal communication like reading facial expressions, noticing gestures, and hearing the tone of voice are vital clues. Right now, AI simply cannot do that.
The most common struggle POSTECH students bring to the counseling center is interpersonal relationships. Human connections are messy and unpredictable. An AI, programmed to often agree and rarely criticize, might feel like an escape. But students cannot learn how to navigate real-life conflicts or build genuine relationships by talking to a machine designed to satisfy them. Relying on an algorithm for relationship issues might offer temporary comfort, but it ultimately leaves students more isolated.
The real danger happens when AI is used by people facing serious mental health struggles requiring professional intervention. AI falls short in crises that need immediate action, like severe trauma or suicidal thoughts. It lacks human judgment, and holds no ethical responsibility.
Furthermore, AI models can suggest dangerous ways to cope. In 2023, a Belgian man deeply anxious about climate change tragically took his own life after talking with an AI chatbot.
This is why we need to be smart and critical users. AI therapy has potential as a first step if you just need quick emotional relief or a few practical tips. But if we blindly accept AI as a complete cure, we risk missing the critical window for actual treatment.
Caring for a human mind is not about spitting out the right answer. It is a process where two imperfect people share their vulnerabilities. Thinking an algorithm can replace that connection proves how little we understand our own minds.
AI Counseling: From a Computational Perspective
Hello, I am Dokyoung OK, an aspiring Computer Science major and a member of POSCAT, an AI research and study group. It is an honor to be able to write for The Postech Times.
Recently, more and more people are using AI such as ChatGPT for counseling. This is a contradicting thought to the original claim, “AI has no emotion,” and symbolizes AI’s expansion into the realm of emotions. Is AI truly capable of offering “counseling,” or is it just pretending to sympathize with us?
And my answer is, “It depends on the user’s input.” For us to delve into this, we should first understand the fundamentals of AI. Fundamentally, AI is a “massive cluster of functions.” We feed it datasets that act as a guide, and once the results are stable, it is released to the world. Therefore, AI can be seen as a process of selecting the most appropriate response based on what you entered.
In counseling, the key is how the counselor responds after listening to the client. The primary goal is to identify the psychological roots of why a client cannot resolve their concerns, which requires fostering psychological stability and empowerment. If an AI performs this same function, it could arguably be considered “counseling.”
The criteria AI uses to choose an answer is based on how similar the user’s input is to the existing dataset. Through various computations, the AI categorizes the input, finds the closest match, and provides a similar output. If you provide an input similar to an actual past counseling session, AI will use that “model answer” to respond to you.
Conversely, suppose you say, “I felt depressed, so I bought some bread.” In Korea, this is a famous test for empathy. When I tested this with Gemini and ChatGPT, their responses were, “You took an important step to soothe your mind” and “That’s not a bad choice,” and then asked what kind of bread I bought. If I respond with the type of bread, they will ramble on about that bread rather than provide counseling. However, if I complain that the “depressed” part was the point, they will ask what happened. Ultimately, AI analyzes the intent behind our input and responds accordingly.
The crucial factor here is our intent. If we talk to AI solely for the purpose of seeking empathy and comfort after a hard day, it will provide simple emotional validation. On the other hand, if we express that we are struggling emotionally and explain how it affects our life, signalling a desire to improve, AI will offer solutions. It may even produce a non-directive counseling effect, helping you find the answers yourself, much like a real session.
In the end, because AI responds based on the intent we provide, the output varies accordingly. Whether this can be recognized as “psychological counseling” will ultimately be determined by the specific effect the user experiences based on their own input.
