We all use words or language to describe our experiences to other people, whether it is something external such as describing a recent holiday or a deeply personal internal experience such as telling someone how we are feeling in a moment of grief or loneliness. We share these things in the hope that we will be understood and that the person hearing our words will get our meaning and will “see” us.
It is difficult sometimes to find the right words to express or explain something, and it can leave us feeling frustrated or even hurt if we feel we have been misunderstood. But those moments where we tell another person how we are feeling and they show us that they have truly listened and have understood without judgement can bring feelings of validation so strong that they can be life changing and life affirming. This depth of feeling comes about because our words aren’t just describing our feelings as something separate from ourselves, these words are describing our deepest emotional experience, they symbolise our private internal world.
Imagine then that in a moment of abject grief or deep loneliness we send a text message describing our feelings and get a response that shows us empathy, that communicates acceptance and validation. In that moment we might feel relief that someone else out there understands us and “sees” us. We might feel a little less alone, a little less hopeless.
This experience of validation and of being understood is very meaningful for people; you could argue that counselling or therapy is made up of a series of such experiences, which over time help us to come to terms with difficult emotions, situations and even to bring about real lasting change.
Does it matter then if the validation comes from an AI chatbot and not a person? If we feel we are being understood and empathised with is it any less valid or real if it doesn’t come from another human being? And going further, can AI be used as a therapist?
Before I get into these questions, I think it’s useful to understand how an AI chatbot uses words in an entirely different way from humans. A chatbot is a front-end interface (such as a website or app) that users type into. The chatbot then sends this prompt to the underlying Large Language Model (LLM) which uses a mathematical model to predict which words are most likely to come next based on the vast amount of data it has been trained on. It then sends the generated response back to the chatbot for the user to see. Although there is the appearance of understanding, there is instead a super-advanced prediction system that puts words together based on the information it has been given.
I want to add at this point, that I do use an AI chatbot in non-therapeutic ways. I use it to help me in my hobbies, to help me improve my website and to develop and clarify ideas that I may use in blog posts.
So, does it matter where validation comes from? Can AI be used as a therapist? My personal feeling is that yes, it does matter where validation comes from and I do not believe that chatbots such as ChatGPT, Claude and Gemini and their underlying LLMs can safely replace a human therapist.
The main reason for this comes down to the understanding that the relationship between therapist and client is key. Research has shown that the quality of the therapeutic relationship between therapist and client is the best indicator of successful outcomes for the client. The stronger the relationship the more likely there will be positive outcomes. This relationship is complex, nuanced and most importantly reciprocal. It is built on genuine person-to-person contact and contains within it vulnerability from both parties, affective empathy (where the therapist feels the clients’ feelings with them), and very often transference (where the client unconsciously projects feelings from an earlier relationship into the therapist and then reacts to these feelings) which a counsellor will be trained to help the client work through.
None of these are possible with a chatbot, there is no person-to-person contact because the chatbot isn’t a person with a rich, lived experience to draw on. Whilst a chatbot can display cognitive empathy (it can recognise how you feel), it cannot feel for itself and so cannot use affective empathy. Finally, whilst users can project human emotions, thoughts and motivations into a chatbot, it cannot recognise these and work with them in the way that a trained counsellor can. Based on this, I believe that it is not possible for this hugely important therapeutic relationship to exist between an AI chatbot and a user.
The second reason why I don’t believe that AI chatbots can be used as therapists is the ELIZA effect. It is named after a chatbot called ELIZA which was programmed in 1966 by Joseph Weizenbaum, an MIT Computer scientist, to mimic a Rogerian therapist. It worked by using an algorithm to identify key words in the users input and then rephrase these into a question to respond back to the user. Weizenbaum quickly discovered that users became emotionally connected to the program and unconsciously attributed to it understanding and motivation that were not possible for it to have. The user was describing their internal emotional experience and the empathic seeming responses they received from ELIZA led them to believe that the chatbot was relating to them in the same way even though this was impossible. It was this unconscious belief or perception that the chatbot had feelings, empathy, understanding and even intelligence that came to be known as the ELIZA effect.
It is true that modern LLMs are far more advanced than ELIZA was but that isn’t where the issue lies. The ELIZA effect or the unconscious inclination to anthropomorphise chatbots hasn’t gone away, it is an innate tendency we all share. In fact, the way that AI companies program their chatbots to mimic human conversation and to evoke emotional responses from users can sometimes lead to an emotional dependency that at its worst can lead to exclusion of real relationships with other people, self-harm and suicide as has been alleged in a number of tragic cases.
It is a core tenet of my ethics as a counsellor to put my client’s well-being and safety at the centre of my work. I understand that the anonymity and lack of judgement that users may feel when using an AI chatbot can be positive and they may be used to practise vulnerability or as an interactive journal that can help to organise thoughts or feelings in preparation for a therapy session. However, the risk of emotional dependency without having another human who can help a client work through this means that I do not believe that AI chatbots are a safe replacement for a human therapist.
It is important to me at this point, to recognise that human therapists are not perfect. There have been a number of news reports in the last few years detailing the harm that has been caused to people by those who are unscrupulous, predatory or unqualified and it is vitally important that when choosing a prospective therapist or counsellor we research anyone we are thinking of working with. It helps to use recognised directories and to focus on counsellors who are members of a recognised ethical body such as the BACP or the NCPS.
It is also important to acknowledge that some people may feel that using an AI chatbot might be their only option in terms of finding mental health advice, support or validation. There are a number of potential barriers for people wanting to find support including cost, geographic location, lack of local services, long waiting lists, disability, difficulties with sharing emotions, fear of judgement and shame. Any of these can make it feel impossible to find support outside of an AI chatbot and it is important to recognise these very genuine barriers. If this sounds like you, I realise that my arguments here may not help you when you feel you have very limited choices if any at all. I understand that point of view, and I accept that it is the lived reality for many people. If you are in this position, I can only encourage you to look for local and national charities or services that might be able to help you to work around these barriers.
It is likely that there are many people who do have access to human therapists, but who choose to use chatbots as a therapist instead. If this sounds like you, I wonder how you might feel to talk to a human counsellor rather than the chatbot? What would be your biggest worry about doing this? I wonder how different it might feel to receive validation from a human therapist rather than from an AI one? I don’t know the answers to these questions, the only way to really know would be to make the change to a human counsellor or therapist and to see for yourself. You might be reading this post and feel that I am wrong. Perhaps you are using an AI chatbot as a therapist and feel that you’re happy with it, that you feel better than before you started using it, and for some people that may be enough. But if you are using an AI chatbot as a therapist and you feel better because of it, please remember that feeling better, whilst good, is not always the same as getting better.
I will end this post with a plea: if you are dealing with a difficult situation, if you feel lost or are grieving and you want to talk to someone, please reach out to a qualified counsellor, either directly or through a charity or local organisation. It can be scary to take that step, but it’s the first step in a journey towards being part of a therapeutic relationship that will hopefully help you towards your own positive outcome.
If you would like to know more about starting counselling with me or how we might work together, please get in touch and we can arrange a free 20-minute telephone call.