When AI Feels Human: Why We Can Experience a Chatbot as a Real Therapist
Part four in a psychosexual therapist's exploration of AI and mental health care
Have you ever thanked a chatbot, worried you had offended it, or felt surprisingly understood by its reply? I have!
These reactions can feel silly or humorous, but they are not signs of our gullibility. They reflect something deeply human: we are wired to scan for signs of relationship. When something uses language, remembers details, responds warmly and appears interested in us, our instincts naturally come into play and we may try to engage with it, as if it were human.
It's natural to anthropomorphise
We routinely attribute human thoughts, feelings and intentions to non-human things. We name cars, apologise to the furniture we bump into, talk to pets and swear at electronic devices when they are 'being difficult'. Psychologists call this anthropomorphism — the habit of treating non-human things as if they had a mind.
This isn't a flaw. It's how we survive...and connect. We detect another's intentions by picking up clues, or 'cues'. It's a fast way to predict what someone will do next. Anything that looks and sounds like a person gets treated like one. But cues are not facts. And AI is now extraordinarily good at producing exactly those cues, even if in fact, it isn't human.
Why AI trips that wire so easily
Back in the 1990s, researchers Clifford Nass and Byron Reeves ran a series of experiments showing something surprising: people apply social rules to computers. They're polite to them, worry about hurting their feelings, and judge them as if they had personalities — even when they know perfectly well it's just a machine. Nass and Reeves called this "Computers Are Social Actors." Their conclusion was that our responses to media are "fundamentally social and natural" — we can't easily switch the social brain off. ACM
It doesn't take much for AI to trigger this:
It talks to us like a human. It says "I" and "you," creating the grammar of a relationship. It says "us", giving the sense that it shares our human experience.
It answers instantly, so the conversation feels like a back-and-forth, not a search.
It remembers what you said earlier and refers back to it, which feels uncannily like being listened to.
It mirrors your tone and language, which is exactly how we signal closeness to each other.
It never gets bored, distracted, or impatient with you.
Each of these is a small social signal. Together they add up to something that feels like being in the presence of another person.
The ELIZA effect & the illusion of feeling heard
In the 1960s, a computer scientist named Joseph Weizenbaum built a simple program called ELIZA that imitated a therapist. It worked by spotting keywords and turning a person's statements back into questions. That was basically it — no understanding, no memory of you as a person. Yet, in doing so, some users felt that ELIZA understood them. https://dl.acm.org/doi/epdf/10.1145/365153.365168
The ELIZA effect is the name for our tendency to read real understanding into a machine's words because the words sound like understanding. Modern AI strengthens this effect because its replies are more fluent, personalised and specific to the individual's story. That doesn't make the effect weaker — it makes it far more convincing.
AI chatbots vs a real therapist?
Talking to AI may sometimes feel safer than talking to another person. It does not visibly flinch, interrupt, become embarrassed or reveal our secrets. This can be particularly relevant where shame, fear of judgement and uncertainty about what is "normal" may make disclosure difficult. This anonymity, availability and empathic design appear to help people develop what researchers call a digital therapeutic alliance with AI. JMIR Mental Health
Our perception of an AI chatbot as warm or empathic can support disclosure, but it can also increase user trust beyond what the system's reliability and safety warrants disclosing. Read more in my previous blog: The Pro's and Con's of AI in Mental Health
Staying curious and aware of our 'relationship' with AI chatbots
Anthropomorphising AI is not automatically unhealthy. Concerns may arise when the system becomes a person's primary relationship, discourages human contact or gains disproportionate influence over important decisions. The feelings we experience in working with an AI chatbot may be real, even if the experience is not a human relationship.
It may help to ask ourselves:
For what things do I feel the AI chatbot is reliably capable of providing?
How can I keep in mind an AI chatbot's usefulness as a tool versus belief in it as an empathic mind?
What need is the use of the chatbot meeting for me?
Does it support or replace human relationships?
Do I understand how my information may be stored or used?
Can I disagree with the system and step away comfortably?
Understanding our natural inclinations to anthropomorphise and the programming of AI chatbots to enhance our ability to do so allows clients and therapists to discuss AI relationships without ridicule or alarm. AI can offer language, structure and a sense of immediate attention. What it cannot currently offer is reciprocal care, professional responsibility, or a collaborative human relationship.
Comments