Are we becoming emotionally dependent on AI?
AI chatbots are convincing mimics of human empathy. Users are turning to AI for friendship, support and therapy. But are we prepared for emotional dependence on AI?
Recently I attended a conference, where the panel was asked how artificial intelligence was affecting their industry. One speaker went on a tangent, saying that AI is great for lonely people. She didn’t realise how dystopian this sounded to the audience. What kind of world are we living in where ChatGPT, Gemini and Claude are comforting people who need connection with real human beings?
Bharat Goel wrote on Substack about the Replika incident. The makers of chatbot Replika deleted the code that was repsonsible for its empathy and emotions. Users responded with an outpouring of grief, feeling like they had lost a real relationship. It is a reality that many users will believe AI’s false display of emotions, as we have a tendency to attribute human characteristics to anthropomorphised objects.
More and more, people are turning to ChatGPT as a therapist and confidant. They’re spilling their innermost thoughts to a chatbot because it offers a level of reassurance, that can often be delusional but at least it’s supportive. Is this what AI was built for? In fact, it was. AI chatbots are built to simulate empathy by providing non-judgemental responses and constant validation. They are also programmed to remember information we provide. We feel special when a person remembers something personal about us. When chatbots pull out facts about us saved in code, we can feel the same way without differentiating between human memory and programmed software.
ChatGPT is a language learning model. It’s build to create text content based off a data set it has been provided, pulled from the internet. ChatGPT can copy the real responses of internet users, which can look a lot like empathy, but it’s not genuine. Also, if ChatGPT doesn’t know the answer to your complex emotional or social issue, it won’t tell you that. AI is programmed to ‘hallucinate’ or make something up based on the material it has consumed when it cannot see a clear answer. The thing about AI platforms is that they’re designed to keep you using the platform, rather than providing meaningful life advice and insights.
Obviously, time spent with AI is time you could have spent connecting with people in real life. But the digital world has fractured our communities and support systems. We are connected to more people than ever, but we’re also more unreachable. People are connecting in real life less, and engaging in more social media, leading to parasocial exhaustion. If you only have ChatGPT to turn to, you have a big problem.
While therapy speak has permeated our culture and we are all familiar with concepts like boundaries, therapy remains inconvenient, expensive, and unreachable for many. In Australia, we can get a rebate for up to ten psychology sessions a year. However, you still have to pay the full bill upfront. For many of us, ten sessions a year barely scrapes the surface. I can really see why people turn to ChatGPT for therapy, but it’s so dangerous to let a language learning model shape our minds.
Unlike a therapist, ChatGPT is instantly accessible and available 24/7. When you can’t sleep or have a bad day, you can’t call your therapist. ChatGPT doesn’t cost anything, but it also does not judge. But AI sucks the nuance and human connection out of therapy. It looks like support, but really it’s the equivalent to using Google to reassure your feelings. The famously agreeable ChatGPT won’t push you out of your comfort zone or challenge your behaviour or thought processes.
AI provides a kind of pseudo-intimacy, which tends to rather sycophantic. Chatbot responses appear to be laced with the reciprocity, respect and support that are at the heart of connection with a therapist or a friend. But the connection is one-sided. Researchers have called this ‘interactive parasociality’. Unlike the parasocial relationship you might have with Billie Eillish, chatbots actually respond to you. Yet, they’re still giving you nothing emotionally.
AI clearly has a lot of drawbacks, but in some ways it can foster connection. Users can practice conversations that they might want to have in real life. Of course, AI is no match for the unpredictability of real life relationships. If you’re learning a language, AI provides a free conversation buddy that corrects your sentences. Chatbots might also have a place in providing relief to dementia patients instead of pet therapy, which might not always be safe for animals.
Research has found that moderate levels of voice interactions with ChatGPT can reduce loneliness. However, the same study also linked heavy AI usage with increased feelings of loneliness. For people with limited access to communities and social networks, AI can become a substitute rather than a supplement to human connection.
Regulation of AI has not come to grips with the social ramifications of allowing people with unrestricted access to what is essentially technology companies masquerading as humanity. We can’t expect users to predict how AI could affect them emotionally, and self regulate their use. Researchers have called for AI interfaces to “avoid anthropomorphic deception or the cultivation of false emotional reciprocity.” The reality is that vulnerable people are becoming emotionally dependent on interfaces created by technology companies for profit. Bottling code and calling it intimacy is a harrowing answer to technological snake oil.