Artificial intelligence (AI)-based chatbots are increasingly used as an accessible way to talk about one's mental state. However, as Professor Warren Mansell of the School of Population Health at Curtin University (Australia) notes, such conversations can be beneficial but can also intensify anxious or harmful thoughts and create emotional dependence on the bot. He set out his views in an article published in the journal Digital Health.
Mansell is not a detached observer: back in 2013 he was among the first to evaluate the work of a therapeutic chatbot. In the new work he proposes a framework that explains why a conversation with AI produces different effects. In his view, the result depends on the person's emotional state, on what they expect from the interaction, and on how the chatbot responds.
The discussion concerns two possible paths a conversation can take. In one case the bot helps a person explore a problem and look at it differently. In the other, the conversation gets stuck on repetitive thoughts and the search for reassurance, which can keep a person in a closed loop.
Mansell stresses that what helps is not unconditional agreement with the user but assistance in exploring feelings and testing one's own assumptions. "For some, a chatbot can become a space where they can speak out and make sense of problems without fear of judgment. For others, repeated reassurance or confirmation of existing beliefs can make it harder to look at the situation from a different angle," he explains.
Research in this area yields contradictory results. Some studies show that chatbots can reduce symptoms of depression and anxiety, while others warn about harmful responses, emotional dependence and the fact that AI reinforces the user's existing beliefs, even if they are dangerous.
According to Mansell, the same bot can help one person and harm another. Moreover, the effect for the same user can change over time depending on how they conduct the dialogue. Example: a person with anxiety uses a chatbot to think through worries and consider alternative views of the situation. But if the conversation again and again focuses on seeking reassurance or confirming fears, thinking is more likely to narrow than to help cope with the problem.
According to the author, such risks are of particular concern for people who are already vulnerable and most in need of psychological support. The accessibility, low cost and round-the-clock operation of chatbots are obvious advantages, but the benefit, Mansell believes, depends on whether they encourage meaningful reflection or simply provide immediate reassurance in order to retain the user.
Developers, the researcher believes, should create bots that help think through problems, ask thoughtful questions and offer different points of view, rather than simply saying what the person wants to hear. In addition, it is important to assess how a conversation develops over time, rather than judging effectiveness only by whether the person felt better immediately after the interaction.
Mansell advises approaching conversations with AI with a clear purpose and taking into account how the use of chatbots fits with other aspects of life and relationships. The framework he proposes has not yet been tested - further research is needed to establish why some conversations with AI support mental health while others can intensify psychological distress (a state of severe stress and emotional tension). Understanding these differences, the author concludes, is important for creating safer and more effective technologies.