Artificial intelligence chatbots could help people better understand and manage their mental health, but they could also reinforce harmful thoughts and increase emotional dependence, a Curtin University researcher warns.
Published in Digital Health, the paper compares AI conversations that follow two possible paths: one that helps people explore their problems and consider new perspectives, and another that traps them in a cycle of repeated thoughts and reassurance.
Author and Professor of Mental Health Warren Mansell, from the Curtin School of Population Health, has proposed a new framework to explain why AI conversations can have very different effects on people depending on their emotional state, what they hope to gain from the interaction and how the chatbot responds. Professor Mansell was the first to evaluate a therapy chatbot, back in 2013, and his research has focused on trying to measure and understand how a person's attention either narrows or broadens during a conversation.
Professor Mansell said chatbots could be helpful when they helped people to explore their feelings and question their assumptions, rather than simply agreeing with everything they said.
"AI conversations can move in very different directions depending on what a person brings to the interaction and how the chatbot responds," Professor Mansell said.
"For some people, a chatbot may provide a space to express their feelings and explore problems without fear of judgement. For others, repeatedly receiving reassurance or having their existing beliefs confirmed could make it harder to see things differently."
Research into AI and mental health has produced contrasting results. Some studies suggest chatbots can reduce symptoms of depression and anxiety, while others have raised concerns about harmful responses, emotional dependence and AI reinforcing users' existing beliefs, even when those beliefs could be damaging.
Professor Mansell said the same chatbot could potentially help one person while having a negative effect on another or even affect the same person differently depending on how they use it over time.
For example, someone experiencing anxiety might use a chatbot to explore their worries and consider alternative ways of looking at a situation. However, if the conversation repeatedly focuses on seeking reassurance or confirming their fears, it could narrow their thinking rather than help them work through the problem.
The paper suggests the risks could be particularly concerning for people who are already vulnerable and most in need of psychological support.
While AI chatbots are often accessible, inexpensive and available at any time, Professor Mansell argues their benefits may depend on whether they encourage meaningful reflection or simply provide immediate reassurance to keep users engaged.
He said developers should design chatbots that encourage people to think through their problems, ask thoughtful questions and consider different perspectives, rather than simply telling users what they want to hear.
The proposed framework also highlights the importance of understanding how conversations develop over time, rather than judging a chatbot's effectiveness solely on whether users feel better immediately after an interaction.
Professor Mansell recommends people approach AI conversations with a clear purpose and consider how their use of chatbots fits alongside other priorities and relationships in their lives.
The proposed framework is yet to be tested, and further research is needed to establish why some AI conversations may support mental health while others could increase psychological distress.
As AI becomes an increasingly accessible source of mental health support, Professor Mansell said understanding these differences would be important for developing safer and more effective technology.
The paper 'Up the mountain or down the rabbit hole? A simple rubric to conceptualise the benefits and risks of AI conversations as dynamic changes in awareness' was published in Digital Health.
