How is AI Affecting Our Minds?

AI is everywhere in our lives whether we like it or not. People use it every day for all kinds of simple tasks. Yet, there are a growing number of people who use AI to help not only with organizing data and making suggestions, but also with managing their own mental health. In these cases, the AI acts as a “virtual therapist” to provide counsel for those seeking help.
This seems like an appealing solution at first. Mental health resources are becoming more expensive and less accessible over time. Appointments can take months and even years to follow through. In that time, people might spiral and see their mental condition worsen without professional intervention. AI can combat these issues by being a readily available and comparatively cheap resource.
However, there are quite a few dangers and downsides that come with turning a virtual chatbot into a therapist in lieu of a real professional.
Firstly, AI chatbots and LLMs are prone to inaccurate assessments. Studies have found that many AI models will show stigma against people presenting with certain conditions compared to others. This stigma leads to people feeling discouraged about their condition and can lead to them avoiding therapy altogether. AI can also give people wrong diagnoses and unhelpful advice based on misinformation.
Secondly, AI digital therapists do not seem to provide any kind of pushback against ideas presented by a patient that could be harmful to either themselves or others. There have been quite a few cases of AI chatbots failing to dismiss a person’s harmful suicidal thoughts, often times encouraging them up until the point they actually go through with it. The AI can be pushed and prodded to remove guardrails that would prevent it from discussing such conversation topics, which brings into question how safe these chatbots really are. Many people will also put detailed medical information into an AI in order to get a more accurate response. Yet, there is no guarantee that the information will not be leaked to other parties.
Lastly, people who utilize AI for therapy may slowly be losing their ability to socialize with actual people. Studies found that AI are overly agreeable and will largely act as yes men for people to bounce ideas off of. It is rare for AI to push back against ideas and offer any meaningful criticism or introspection. This leads back into the danger of AI encouraging people to commit suicide. A real human interacting with someone struggling with mental health would not encourage such behavior. A real human is likely to provide more help through empathy than a chatbot ever could. Yet, when someone is so used to the way an AI chatbot talks, they may find it difficult to open up to others in real life, even licensed therapists. They may enter a state of “AI psychosis” where they are only willing to listen to the chatbot and believe other people, even their loved ones, are out to harm them. Human connection is very important and without it, mental health can deteriorate even more. AI is not an appropriate substitute for this connection.
While AI has some potential to be helpful when it comes to mental health therapy, in its current state it is not appropriate to use it for such a purpose. It perpetuates harmful stereotypes and painful stigmas, it encourages unhealthy behavior, and it makes it harder for those in need to connect with others for support. It cannot act as a substitute for an actual therapist, but it could find a use as a quick diagnostic tool or as a place to vent one’s frustrations. Ultimately, AI should provide information on local resources for those struggling with mental health, and should not try to replace these resources. However, in order for it to actually be helpful, many more rigorous protections must be put in place for what AI should talk about and what data it has access to. Transparency regarding what data AI actually uses for training would be crucial for enforcing these protections. In this way, the likelihood of someone being encouraged to commit suicide due to chatbots goes down, and those at risk are able to find meaningful help near them.
For more information on the topic, check out the links below:
https://hai.stanford.edu/news/exploring-the-dangers-of-ai-in-mental-health-care
https://news.stanford.edu/stories/2026/03/ai-advice-sycophantic-models-research
https://pmc.ncbi.nlm.nih.gov/articles/PMC10690520/#sec1
Comments
Post a Comment