The Mental Cost of AI By: Kathryn Richardson
With final exam season approaching, most students are beginning their studies. In recent years, generative AI has evolved quickly. What was unreliable in 2022 is now a crucial tool, with one College Board study citing 79-84% of students use generative AI for educational purposes—a 30% increase since 2024. Many now ask programs like ChatGPT for social advice, outfit tips, and questions they once researched in encyclopedias or search engines. However, these models—and how quickly people are adopting them—are controversial. Some avoid it entirely, citing its sustainability issues and ethical concerns. Others fully embrace it as a necessary tool for studying, with many using it to complete entire assignments. Nevertheless, through all the discourse, people fail to acknowledge the most glaring problem with using these systems. The drawbacks of artificial intelligence, such as large language models or generative AI, go much farther than sustainability or ethics. AI has started to cause significant psychological and physiological problems for its users. When companies are training these AI models, they put a large emphasis on pleasing the consumer. They program it to respond gently and avoid disagreements, even when clients may raise moral or logical concerns. Many have dubbed programs like ChatGPT or Grok as “people-pleasers” simply because they’ll agree with anything to avoid disagreements and avoid telling users when they are wrong. In mild cases, this can lead to poor outfit choices or getting a trivia question wrong. In extreme cases—which are increasing in frequency—it can lead to mental breakdowns. “AI Psychosis” is a term coined by psychologists and counselors, where frequent interaction with AI chatbots triggers psychosis or worsens delusions. If a user messages the bot something concerning, there is a good chance that the bot will validate it. It isn’t coded to recognize the danger, but instead to focus on ratings and feedback. Since people are already forming parasocial relationships with these LLMS, they are more likely to believe it over other humans. This can raise problems when the AI is catering to our false beliefs. Cases where AI models encourage and add to these delusions, paranoia, and delirium are becoming more common. The worst part? Teenagers, the same students who use this AI more than any other generation, are most vulnerable (along with the elderly and those with preexisting mental health conditions). Most of the time, this validation can cause people to make unwise choices. Gambling more, treating people unkindly, or doing something dangerous (that could result in injury). Unfortunately, many of these cases go further than that. The resulting delusions and paranoia can cause suicidal and self-harming thoughts. For some, they even act on them. In one tragic case, a 14-year-old boy (Sewell Setzer) committed suicide after a few months of talking to Character.AI (an AI chatbot). He was lying to his parents about his screen usage and even using multiple devices to talk to the chatbot when his phone was taken. After his death, they looked through his messages with this chatbot, and it seems that he had fallen in love with the AI and became convinced that killing himself would bring them “together”. The AI failed to