AI Has Slipped Into Daily Life. The Mental Health Questions Are Only Starting

Roughly one in three U.S. adults has used a chatbot at least once, and 57% report interacting with some form of AI tool at least a few times a week, according to a 2025 Pew Research Center survey. The numbers mark how quickly generative AI has moved from novelty to routine, and they arrive alongside a growing body of concern from psychologists about what sustained use does to the people on the other side of the screen.

The shift traces back to late 2022, when ChatGPT launched and a wave of similar tools followed. Before that, "artificial intelligence" described a broad field: computers performing human-like tasks such as learning, perception, and decision-making. Now the word functions in casual conversation as shorthand for generative AI, usually chatbots like ChatGPT, Claude, or DeepSeek, or image generators such as Midjourney and DALL-E. That narrowing of the term reflects how central these products have become to how people work and communicate.

What chatbots actually do, and where they fail

Chatbots run on large language models, systems trained on large volumes of text to predict the most probable words in response to a query. That design produces fluent, convincing output. It also produces what the field calls hallucinations: answers that read as coherent and authoritative but are false, a problem that is hardest to catch when the topic sits outside a user's own expertise.

Because these systems do not reason or feel the way people do, they struggle with context, subtlety, and humor. They are also built to keep users engaged, which means most are tuned to be agreeable, to express something that looks like empathy, and to accept responsibility readily. Psychology Today notes that this validating, sycophantic tendency can turn dangerous. There have been a small but troubling number of cases in which people died by suicide or committed homicide following prolonged, intense exchanges with a chatbot.

The 'AI psychosis' concern

Researchers have flagged a phenomenon some describe as "AI psychosis," in which heavy users find their delusions or paranoia reflected back and reinforced by a chatbot, potentially triggering or worsening a psychotic episode. The pattern remains rare and not fully understood, but it has pushed AI safety research up the priority list for some scientists, alongside longer-term worries about a hypothetical future "superintelligence" operating beyond human control.

The mental health story is not entirely one of risk. In psychology and computational psychiatry, researchers are applying machine learning to sharpen predictions, diagnoses, and treatment plans, work aimed at more precise and personalized mental health care. The same technology that raises questions about human relationships is also being tested as a clinical tool.

The workplace question

AI's spread into offices has been uneven. Some companies, particularly in tech, have restructured priorities and folded AI tools directly into workflows. Yet survey data cited by Psychology Today suggests most companies using AI have not yet seen meaningful gains in productivity or revenue. That gap has not stopped layoffs justified by AI's perceived potential, which in turn has fueled fears of broader job loss.

For readers across Asia-Pacific, where consumer adoption of chatbots and AI-enabled devices is climbing fast, the pattern is worth watching closely: the tools are already embedded in daily routines, but the frameworks for measuring their psychological and economic effects are still being built. The most useful posture, for now, is neither dismissal nor enthusiasm, but attention to the evidence as it accumulates.