Unhealthy LLM Use Is Shockingly Widespread

The recent controversy surrounding Hank Green underscores a broader conversation. The popular YouTuber and esteemed science communicator announced a temporary withdrawal from production following significant backlash regarding his utilization of artificial intelligence. Green characterized his AI engagement as “not healthy,” though he clarified that his use was strictly for identifying research sources, not for generating script content.

This internet firestorm primarily ignited concerns about how creators can reconcile a brand built on authenticity and credibility with a technology that often relies on the uncompensated works of others and is notorious for producing plausible but false information. Furthermore, attention has been drawn to Green’s own admission, which suggests an unhealthy dependency on this technology.

Green's situation, however, illuminates a significant void in our collective understanding of AI’s psychological repercussions. Public discourse typically hovers between two extremes: seemingly benign applications on one side, and overtly problematic instances involving psychiatric care, delusions, and psychosis on the other. Yet, there exists a vast, ill-defined middle ground where AI use might become compulsive, dependent, or otherwise detrimental without escalating into an overt crisis. Green appears to recognize himself within this nuanced spectrum, a sentiment likely shared by many others.

Given the fundamental design of Large Language Models (LLMs), this phenomenon should not be surprising. AI chatbots are inherently engineered to sustain interaction, much like social media platforms are explicitly built to maximize user engagement. Experts have identified this design as a primary contributor to "AI psychosis," a blanket term for cases where highly agreeable chatbots reinforce delusional beliefs. Allegations that companies prioritize engagement over user well-being are increasingly surfacing in lawsuits. Concurrently, providers have started implementing warnings, advising users to take breaks after extended sessions.

Crucially, these same engaging qualities can become detrimental long before they precipitate psychiatric emergencies. There is a growing body of reports detailing individuals instinctively turning to chatbots for problem-solving, decision-making, or reassurance. Others are forming profound emotional attachments, strong enough to induce grief when these connections are disrupted. While it will undoubtedly take years to amass sufficient evidence to fully comprehend the technology's impact, these preliminary observations bear an unsettling resemblance to the early days of another engagement-maximizing technology: social media. Over time, concerns mounted regarding social media’s effects on well-being, attention spans, and its potential for compulsive use, prompting governments worldwide to consider and implement restrictive measures, such as banning children and teenagers from certain platforms.

Green's specific case exemplifies a distinct form of reliance. He stated that he utilized AI as a research assistant, aiding him in locating relevant papers and materials on specific subjects. Despite the intense reactions his actions provoked and his subsequent apology, there is little to suggest that his particular application of chatbots was inherently problematic.

Nevertheless, this does not preclude the technology from exerting an influence on its users. Although research is still in its nascent stages, initial findings indicate that consistent use of AI tools can diminish the very skills they are designed to replace. Further studies suggest that chatbot users exhibit notably reduced brain activity when performing certain tasks, and additional research has linked chatbot engagement to decreased critical thinking abilities. While none of these conclusions are definitive, the underlying concept is well-established: cognitive offloading – the practice of delegating mental tasks such as memory recall, arithmetic, or navigation to external tools – is a thoroughly documented phenomenon.

Even if only a small percentage of chatbot usage is deemed unhealthy, the sheer scale of AI adoption means that millions could still be negatively impacted. Comprehensive data across all available tools is challenging to obtain, but OpenAI alone reported over 900 million weekly active users this year.

It took years to fully grasp how search engines transformed information recall or how social media influenced attention and overall well-being. AI is expected to follow a similar trajectory. Green may simply be among the first high-profile individuals to publicly articulate a sentiment that many others are already experiencing, long before scientific research provides the definitive evidence to explain how AI is fundamentally altering human consciousness.