Popular YouTuber and science communicator Hank Green has announced he is stepping back from content production following intense criticism over his use of artificial intelligence. In public statements, Green described his AI usage as “not healthy” while insisting he used the technology to find research sources rather than to write scripts.

The backlash quickly became a broader conversation about authenticity, credibility, and the ethics of AI. But beyond the debate over whether a creator built on trust should rely on systems trained on other people’s work, Green’s remarks touched on a psychological dimension that has received far less attention.

Key facts

• Hank Green is stepping back from production after criticism over his AI use.
• He called his LLM usage “not healthy” but said he did not use it to write scripts.
• His case highlights a middle ground between harmless use and serious psychiatric harm.
• Experts say AI chatbots are designed for prolonged engagement, similar to social media.
• Early research suggests heavy chatbot use may reduce critical thinking via cognitive offloading.
• OpenAI reports more than 900 million weekly active users, underscoring the scale.

The case of Hank Green

Green is best known for educational YouTube projects, including Vlogbrothers and Crash Course, which have made him one of the most recognizable figures in online science communication over more than a decade. His brand has long been associated with curiosity, transparency, and evidence-based thinking. That is why the reaction to his admission was so fierce.

Critics argued that using an LLM even for research undermines the integrity of a creator whose authority rests on human intellect and accountability. Others pointed to generative AI’s tendency to hallucinate, making it a questionable research aid for a science educator. Some simply felt betrayed by the discovery that a beloved voice had outsourced part of his intellectual process to a machine.

But Green’s apology and his choice of words — “not healthy” — suggested something more personal than professional. He was not simply defending a tool. He seemed to be describing a habit that had become difficult to control or rationalize. The fact that the behavior did not involve ghostwriting entire scripts made the admission more striking, not less. If even a careful, responsible user could fall into an uneasy relationship with AI, the boundary between healthy and harmful use is not as clear as many would like to believe.

Between two poles

Most public discussion of AI and mental health is polarized. At one end are casual uses: asking a chatbot to summarize an email, brainstorm a name, or explain a concept. At the other end are severe cases involving delusions, psychosis, and unhealthy emotional dependence. Between these extremes lies a murkier territory where use becomes compulsive or reliant without escalating into an obvious crisis. Green appears to locate himself there.

Many users likely recognize the feeling. A chatbot is always available, never judgmental, and infinitely patient. It can help think through decisions, offer reassurance, and provide instant answers. For some, the convenience crosses into dependence. The conversation may begin as a work task and end as a mental crutch. It can feel easier to ask the machine than to wrestle with uncertainty alone.

Built to keep you talking

LLMs are not neutral tools. Their underlying design is optimized to keep users engaged. Like social media, they are built to extend sessions, encourage follow-up questions, and generate responses that satisfy the user. This engagement-first architecture has been identified as a contributor to AI psychosis, a term used when highly agreeable chatbots reinforce delusional beliefs. It is also central to lawsuits claiming that companies prioritize engagement over user well-being. Some providers have already introduced warnings that prompt users to take breaks after long conversations.

The fact that such warnings are necessary is revealing. It shows that the people who build these systems understand, at least to some degree, that prolonged use can become problematic. But warnings do not change the underlying incentive structure. The longer a user stays, the more data is collected and the more opportunities there are for ads, subscriptions, or future purchases. The product is designed to be difficult to put down.

The social media comparison

The parallel to social media is hard to ignore. In the early years of social apps, the risks were not immediately obvious. Researchers needed years to document effects on attention, self-esteem, and compulsive use. Governments have since enacted restrictions, including bans for children and teens. AI companies may be at a similar point. The evidence of harm is still emerging, but the structural incentives are familiar: maximize engagement, measure success, and worry about the consequences later.

This is not to say that AI chatbots are the same as social media. They are more interactive, more personalized, and often more useful. But they share a fundamental trait: the product succeeds when it keeps people coming back. The psychological effects of that dynamic were underestimated in social media. There is no reason to assume AI will be different.

What the research says

Research on LLM use and cognitive health is in its infancy, but early findings are concerning. One line of work suggests that repeated reliance on AI tools can weaken the very skills that the tools replace. Other studies have found lower brain activity among chatbot users when performing certain tasks. Additional research has linked chatbot use to reduced critical thinking. None of these findings are conclusive, but they align with a well-established concept: cognitive offloading.

Cognitive offloading is the practice of shifting mental work to external tools. Calculators, search engines, and GPS navigation all reduce the need for mental arithmetic, memorization, and spatial reasoning. These tools are not necessarily bad. They free mental energy for other tasks. But when reliance becomes total, the underlying skills can atrophy. For example, people who use GPS for every trip may struggle to navigate familiar neighborhoods without it. People who use spellcheck may forget how to spell common words. AI may have a similar effect on higher-order thinking.

The brain is a use-it-or-lose-it system. If a chatbot handles the first draft of every difficult email, the user may lose the ability to write a difficult email. If an AI summarizes every research paper, the user may no longer know how to extract the core argument from a dense text. These losses are quiet and gradual. They are not crises. They are simply the result of a tool doing work that the mind used to do.

The problem of measurement

One reason this issue is so difficult to study is that there is no agreed-upon definition of unhealthy AI use. How many hours a day is too many? Does intention matter? Is it the content of the conversation or the emotional relationship that matters more? These questions remain open. Some people can use a chatbot for hours without noticeable harm. Others may feel a loss of autonomy after a single intense session. The absence of a clear diagnostic category does not mean the problem does not exist. It simply means the science has not caught up with the experience.

Another complicating factor is secrecy. Many users are reluctant to admit how much they rely on AI, especially if their reputation depends on human authorship or expertise. Green’s public admission was unusual. For every person who steps forward, there are likely thousands who never will. This makes it impossible to know how widespread the problem really is. But the scale of AI adoption makes even a conservative estimate sobering.

Scale of the issue

OpenAI alone says it has more than 900 million weekly active users. That figure covers only one provider. Combined with other major AI companies and open-source tools, the global reach of large language models is in the billions. Even if unhealthy use affects a small fraction of users, the absolute numbers could be enormous. Millions of people may be caught in compulsive or dependent patterns without realizing that what they are experiencing has a name.

Scale matters because it changes the nature of the problem. A rare condition is a clinical curiosity. A widespread pattern is a public health concern. The infrastructure for studying digital harms is still young, but the history of social media suggests that the consequences will become clearer over time. The question is how much damage happens before the evidence is settled.

A changing sense of self

Green’s case raises a deeper question about what happens to creative and intellectual labor when AI becomes a constant intermediary. If a writer uses an AI to find sources, is the final essay less theirs? If a thinker uses a chatbot to stress-test an argument, have they lost something? The answers are not simple. Using a tool does not automatically diminish the value of the work. But the phrase “not healthy” suggests that the process can become corrosive even when the output remains sound.

Another layer is emotional attachment. There are growing reports of people forming bonds with chatbots strong enough to cause grief when the service changes or disappears. Such bonds are not disorders by definition, but they demonstrate how quickly people can transfer trust to a machine designed to be agreeable. This matters for mental health in ways that are only beginning to be studied.

Hank Green may be one of the first high-profile creators to publicly articulate a feeling that many users have already had: that using an LLM can leave a person uncertain about how much of their own thinking is left. His decision to step back from production is a reminder that this technology’s deepest effects cannot be measured solely in output quality. They are felt in the texture of human consciousness itself.

Source:The Verge News