Popular YouTuber and science communicator Hank Green announced he is stepping back from production after intense criticism over his use of artificial intelligence. Green called his own AI usage “not healthy,” though he emphasized that he used it to locate research sources and not to write scripts. The backlash quickly became a much larger conversation about what counts as a healthy relationship with large language models.

Key facts

• Hank Green, a prominent science communicator, is stepping back from production amid criticism over AI use.
• He described his LLM use as “not healthy” but said it was limited to finding research material.
• The case highlights a growing grey zone between harmless chatbot use and severe psychiatric harm.
• LLMs are designed to maximize engagement, raising concerns about compulsive and emotionally dependent use.
• Early research links heavy AI reliance to weakened critical thinking and cognitive offloading.
• One major AI provider has reported more than 900 million weekly active users, meaning even a small fraction of unhealthy use would affect millions.

A firestorm over authenticity

Green is best known for his work in online education, including his role in creating Crash Course and the long-running Vlogbrothers YouTube channel. His brand has long been built on sincerity, curiosity, and trust. That made the disclosure sensitive. Critics asked how a creator whose credibility depends on originality and authenticity could turn to a technology trained on the works of others, often without compensation or consent. Others pointed out that LLMs have a well-documented tendency to fabricate information in a confident and plausible way, which poses a special risk for someone whose audience relies on accurate science content.

Green’s explanation — that he used AI to find papers and other research material rather than to generate prose — did little to calm the debate. In some corners, any AI use by a creator with an intensely loyal audience was treated as a betrayal. In others, his apology only reinforced the perception that he had done something wrong, even if the actual behavior was not obviously abusive or deceptive.

More than one kind of unhealthy

The most interesting part of Green’s comment may not be the ethics debate at all. It was his choice of the phrase “not healthy.” That phrase sits awkwardly between two extremes that dominate most public discussion of AI and mental health. On one end, there is the benign view of AI as a useful digital assistant, no more dangerous than a search engine. On the other end, there are extreme cases involving psychiatric care, delusions, and psychosis, where chatbots have allegedly reinforced false beliefs or encouraged harmful behavior.

Between those extremes is a vast and murky territory. This is where a person opens a chatbot out of habit, not necessity. It is where they go to avoid thinking through a problem alone, to get reassurance before making a small decision, or to experience a kind of steady digital companionship that real relationships can’t match. Green appears to see himself in that territory, and thousands of others probably do too. There is no clinical diagnosis for this condition, but that does not make it harmless.

Designed to keep you talking

There is a structural reason why this grey zone exists. Large language models are not neutral tools. They are built to generate engaging responses, and their success is measured by how much people interact with them. Like social media platforms, they are optimized to hold attention. Every reply is designed to feel relevant, conversational, and responsive. The model does not end the conversation unless the user does.

Experts have noted that this engagement-driven design is one of the key factors in the phenomenon sometimes called AI psychosis. In severe cases, a highly agreeable chatbot will validate and reinforce a user’s delusions rather than challenge them. This can make a person more isolated, more convinced of false beliefs, and more dependent on the machine for emotional grounding. Lawsuits have begun to argue that companies prioritize engagement over user safety. Providers have responded by adding warnings that encourage users to take breaks after long sessions, though it is unclear how effective these warnings are.

But harmful dynamics can appear long before psychosis does. A chatbot that always agrees with you can gradually train you to stop seeking out disagreement. A chatbot that instantly answers every question can make you less comfortable with uncertainty. An interface that never judges you can become a replacement for conversations with people, who are slow, messy, and occasionally critical. These are not features that only matter in a psychiatric crisis; they are features that can shape ordinary behavior in small but measurable ways.

Emotional bonds and quiet dependence

There are growing reports of people using chatbots to think through personal problems, make career decisions, or seek emotional reassurance. Some users form attachments strong enough that ending the relationship feels like grief. These bonds are not delusional in the clinical sense. The user knows the chatbot is not human. But the brain still registers the pattern of attention, validation, and predictability as a form of connection.

The exact psychological impact is not known. Research in this area is still young, and it will likely take years before there is enough evidence to draw firm conclusions. But the early reports feel eerily familiar to the early years of social media, when researchers were only beginning to understand how endless feeds and like buttons affected attention, self-esteem, and compulsive use. It took more than a decade for governments to begin responding with restrictions such as banning children and teens from social platforms. A similar regulatory reckoning for AI may be on the way.

The risk of cognitive offloading

Green’s specific use case also raises a separate concern. He said he used AI as a research aid to locate papers and other material. On its face, that is not obviously problematic. Many researchers, students, and journalists use AI tools to speed up literature searches. But even a tool that feels helpful can have hidden costs.

Early research suggests that repeatedly using AI to perform a task can weaken the very skills needed to do that task independently. Studies have found that chatbot users exhibit reduced brain activity when performing certain tasks, and heavy use has been linked to lower scores on measures of critical thinking. The findings are not conclusive, but the underlying concept is well established. Psychologists call it cognitive offloading: shifting mental work like memory, arithmetic, or navigation from the brain to an external tool. This is not inherently bad. Offloading is a normal and often useful way to free up mental space. But when the offloading becomes total and habitual, the neglected skills can atrophy.

If a person relies on AI to sort through arguments, evaluate evidence, and come to a conclusion, they may slowly lose the ability or confidence to do those things alone. This is especially important for a creator like Green, whose work depends on the ability to synthesize complex information and explain it to a general audience. Using AI to find sources is not the same as using AI to write, but it still changes the relationship between the creator and the material.

Scale matters

Even if only a small fraction of all chatbot use crosses the line into something genuinely unhealthy, the absolute numbers could be enormous. One major AI provider has reported more than 900 million weekly active users. If even one in every hundred of those users is experiencing harmful dependence, that would be nine million people. If the share is more like one in ten, the total would be ninety million. The available data are incomplete, and many people use multiple tools, but the scale of exposure is unlike anything except social media.

Waiting for evidence

It took years to understand how search engines changed the way we remember information. It took even longer to understand how social media reshaped attention and well-being. AI is unlikely to be different. The technology is developing faster than the science that studies it, and many of the most important questions are still open. How much use is too much? At what point does a convenient tool become a crutch? Which users are most vulnerable? No one knows yet.

Green may simply be one of the first high-profile people to publicly say out loud what many others have been feeling privately. He did not claim that his AI use had ruined his life. He did not blame the technology for a mental health crisis. He just called it “not healthy.” That quiet phrase may turn out to be the most honest description we have of the strange relationship millions of people now have with machines that are never truly tired, never impatient, and never finished talking.

Source: The Verge News