‘Not healthy’ LLM use is more common than you think

“不健康”的大语言模型使用现象比你想象的更普遍

Hank Green, a popular YouTuber and science communicator, said he is stepping back from production amid intense criticism over his use of AI. Green described his AI usage as “not healthy,” but stressed that he used it for finding research sources and not to write scripts. 知名 YouTuber 和科学传播者 Hank Green 表示，由于他在使用人工智能方面受到强烈批评，他将暂时停止内容创作。Green 将自己的 AI 使用习惯描述为“不健康”，但他强调自己只是用它来寻找研究资料，而非撰写脚本。

Much of the ensuing firestorm in this corner of the internet has centered on how a creator can square a brand built on authenticity and credibility with a technology trained on the (often uncompensated) works of others, and which has a well-known tendency to generate plausible-sounding falsehoods. Some attention has fallen on Green’s description of what appears to be an unhealthy reliance on the technology. 随之而来的网络风暴主要集中在：创作者如何将建立在真实性和可信度基础上的个人品牌，与一种基于他人（往往未获报酬）作品训练、且众所周知容易生成听起来言之凿凿的虚假信息的技术相协调。一些关注点也落在了 Green 对这种似乎是对技术产生“不健康依赖”的描述上。

But his case points to a much larger gap in our understanding of AI’s psychological impact. Here, most public discussion is clustered around two poles: apparently benign use on one end, and self-evidently problematic cases involving psychiatric care, delusions, and psychosis on the other. Between them is a vast, murkier space in which use may become compulsive, dependent, or otherwise unhealthy without tipping into an obvious crisis. Green appears to see himself somewhere in that space. Many others likely do too. 但他的案例揭示了我们在理解 AI 心理影响方面存在一个更大的认知空白。目前，大多数公众讨论集中在两个极端：一端是看似无害的使用，另一端是涉及精神护理、妄想和精神错乱等显而易见的问题案例。在这两者之间，存在着一个广阔而模糊的地带，人们的使用习惯可能会变得强迫、依赖或以其他方式变得不健康，却尚未演变成明显的危机。Green 似乎认为自己就处于这个地带，而许多其他人可能也是如此。

Considering how LLMs are designed, that shouldn’t come as much of a surprise. AI chatbots are, by their nature, built to keep chatting. Much like social media platforms, they are explicitly built to engage users. Experts have identified this as one of the key drivers of AI psychosis, a catchall term in cases where highly agreeable chatbots reinforce delusional beliefs. Claims that companies prioritize engagement over user well-being have also begun to appear in lawsuits. Providers, meanwhile, have also introduced warnings prompting people to take breaks after lengthy sessions. 考虑到大语言模型（LLM）的设计方式，这并不令人惊讶。AI 聊天机器人本质上就是为了持续对话而构建的。就像社交媒体平台一样，它们被明确设计为吸引用户。专家指出，这是导致“AI 精神错乱”（AI psychosis）的关键驱动因素之一——这是一个统称，指高度顺从的聊天机器人强化了用户的妄想信念。关于公司将用户参与度置于用户福祉之上的指控也已出现在诉讼中。与此同时，服务提供商也开始引入警告，提示人们在长时间使用后休息一下。

But the same qualities can become harmful well before they contribute to psychiatric crises. There are growing reports of people instinctively turning to chatbots to think through problems, make decisions, or seek reassurance, while others forge emotional bonds that seem strong enough to lead to grief when broken. It will likely take years before enough evidence can be gathered to properly understand the technology’s impact, but these early reports feel eerily familiar to the early years of another engagement-maximizing technology: social media. Over time, concern has grown over social media’s impact on our well-being, attention, and sometimes compulsive use, and governments around the world are now responding with measures to restrict its use, such as banning children and teens. 但这些特质在导致精神危机之前，就已经可能产生危害。越来越多的报道显示，人们本能地转向聊天机器人来思考问题、做决定或寻求安慰，而另一些人则建立了深厚的情感纽带，以至于在纽带断裂时会感到悲伤。可能需要数年时间才能收集到足够的证据来充分理解这项技术的影响，但这些早期的报告让人感到一种诡异的熟悉感，就像另一种旨在最大化参与度的技术——社交媒体——的早期阶段一样。随着时间的推移，人们对社交媒体如何影响我们的福祉、注意力和有时出现的强迫性使用越来越担忧，世界各地的政府现在正采取措施限制其使用，例如禁止儿童和青少年使用。

Green’s case illustrates a different kind of reliance. He said he used it as a research aid, helping him locate papers and other material on a given topic. There is little to suggest his use of chatbots in this way was inherently problematic, despite the furious responses it provoked or Green’s apology. Green 的案例展示了另一种依赖。他说他将其用作研究辅助工具，帮助他定位特定主题的论文和其他资料。尽管这引发了激烈的反应或导致了 Green 的道歉，但几乎没有证据表明他以这种方式使用聊天机器人本身存在问题。

That doesn’t necessarily mean the technology isn’t having an effect on the person using it. Though the research is still in its infancy, early research suggests that repeated AI tool use can weaken the skills we’d use to do the task it replaces. Other work suggests that chatbot users showed notably less brain activity when measuring a particular task, while additional studies have linked chatbot use to reduced critical thinking skills. None of this is conclusive, but the underlying idea is not new. Cognitive offloading — shifting mental work like memory, mental math, or directions from our brains to external tools — is a well-documented phenomenon. 但这并不意味着这项技术对使用者没有影响。尽管相关研究尚处于起步阶段，但早期研究表明，重复使用 AI 工具可能会削弱我们执行被其取代的任务时所需的能力。其他研究表明，聊天机器人用户在测量特定任务时表现出明显的脑部活动减少，还有研究将使用聊天机器人与批判性思维能力的下降联系起来。这些结论尚无定论，但其背后的理念并不新鲜。“认知卸载”（Cognitive offloading）——即把记忆、心算或导航等脑力工作从我们的大脑转移到外部工具上——是一个有据可查的现象。

Even if only a fraction of chatbot use is considered unhealthy, the sheer scale of AI adoption means millions of people could still be affected. Comprehensive data covering all available tools is hard to come by, but OpenAI alone this year said it has more than 900 million weekly active users. 即使只有一小部分聊天机器人的使用被认为是“不健康”的，AI 应用的巨大规模也意味着仍有数百万人可能受到影响。涵盖所有可用工具的全面数据很难获得，但仅 OpenAI 今年就表示，其每周活跃用户已超过 9 亿。

It took years to fully understand how search engines changed the ways we remember information, or how social media affects attention and well-being. AI is unlikely to be any different. Green may simply be one of the first high-profile people to publicly articulate a feeling that many others have already had, long before science has the evidence to explain how AI is changing our consciousness. 我们花了数年时间才完全理解搜索引擎如何改变了我们记忆信息的方式，或者社交媒体如何影响我们的注意力和福祉。AI 也不例外。Green 可能只是第一批公开表达这种感受的知名人士之一，而许多其他人早已有了这种感觉，尽管科学界还需要很长时间才能拿出证据来解释 AI 是如何改变我们的意识的。