AI 会让平庸的品味变成灾难。 在现代历史的大多数时间里，平庸的品味是有“限速器”的。 生产是有成本的。传播需要时间。必须有人愿意署上自己的名字。市场确实会过滤掉大量垃圾，但整体产出速度受制于摩擦。 AI 消除了摩擦，却保留了品味本身。 这两者的结合，才是真正的危险。 生成式模型是在已有大量内容的基础上训练的，然后被奖励去生成在多数人眼中“看起来合理”的东西。它的输出在表面上很专业，却没有任何必须“有意义”的义务。它是无数他人决策的平均值，被剥离了那些决策之所以成立的具体语境。 于是，一种新的平庸诞生了——精致、快速、无限。 你可以在写作中看到这一点。自动补全不断把人推向通用表达、全球可读的句子、安全的形容词。那些标志着真实生活经验的本地细节，会在“建议”中被一点点删掉。久而久之，个人声音会塌缩成同一种“职业化语浆”。 你可以在故事里看到这一点。让模型讲来自 200 个国家的故事，往往只是在同一套情绪模板上更换道具：稳定、怀旧、和解。冲突被打磨掉，没有锋利到可能犯错的东西，也没有鲜活到足以真实存在的东西。 你可以在音乐里看到这一点。当每天可以生成数百万首新曲时，平台就变成了一个带搜索框的垃圾填埋场。问题不再是“如何创作”，而是“如何找到任何值得被关注的东西”。垃圾信息变成策略，洗脑式好听变成漏洞利用。 你可以在电影里看到这一点。数据驱动的娱乐内容，最终会收敛到适合分心观看的作品：清晰的对白、熟悉的叙事弧线、统一的灯光设计，没有任何要求你真正投入注意力的东西。这是一种为“半看不看”而优化的产品，而它反过来训练观众永远半心半意地观看。 但更深层的问题在这里。 平庸的品味不只是生产平庸的作品，它还会改变品味形成的环境本身。 当信息流被平均化的输出填满，参照系就会发生漂移。人们不再意识到缺失了什么。团队不再锻炼判断力，因为机器不断给出“看起来还行”的选项。领导者开始把吞吐量误当成进步，因为仪表盘很热闹、内容持续发布。 真正的灾难发生在上游——发生在感知层面。 在一个拥有无限草稿的世界里，真正稀缺的资源是： * 丢弃的能力 * 感觉到“差一点也不行”的能力 在系统不断试图把一切打磨成安全形态时，保护具体性与锋芒的能力 这，才是品味的本质。 品味，是宁愿让“平均用户”失望，也要取悦“对的人”的决断。品味，是当算法不断把你推向中间值时，仍然坚持一个明确的立场。 AI 让这一点变得更紧迫，而不是更不重要。 真正会赢的团队，会把 AI 当作生成器，然后建立一种对筛选极其残酷的编辑文化。他们会减少指标数量，但更加在意那些能保护质量的指标。他们会训练人们识别“雷同”，点名它、删除它、然后重来。 而大多数组织会反其道而行之。他们会庆祝产出数量，却在不知不觉中丢失了方向与灵魂。 AI 不需要是邪恶的，这一切也会发生。它只需要——足够容易。 AI makes mediocre taste catastrophic. For most of modern history, mediocre taste had a governor. It cost something to produce. It took time to distribute. It required a human to sign their name to it. The market filtered plenty of junk, but the throughput was limited by friction. AI removes the friction and keeps the taste. That combination is the danger. Generative models are trained on the bulk of what already exists, then rewarded for producing what looks plausibly “right” to the most people. The output has surface competence and zero obligation to mean anything. It is the average of a million decisions made by other people, stripped of the context that made those decisions work. So you get a new kind of mediocrity. Polished. Fast. Infinite. You can see it in writing. Autocomplete nudges people toward generic phrasing, toward the globally legible sentence, toward the safe adjective. The local detail that signals lived experience gets edited out by suggestion. Over time, voice collapses into the same professional-sounding paste. You can see it in story. Ask for a tale from 200 countries and you often get the same emotional template with different props. Stability, nostalgia, reconciliation. Conflict sanded down. Nothing sharp enough to be wrong, nothing alive enough to be true. You can see it in music. When millions of new tracks can be generated daily, the platform becomes a landfill with a search bar. The problem stops being creation. The problem becomes finding anything that deserves attention. Spam becomes strategy. Catchiness becomes an exploit. You can see it in film. Data-driven entertainment converges on content that can survive distracted viewing. Clear dialogue, familiar arcs, uniform lighting, nothing that asks you to lean in. It is a product optimized for being half-watched, which trains an audience to keep half-watching. Here’s the deeper issue. Mediocre taste does not just produce mediocre artifacts. It changes the environment that taste is formed inside. If feeds are filled with average outputs, the reference points shift. People stop noticing what is missing. Teams stop building muscles for judgment because the machine keeps giving them plausible options. Leaders start confusing throughput for progress because the dashboards look busy and the content keeps shipping. The catastrophe is upstream. It happens in perception. In a world of infinite drafts, the scarce resource is the ability to discard. The ability to feel when “almost good” is still wrong. The ability to protect specificity when the system keeps trying to sand it down into something safe. That is what taste really is. Taste is the decision to disappoint the average user in order to delight the right one. Taste is insisting on a point of view when the algorithm keeps offering you the center. AI makes this more urgent, not less. The winning teams will treat AI as a generator, then build an editing culture that is ruthless about selection. They will measure fewer things and care more about the ones that preserve quality. They will train people to notice sameness, to name it, to delete it, to start again. Most organizations will do the opposite. They will celebrate output and quietly lose the thread. AI does not need to be evil for this to happen. It only needs to be easy.