警惕 AI 写作的隐性代价

Machine-made prose

The perils of AI writing

Humans should not surrender authorship to large language models

When it comes to the written word, no writer is a neutral party. This newspaper has applauded technological progress for nearly two centuries. It would be churlish to turn Luddite only now, once the wolves of automation have turned up at our own door. And yet. Both writers and readers should be wary of passing the pen to AI.

谈及文字创作，没有哪个作者能置身事外。本报近两个世纪以来始终为技术进步喝彩。自动化的浪潮终于叩响自家大门，此刻如果摇身变成卢德分子，未免太过无礼。但即便如此，作者与读者都应当警惕，不要把笔交到 AI 手中。（Luddite卢德分子；抵触新技术的人（背景：19 世纪英国工人捣毁纺织机器，反对工业自动化；现在泛指抗拒科技进步的人）；the wolves of automation 自动化的洪流（比喻，狼象征来势汹汹的冲击）；And yet. 单独短句，外刊经典写法：但事实并非如此 / 可是）

Admittedly, in some sorts of writing an AI takeover would be welcome: customer-service complaints,say, or technical write-ups where clarity is all. What matters in such cases is disclosing the use of AI and making sure that the human being whose name is on the page still takes responsibility. Writers of all sorts would be doing their readers a disservice if they did not use AI models for research, proofreading and, when appropriate, feedback.

诚然，在某些类型的写作场景里，由 AI 接手是件好事：比如客服投诉回复，或是只求清晰易懂的技术文档。这类场景的关键在于披露 AI 的使用情况，并且确保署名且依然承担相应责任。各类写作如果不利用 AI 做资料检索、校对，或是在恰当场景下用来获取修改建议，便辜负了读者。（write-up 书面记述、报告文稿；do sb a disservice 损害某人利益；辜负某人）

Elsewhere, however--in novels, speeches and even news articles—surrendering human authorship would be a mistake.For the time being, the most popular gripe about AI writing is the clumsiness of the prose itself. That complaint rests on shaky foundations. The Luddites grumbled, once, about the poor workmanship of machine-made stockings. But then the stockings got better. Doubtless, AI sentences will, too.

然后在其他领域，比如小说、演讲和新闻文章，让AI写作将是个错误。眼下，大家对 AI 写作最普遍的吐槽，是其文字本身生硬笨拙，但这项批评站不住脚。当年卢德派也曾抱怨机器织出的长袜做工粗糙，可后来机器织袜的品质不断提升。毫无疑问，AI 生成的文字同样会越写越好。（gripe=grumble 抱怨）

The medium and the message

Any serious case against AI writing must rest, therefore, on the ways that words are not stockings: how they carry meaning and significance beyond what ends up on the page. The act of writing tends to be painful, even for many of those who do it for a living. But much of that pain is really the pain of thinking, of crystallising thoughts in text. Sidestepping that is a sure-fire route to less rigorous, let alone original, human thought. Students who outsource their essays to Claude may lose the habit of reasoning. Politicians who let AI craft their arguments may be less mindful of their consequences.

媒介与讯息（源自传播学大师麦克卢汉经典名句：The medium is the message  媒介即讯息）

因此，任何反对用 AI 写作的严肃论点，都必须立足于文字与长袜的不同之处：文字承载的意义与价值，远不止纸上呈现的内容。写作这件事往往充满煎熬，即便对以文字谋生的许多人也是如此。但这份煎熬，很大程度上是思考的痛苦，将想法凝练成文字的这个过程痛苦。如果绕开这一过程，必然会造成人类思考严谨性下降，更谈不上原创思考。把论文交给 Claude 代笔的学生，可能会失去推理习惯；让 AI 撰写论证内容的政客，或许对其政策后果缺少一些审慎考量。（a sure-fire route to… 势必会走向…… 的途径）

Another big concern is with the consumers of the written word--or readers, as they are better known. AI writing breaks the expectations of a literate culture. Human text takes longer to write than to read. This offers a certain guarantee if not of quality--the world still swims in dire prose--then at least of effort. Someone, somewhere, thought something was worth saying. Economists call that a costly signal.

另一大担忧关乎我们熟悉的读者。AI 写作打破了文字文明长久以来形成的预期。写作花费的时间比阅读更多，因此即便不能保证质量（世上依旧充斥着糟糕的文笔），至少能保证投入了心血。在某个地点、某人认为某件事值得诉诸文字。经济学家将此称为高成本信号。

Whatever French theorists have to say about the death of the author, readers still want to know that a person wrote the words in front of them. Take poetry. Even a few years ago, when AI was less powerful, researchers were already finding that many people, in blind tests of AI  and human-written poetry, preferred the roborhymes. But making sense of a poem relies on the promise that a human wrote it, and thus that it meant something to someone.

不管法国理论家如何论述 “作者之死”（the death of the author 作者之死，罗兰・巴特提出的文学理论），读者潜意识里依然在意创作者是不是人。以诗歌为例：就在几年前，AI 能力尚弱的时候，研究人员就已经发现，在 AI 诗歌和人类诗歌的盲测实验中，不少人更喜欢 AI 生成的机器人诗句。但读懂一首诗，前提在于相信这首诗由人创作 --因而这份文字对某个真实的人而言，承载着某种意义。

Even when AI is not used, it can sully readers’ experiences. They may stumble on a turn of phrase that sounds suspiciously machine made, and give up on reading the rest of the text. Or the proliferation of AI-generated fakery may make them distrust real news. Polluting the information environment like that is what economists call an externality.

即便不使用AI创作，AI 本身也会破坏读者的阅读体验。读者可能读到某句措辞，会疑似机器生成，于是放弃继续阅读全文。大量泛滥的 AI文本内容充斥着现在的阅读环境，读者即便看的是真实新闻也容易产生怀疑。这种AI创作污染讯息的情形，就是经济学家所说的外部性（stumble on 偶然发现；无意间读到；a turn of phrase 措辞，表达方式，文笔）

Technology will remake many of these norms. AI offers immense opportunities, too, many of which no one can yet fully grasp. Yet liberal societies negotiate these momentous changes not by passively surrendering to technologies, but by adapting to them. The printing press set off religious wars in early modern Europe. In the 1930s the radio heralded fascism. Few associate either with such high drama today. Societies are grappling with plenty of other technologies that pose risks, such as fattier foods and more potent illegal drugs. 

技术会重塑诸多社会规范。AI 同样蕴藏巨大机遇，其中很多机遇目前尚无人能完全洞悉。但自由社会应对这类重大变革的方式，不是被动向技术妥协，而是主动适应。印刷机曾在近代早期的欧洲引发宗教战争；20 世纪 30 年代，广播也曾为法西斯主义的兴起推波助澜。如今很少有人会把这两项技术和这般剧烈的动荡联系在一起。社会也一直在努力解决其他诸多存在风险的技术产物，例如高脂食品与药效更强的违禁毒品。

So it would be hubristic to try to chart out exactly how societies ought in future to approach the use of AI in writing. But any writer knows to start with a first draft. This is ours.

因此，如果试图精准描绘出社会未来应对AI 写作的具体方法，不免显得狂妄自大。但任何创作人都懂得先完成初稿。这篇，便是我们的初稿。