文艺批评

齐泽克理论著作《连线大脑里的黑格尔》英文版封面

关于聊天机器人可能会导致传统学生论文（essay）终结的话，已经说了很多。但是，一个更值得关注的问题是，当人类对话者使用侵略性、性别歧视或种族歧视的言辞，促使机器人展现自己满嘴脏话的幻想作为回应时，聊天机器人应该如何回复？人工智能是否应该被编程，使其回答与对方提出的问题处于同一水准线上？

如果我们决定需要采取某种形式的管控，我们必须确定审查应该进行到何种程度。一些群体认为“有攻击性”的政治立场会被禁止吗？呼吁与约旦河西岸的巴勒斯坦人团结在一起，或者声称以色列是种族隔离国家（美国前总统吉米·卡特曾将这个说法写进一本书的标题）呢？它们会被视为“反犹”言论而遭屏蔽吗？

齐泽克回应批判的文集《齐泽克回击！》，

Bloomsbury, 2023

问题并没有在这里结束。正如艺术家和作家詹姆斯·布里德尔（James Bridle）所警告的，新的人工智能“基于对现有文化的大规模挪用”，相信它们“真的有知识或有意义是非常危险的”。因此，我们也应该非常警惕新的人工智能图像生成器。布里德尔观察到，“在试图理解和复制人类视觉文化整体的过程中，[它们]似乎也再造了我们最晦暗的恐惧（fears）。也许这只是一个迹象，表明这些系统在模仿人类意识方面非常出色，甚至可以深入到潜伏在存在（existence）的深处的惊怖（horror）：我们对于肮脏、死亡和腐败的恐惧。”

但新的人工智能在模拟人类意识方面究竟有多好呢？最近有家酒吧用下面的话宣传一款特价饮料：“花两杯的价格买一杯啤酒，免费获赠第二杯！”对任何人类的一员来说，这显然是个笑话。经典的“买一送一”特价口号被重新设计，从而自我取消。这种玩世不恭的表达会被人称赞为滑稽的诚实，一切的一切还是为了促进销售。聊天机器人能理解其中的含义吗？

“Fuck”也提出了一个类似的难题。尽管它指向的是大多数人都喜欢做的事情（性交），但它也经常带有消极意味（“我们完蛋了！”[We’re fucked!]“去你的吧！[Go fuck yourself!]”）。语言和现实是混乱的。人工智能是否已经准备好去辨别这些差异？

德国诗人海因里希·冯·克莱斯（Heinrich von Kleist）在他1805年的文章《论在言语过程中思想的逐渐形成》（On the Gradual Formation of Thoughts in the Process of Speech，该文首次发表于克莱斯去世后的1878年）中，颠覆了以下常识：一个人在没有清楚要说什么之前，不应该张开嘴巴说话。他写道：“如果一个想法表达得很模糊，那并不意味着这个想法在构思时就是混乱的。相反，很可能那些表达【expressed】得最令人困惑的观念，正是那些思考得最清晰的。”

描述齐泽克公共形象生成的著作《斯拉沃热如何变为齐泽克》，University of Chicago Press，2023

语言与思想之间的关系非常复杂。斯大林在1930年代初的一次演讲中，提出了许多激进措施，用来“发现并毫不留情地打击那些仅在思想上反对集体化的人——是的，我是这个意思，我们应该打击人们的思想”。我们可以肯定地认为，这不是事先准备好的话。在被当时的现场氛围感染后，斯大林立刻意识到他刚才说了什么。但是，他没有收回自己的话，而是决定坚持他的夸张说法。

正如雅克·拉康后来所说，这是一个真理在表述（enunciation）行为中意外显现【原文中为“emerging”，即复杂性科学中“涌现”一词】的例子。路易·阿尔都塞在捕获（prise）和惊讶（surprise）之间的互动中发现了类似的现象。一个突然抓住（“捕获”）某个想法的人会对她所完成的事感到惊讶。我再问一次，有任何聊天机器人能做到这点吗？

论文《大型语言模型的涌现能力》

问题不在于聊天机器人很笨，问题在于它们不够“笨”。问题不在于它们天真（缺乏讽刺和反思），问题在于它们不够天真（不理解天真掩盖着敏锐）。因此，真正的危险不是人们会把聊天机器人误认为是真人，而是与聊天机器人交流会让真人说话像聊天机器人一样——丢失掉所有的细微差别和讽刺，痴迷于准确地说出自己想说的话。

我年轻的时候，有个朋友在一次创伤性经历后去一位精神分析师那接受治疗。这位朋友对于分析师对患者的期待有一种陈腐的看法，所以他在第一次治疗时表演了一些虚假的“自由联想”，谈论他如何憎恨他的父亲，希望父亲死去。分析师的反应非常巧妙：他采取了一种天真的“前弗洛伊德”立场，指责我的朋友不尊重他的父亲（“你怎么能这样谈论让你成为你的那个人？”）。这种假装的天真传达了一个明确的信息：我不相信你的虚假“联想”。聊天机器人能理解这种潜台词吗？

很可能不能，因为这就像罗文·威廉姆斯（Rowan Williams）对陀思妥耶夫斯基《白痴》里的梅什金公爵的解读。根据标准的解读，“白痴”梅什金，是一个圣洁的、“极其善良美好的人”，他因现实世界的残酷和激情而被迫陷入孤立的疯狂状态。但在威廉姆斯激进的重读中，梅什金代表着风暴之眼：尽管他是善良圣洁的，但鉴于他在围绕着他的复杂关系网络中的角色，正是他触发了他所目睹的浩劫和死亡。

《连线大脑里的黑格尔》中译本，

将在今年由西北大学出版社出版

问题不仅仅在于梅什金是一个天真的傻瓜，问题在于他特有的愚钝让他无法意识到他对他人的灾难性影响。他是一个说话像聊天机器人一样的扁平的人【英语原文“He is a flat person who literally talks like a chatbot.”，译文删掉了literally】。他的“好”在于，他像聊天机器人一样，不带讽刺地应对挑战，提供毫无反思的陈词滥调，从字面上理解一切，依赖心智的自动完成（mental auto-complete）而非真正的思想形成（authentic idea-formation）。因此，新的聊天机器人将与各种意识形态拥趸（ideologues）相处得非常好，从今天的“觉醒”（woke）群体到宁愿继续沉睡的“让美国再次伟大”（MAGA）民族主义者。

原文2023年3月23日发表于Project Syndicate，

译文2023年3月30日首发于“远读”公众号

END

英语原文

向上滑动阅览

There is nothing new about “chatbots” that are capable of maintaining a conversation in natural language, understanding a user’s basic intent, and offering responses based on preset rules and data. But the capacity of such chatbots has been dramatically augmented in recent months, leading to handwringing and panic in many circles.

Much has been said about chatbots auguring the end of the traditional student essay. But an issue that warrants closer attention is how chatbots should respond when human interlocutors use aggressive, sexist, or racist remarks to prompt the bot to present its own foul-mouthed fantasies in return. Should AIs be programmed to answer at the same level as the questions that are being posed?

If we decide that some kind of regulation is in order, we must then determine how far the censorship should go. Will political positions that some cohorts deem “offensive” be prohibited? What about expressions of solidarity with West Bank Palestinians, or the claim that Israel is an apartheid state (which former US President Jimmy Carter once put into the title of a book)? Will these be blocked as “anti-Semitic”?

The problem does not end there. As the artist and writer James Bridle warns, the new AIs are “based on the wholesale appropriation of existing culture,” and the belief that they are “actually knowledgeable or meaningful is actively dangerous.” Hence, we should also be very wary of the new AI image generators. “In their attempt to understand and replicate the entirety of human visual culture,” Bridle observes, “[they] seem to have recreated our darkest fears as well. Perhaps this is just a sign that these systems are very good indeed at aping human consciousness, all the way down to the horror that lurks in the depths of existence: our fears of filth, death, and corruption.”

But just how good are the new AIs at approximating human consciousness? Consider the bar that recently advertised a drink special with the following terms: “Buy one beer for the price of two and receive a second beer absolutely free!” To any human, this is obviously a joke. The classic “buy one, get one” special is reformulated to cancel itself out. It is an expression of cynicism that will be appreciated as comic honesty, all to boost sales. Would a chatbot pick up on any of this?2

“Fuck” presents a similar problem. Although it designates something that most people enjoy doing (copulation), it also often acquires a negative valence (“We’re fucked!” “Go fuck yourself!”). Language and reality are messy. Is AI ready to discern such differences?

In his 1805 essay “On the gradual formation of thoughts in the process of speech” (first published posthumously in 1878), the German poet Heinrich von Kleist inverts the common wisdom that one should not open one’s mouth to speak unless one has a clear idea of what to say: “If therefore a thought is expressed in a fuzzy way, then it does not at all follow that this thought was conceived in a confused way. On the contrary, it is quite possible that the ideas that are expressed in the most confusing fashion are the ones that were thought out most clearly.”

The relationship between language and thought is extraordinarily complicated. In a passage from one of Stalin’s speeches from the early 1930s, he proposes radical measures to “detect and fight without mercy even those who oppose collectivization only in their thoughts – yes, I mean this, we should fight even people’s thoughts.” One can safely presume that this passage was not prepared in advance. After getting caught up in the moment, Stalin immediately became aware of what he had just said. But instead of backpedaling, he decided to stick with his hyperbole.1

As Jacques Lacan later put it, this was a case of truth emerging by surprise through the act of enunciation. Louis Althusser identified a similar phenomenon in the interplay between prise and surprise. Someone who suddenly grasps (“prise”) an idea will be surprised by what she has accomplished. Again, can any chatbot do this?

The problem is not that chatbots are stupid; it is that they are not “stupid” enough. It is not that they are naive (missing irony and reflexivity); it is that they are not naive enough (missing when naivety is masking perspicacity). The real danger, then, is not that people will mistake a chatbot for a real person; it is that communicating with chatbots will make real persons talk like chatbots – missing all the nuances and ironies, obsessively saying only precisely what one thinks one wants to say.

When I was younger, a friend went to a psychoanalyst for treatment following a traumatic experience. This friend’s idea of what such analysts expect from their patients was a cliché, so he spent his first session delivering fake “free associations” about how he hated his father and wanted him dead. The analyst’s reaction was ingenious: he adopted a naive “pre-Freudian” stance and reproached my friend for not respecting his father (“How can you talk like that about the person who made you what you are?”). This feigned naivety sent a clear message: I don’t buy your fake “associations.” Would a chatbot be able to pick up on this subtext?

Most likely, it would not, because it is like Rowan Williams’s interpretation of Prince Myshkin in Dostoyevsky’s The Idiot. According to the standard reading, Myshkin, “the idiot,” is a saintly, “positively good and beautiful man” who is driven into isolated madness by the harsh brutalities and passions of the real world. But in Williams’s radical re-reading, Myshkin represents the eye of a storm: good and saintly though he may be, he is the one who triggers the havoc and death that he witnesses, owing to his role in the complex network of relationships around him.1

It is not just that Myshkin is a naive simpleton. It is that his particular kind of obtuseness leaves him unaware of his disastrous effects on others. He is a flat person who literally talks like a chatbot. His “goodness” lies in the fact that, like a chatbot, he reacts to challenges without irony, offering platitudes bereft of any reflexivity, taking everything literally and relying on a mental auto-complete rather than authentic idea-formation. For this reason, the new chatbots will get along very well with ideologues of all stripes, from today’s “woke” crowd to “MAGA” nationalists who prefer to remain asleep.

或许你想看

●文艺批评 | 杨庆祥：与AI的角力——一份诗学和思想实验的提纲

● 文艺批评 | 张历君：时间的政治 ——论鲁迅杂文中的“技术化观视”及其“教导姿态”(上)

● 文艺批评 | 戴锦华：我的“远虑”和“近忧”——技术革命及其后果

● 文艺批评 | 王坤宇：后人类时代的媒介——身体

● 文艺批评 | 张净雨：数字后人类身体的美学建构与意义生成

大时代呼唤真的批评家

转载文章请说明来源，使用编者按也请说明情况

编辑 | 提燈

图源 | 网络返回搜狐，查看更多