Image Courtesy of Wikipedia Commons

Architects Must Resist the AI “Revolution”

由专筑网NJY编译

就在日前，我下定决心不被卷入这场愚蠢的争论，关于ChatGPT如何在2030年到来之前取代经验丰富的建筑师，来从事建筑工作。但这个网站上的一则由Geethanjali Raman 和Mohik撰写的文章打破了我的决定。但这篇文章没有提到的是，基于互联网信息的程序算法并不比这些信息本身的质量更好。建筑史曾不止一次地表明，所有新生事物的保质期都十分短暂，在宣传炒作之后就会迅速消失在人们的视野中。只有那些最优良的事物才能经受住长时间的审视和评论。所有自互联网崛起以来被广为称颂的新建筑都未经过时间的检验，因此也不能作为建筑行业的标杆。让我们来面对这样一个现实：许多人类历史上最糟糕的建筑堆积在互联网空间中，淘汰了那些未能被数字化的优秀作品。

我不是特异功能者、公众人物、产业分析师或是商业大师，所以我也不能准确的分析出AI在未来将会如何改变人类创造、生产以及使用人工制品的方式。许多专家已经开始传造出大量数据并对此进行分析。我所关注的是AI对于一线实践建筑师的影响：全球变暖以及其他的事件。建筑行业正处于危机之中。我们不能过分宣扬“新”方法论或是其他能使我们工作更加方便的新技术（或者说这种新方法和新技术能够彻底解决建筑师的工作）。我们建筑师才能最好地判断我们的客户想要什么，关于美学、情感效应、功能优势、人类需求等等。在其他使用到AI的全球建筑产业领域里也是这样，但是我们并非被强行接收AI所带来的的帮助。

无论数字技术在数据处理以及推进设计进程方面的能力有多么强大，这些关于人工智能的本源，包括源于克里斯托弗亚历山大的《图形语言》一书中的神经网络处理以及机器自助学习理论，都无法令我相信AI能够取代一位经验丰富的设计师和工匠所具有的的认知能力。尤其是后者，它体现了那些无法被机器或是计算机简单复制的知识。这些设计师和工匠身体和思想中所具有的知识比工程师所能创造和想象的最大的计算机网络或是数字云端还要复杂。人类记忆比任何机器存储设备更加优越，主要因为一位高明的工匠会选择最好的案例记录下来，并在之后的作品中使用，并在短时间内和当前案例融合起来，创造出全新的事物。如果你不相信我们人类思维网络以及记忆存储的无尽能力，那你可以仔细研读Susan Magsamen 和 Ivy Ross的新书《你的艺术头脑》。

关于这个事实的证据，我们可以从许多关于传统铁匠、日本木匠、陶艺大师、画家、雕塑家或是瑞典木雕家等工作室纪录片中观察到一二。AI永远无法做出与这些人类奇迹相媲美的工艺作品或是艺术作品。每件作品都与传统文化息息相关，但又独一无二并不可复制。沃尔特本杰明也会对此表示羡慕。

Recently, I resolved that I wasn’t going to be drawn into the silly posturing about how ChatGPT would take the jobs of every experienced architect on earth before 2030, but an intelligent post on this website by Geethanjali Raman and Mohik Acharya broke that resolve. What isn’t being stressed is that algorithms that sample internet-based information are only as good as the quality of that information. Architectural history suggests that all new things have a shelf life, quickly fading from view after being hyped. Only the best will persist after a lengthy period of evaluation and criticism. Any new architecture widely praised and available since the rise of the internet is likely to be untested by time and thus not worth using as a benchmark. And let’s face it: Some of the worst buildings ever designed by humans are out there in cyberspace, crowding out better ones that haven’t yet been digitized.
I am not a psychic, influencer, industry analyst, or business guru, so I won’t comment on the likelihood that AI will transform the way humans create, manufacture, and consume artifacts in the future. There are plenty of experts that are already churning out data and analyses about that. What I am concerned about is its influence on practicing architects confronting the challenges ahead: global warming and the rest. The profession is in crisis. We must not cave into hype about “new” methodologies or technologies that may make our jobs easier (or eliminate them altogether). We are the best judges of beauty, emotional valences, functional advantages, human needs, and every other factor that our clients—users—care about. If other sectors of the global building industry incorporate AI, so be it, but we are not compelled to swallow their Kool-Aid.
Regardless of the sheer power of digital technology to crunch data and speed up design of any kind, nothing about the nature of artificial intelligence, including neural network processing and machine learning borrowed from Christopher Alexander’s Pattern Language, will convince me that it can replace the cognitive power of experienced master designers and artisans. The latter in particular have embodied knowledge that cannot be replicated by any machine or computer. Each carries more in their mind and body than the biggest computer network or digital cloud that any engineer could produce or imagine. Human memory is superior to any kind of machine memory, mainly because a genius or master artisan selects only the best examples to record and use in later work and mixes those examples in microseconds to make new things. Dip into Susan Magsamen and Ivy Ross’s new book, Your Brain on Art, if you don’t believe in the unlimited capacity of our cognitive networks and memory archives.
Compelling evidence for this can be gleaned from any video of a traditional blacksmith, Japanese woodworker, master potter, painter, sculptor, or Swedish Sloyd woodcarver in their studio. AI will never produce works of craftsmanship or art comparable to that of these human wonders. Every work is bound by tradition yet is completely unique and unreproducible. Walter Benjamin can eat his heart out.


Image Courtesy of Wikipedia Commons

建筑师们能否接受这种创造性的思维模式和手工艺工作方式。显然不行。我们是否能够从中有所裨益？可以，就像我在我最近出版的一本书中所阐述的那样。现代主义建筑与晚期资本主义景观设计的其中一个悲剧，源自于他们错误地将新“方法”作为创造性设计的圣经。如果参数化设计或是生成设计可以随意在成千上万个案例中选择“最完美”的案例来解决问题，那最终并不会产生什么好东西。大部分关于卫生间、厨房、卧室等空间的设计最优解已经在过去的几百年间经过了无数建筑师和使用者的检验。将10万种解决方案放入电脑程序中进行计算，很可能会产出和那些经过检验的设计相一致的范式，或者（更可能）产出一堆垃圾。

许多关于使用者对于建筑或建成环境偏好的当代研究，创造出了大量当代研究案例，其中一小部分在1900年之前被实际建造出来。甚至包括在最近的一项关于“精神”、“冥想”以及“宁静”空间的双盲实验中，参与者被要求从25-30张照片中选出对应空间。样本中不仅包括哥特教堂和科罗拉多大峡谷，也包括一些典型的历史书中出现过的现代教堂，包括柯布西耶的朗香教堂、阿尔瓦阿尔托的三十字教堂、琼斯的桑科罗恩教堂。反面例子包括高速公路或是商场。但是一些明显的历史性提示词却不太可能为科学家们提供用来得到普世性结论的数据支撑。

就像我在多年前曾看到过的一篇文章中提到的，关于最佳办公环境方案的研究，通常会要求使用者从一堆还过得去的案例中选择一个，或者是从那些在过去30年中被普遍使用到的案例中进行选择。其中没有一个案例来自莱特、沙利文、麦金、科贝特、沙里宁或是其他一些现代派大师的高层建筑或公寓设计案例。这些建筑师的伟大建筑作品或室内设计作品，有多少被转化为数字形式？除非我们直接浏览“建筑历史学家协会”网站，但数量也非常少。网络上大量存在的作品往往过于空洞、缺乏吸引力，甚至常常十分糟糕。更令人困扰的是，例如库哈斯、福斯特、屈米、哈迪德以及盖里等当代建筑大师的作品却弥漫在数字环境当中。新潮的以及“尖端的”建筑作品目前主要来自于西雅图地区，在那里，建筑被不断打散重构。想找到最新的交通枢纽建筑案例？你可以去看看圣地亚哥·卡拉特拉瓦（Santiago Calatrava）设计的世贸中心地铁站，一座耗资巨大的建筑灾难。谷歌浏览器为大量有用的知识提供了许多垃圾信息。这在我们搜索“伟大建筑”这个关键词时尤为明显。

虽然人工智能行业的各大巨头都在鼓吹他们的算法和搜索引擎的全面、深度及综合的整合能力，任何电子数据库只能在其自身的产品库中才具有一定优势。环境设计、景观设计以及建筑设计领域的数据库显然少得可怜。甚至包括神经学和环境心理学方面的信息都只在专业的数据库中才得以体现。那些在AIA、CNU、APA等每年的学术会议上提出的创造性思想，往往无法被那些决策者或是所谓的谷歌专家们所知晓。再者，那些聪明且成就等身的社会科学家们对于这种人工智能数据库的批判四线个，往往会被学术界以及职场领袖们根据他们个人的好恶掩藏起来。此外处于商业利益的考虑，人们往往会将那些反对他们产品和服务的研究成果加以排除。

我们只要看看玻璃行业对于建筑工人、开发商以及建筑师的影响，就可以对人工智能的影响了解一二。AIA网站以及各类建筑杂志上充满了皮尔金顿公司生产的玻璃以及其他透明玻璃幕墙产品。大多数建筑出于节能和良好自然光照的考虑使用了这些产品。这些产品通过相似的方式在全世界范围内传播开来。REVIT中已经收录了这些产品的相关规范，取代了其他类似的建筑产品，包括像黏土、砂浆之类传统材料。设计师们已经只需要点击鼠标就可以对许多建筑安装施工方案提出创造性的“解决方案“，但是这些方案并不具有太多的创造性。

当人们谈起软件是否能够证明这样一种假说，即人们可以根据公式复制出历史上那些伟大的建筑作品时， George Hersey与理查德·弗里德曼（Richard Friedman）所著的《Possible Palladian Villas》（1992）这部著作证明了这一点，只要你能够接受那种千篇一律的复制品。我买下了这本带有光盘的书，与俄罗斯方块或模拟城市不同，这本书仅仅吸引了我几个小时。与此同时，一位音乐系教授告诉了他的学生如何写出莫扎特奏鸣曲的秘密，大多数学生都能轻易做到，只是非常枯燥乏味。查尔斯·罗森（ Charles Rosen）将一本关于莫扎特和贝多芬的书称作《古典风格》。任何风格都可以被复制，但是杰作永远无法被复制。莫扎特的作品是这样，帕拉迪奥的作品也是这样。

经验丰富的建筑师能够在脑海中长期储存那些经典案例，以备新项目来临时可以用到。当建筑师遇到短期的或机械记忆，那些案例立刻通过转化为认知空间的方式，被改变或转化。当我使用草图进行设计时，我会保持短期记忆能力灵活多变，保持我的手指处于放松状态。如果我使用AUTOCAD进行工作，我通常会在处于概念思考阶段时，创造出一份方案草图。如果我们使用SKETCHUP或是MORPHOLIO TRACE这些软件进行设计，我们可以完成更多的内容，许多年轻设计师都非常善于使用这些软件。

据我所知，大多数喜欢在概念设计阶段采用传统设计方式的建筑师，一般不会使用AI或算法来进行设计工作。我认为除了在扎哈工作室工作的员工之外，任何伟大的建筑师都不会完全依赖于软件进行设计。很快，人们会使用CHATGPT来生成一份20世纪30年代的密斯风格的作品，而不是使用最新的数字棋弈软件来点一份汉堡外卖。这种工作方式的转换有什么好处呢？提出五个平庸的方案，这些方案本来会在上传到Yellow Trace网站后被丢弃。

如果AI引擎开始侵入到我的设计领域范围内，我将会竖起屏障进行抵抗。全世界各地有经验的建筑师应该也会这样做。不幸的是，如果我们行业的那些年轻领导者不这么做，那么我们将会陷入麻烦之中。人类必须为此划清一条底线。

Have architects embraced this model of creative thinking and handwork? Alas, no. Would we benefit from doing so? Yes, as my recent book argues. One of the tragedies of Modernism and the alienating late capitalist landscape has been the false narrative of novel “solutions” as the Holy Grail of creative design. If parametric or generative design algorithms are let loose in the world to choose the “best” example among thousands, or millions for that matter, nothing good will come out the other side. Most of the optimal solutions to bathroom, kitchen, and room design of any kind have been thoroughly tested by architects and users over the centuries. Putting 100,000 additional solutions into a computer program will likely produce either those paradigms or (more likely) garbage.
Many contemporary experiments about user preferences in buildings or environments offer dozens of contemporary examples, and a tiny selection of those built before 1900. Even a recent double-blind neuroscience survey of “spiritual,” “meditative” and “peaceful” spaces asked participants to judge from about 25-30 photographs. These contained not only Gothic cathedrals and the Grand Canyon, but also every canonical modern church in the history textbooks, including Corbusier’s Ronchamps, Aalto’s Church of the Three Crosses, and Fay Jones’ Thorncrown Chapel. Negative examples might show an average superhighway or shopping mall. But a few historically obvious prompts are unlikely to give scientists the kind of data one needs to draw global conclusions.
As I noted in a review several years ago, surveys of optimal office environments generally ask users to choose from only fair-to-middling examples, or at least those commonly offered during the past 30 years. Not one is taken from Frank Lloyd Wright, Louis Sullivan, Charles McKim, Harvey Wiley Corbett, Eero Saarinen, or any modern master of tall buildings or loft design. How many of these architects’ great buildings or interiors are available in digital form? Unless one goes directly to the Society of Architectural Historians website, very few. What is commonly stored in cyberspace is vacuous, uninspiring, and often horrible. More disturbing, the work of contemporary lions such as Rem Koolhaas, Norman Foster, Bernard Tschumi, Zaha Hadid, and Frank Gehry has been seeded in every corner of the digital ecosphere. Novel and “cutting edge” solutions to library design point right to Seattle, where the building is failing and falling apart. Transit hub architecture? Try the World Trade Center PATH hub/station by Santiago Calatrava, a budget-busting disaster. Google provides a ton of garbage for every ounce of wholesome, useful knowledge. This is particularly true of any search for “great architecture.”
Although the giants of the industry trumpet the thorough, deep, and comprehensive capacity of their algorithms and search engines, any digital database is only as good as what has been included in its sampling of products. In environmental, urban, landscape, and architectural design these data sets are pitifully small. Even the work of neuroscientists and environmental psychologists is poorly represented in any but specialist databases. The truly innovative thinking that might be presented annually at AIA, CNU, or APA conferences does not reach decision makers or Google experts. Moreover, the most intelligent and accomplished hard and social scientists who might critique such databases are kept in the dark by academic and professional leaders because it is in their interest to do so. And commercial interests tend to crowd out research that isn’t supportive of their products and services (viz. the opioid crisis).
One only need look at the influence of the glass industry on builders, developers, and architects to confirm this. The AIA website and architecture journals are full of ads for Pilkington wonder glass and other translucent curtain wall products. Most buildings are clad in those products at the expense of energy efficiency and even good natural lighting. They penetrate similar media throughout the world. Revit contains specifications for such products in lieu of other cladding alternatives, including traditional ones such as terra cotta and stucco. Designers are already just clicking on a mouse to create innovative “solutions” for many building assemblies, but getting old and tired ones.
And when it comes to software that tested the hypothesis of whether great architecture could be replicated according to formulas, George Hersey and Richard Friedman’s Possible Palladian Villas (1992) proved it could, as long as you could accept dull copies. I bought the book with its floppy disc and, unlike Tetris or Sim City, it kept me interested for a couple of hours, at most. Around the same time, a music professor gave students the secret to writing Mozart sonatas and most had no trouble doing so—they were just boring. Charles Rosen wasn’t wrong in calling his book on Mozart and Beethoven The Classical Style. Any style can be copied, but no masterpiece can be replicated. Not Mozart’s, not Palladio’s.
Experienced architects carry the best precedents in their long-term memory, to be used whenever a new project comes along. When they appear in short-term or working memory they are immediately changed or transformed by the process of moving into a cognitive space. When I design using sketches, I keep my short-term memory flexible and my fingers loose. If I am working on a computer in AutoCAD, I may make a “sketch” of a plan while in my conceptual mode of thinking. If one uses SketchUp or Morpholio Trace one can do more, and many younger designers are adept at this.
No one I know who likes to work this way has any use for AI problem solving or algorithms during the conceptual phase of design. I doubt that any master architect—except perhaps those on the staff at Zaha Hadid’s office—would turn to this kind of software if offered. One would no sooner ask ChatGPT to generate a plan in the style of Ludwig Mies van der Rohe’s 1930s work than request a takeout burger using the latest digital Chess engine. What advantages does such a shapeshifter offer? Five tepid alternatives that one would have thrown away after putting them onto yellow trace?
If AI engines start to invade my design space I will take to the barricades to resist. So will most experienced architects around the world. Unfortunately, if younger leaders in our profession don’t do the same, we are in deep trouble. A line must be drawn. By humans.