Persons of Interest

Refik Anadol, The Art World’s Happy Warrior for A.I.

雷菲克·阿纳多尔(Refik Anadol),艺术界的人工智能快乐战士
His new museum, Dataland, is a joyful monument to the technology. Is he a visionary, or Silicon Valley’s court painter?
他的新博物馆「数据之地」(Dataland)是对这项技术的一座欢欣纪念碑。他究竟是位远见者,还是硅谷的御用画师?
Animated portrait of Refik Anandol
Illustration by Vivek Thakker; Source photograph from Getty

“All right!” Refik Anadol said, as the escalator carried us downward. “We are entering the dream of the machine.” We had arrived in the cavernous first gallery of Dataland, the twenty-five-thousand-square-foot space, in downtown Los Angeles, that Anadol calls the first museum of A.I. art. Space-age music blared night-club-loud as pictures of birds, plants, and flowers cascaded down the walls. This array was a small sample of the half-billion images—and the hundred thousand hours of audio, including birdsong, rain, and even silence—on which Anadol has trained the Large Nature Model, an A.I. model that powers “Machine Dreams: Rainforest,” Dataland’s inaugural show. The pictures swooped around and beneath us like a cloud of starlings, and an earthy, slightly metallic smell emanated from the diffusers we wore around our necks, which, along with a biometric wristband, each Dataland visitor receives upon arrival. The aroma is one of twelve scents—formulated by the L.N.M., with a little help from L’Oréal Luxe—that perfume visitors’ trips through the museum’s five rooms. Anadol calls it “the scent of data.”

「好嘞!」雷菲克·阿纳多尔在自动扶梯载着我们下行时说道。「我们正在进入机器的梦境。」我们抵达了数据之地那洞穴般的第一展厅。这座位于洛杉矶市中心、占地两万五千平方英尺的空间,被阿纳多尔称为首座人工智能艺术博物馆。太空时代风格的音乐以夜店般的音量轰鸣,鸟、植物和花朵的影像从墙壁上倾泻而下。这组阵列只是阿纳多尔训练「大型自然模型」(Large Nature Model)所用五亿张图像——以及十万小时音频,包括鸟鸣、雨声乃至寂静——中的一小部分样本。该模型驱动着数据之地的开幕展「机器之梦:雨林」。影像如椋鸟群般在我们周围和脚下翻飞,我们脖子上挂着的扩散器散发出一种泥土般的、略带金属味的气息。每位数据之地的访客在抵达时都会领到这样一个扩散器,外加一条生物识别腕带。这种香气是十二种气味之一——由大型自然模型在欧莱雅奢品(L'Oréal Luxe)的一点协助下调配而成——在博物馆的五个展厅中为访客的旅程增添嗅觉体验。阿纳多尔称之为「数据的气味」。

Data is Anadol’s alpha and omega, his medium and his muse. He often works with large archives, or, as with the L.N.M., scientific data gathered by organizations such as the Smithsonian Institution and London’s Natural History Museum, and by his own researchers in rain forests across the world. “Visualization” is too humble a word for the operations that Anadol performs, though in recent years his work has been unusually visible. His art has lit up the Sphere, in Las Vegas; the stage of the Grammy Awards, in L.A.; and the halls of the World Economic Forum at Davos and the United Nations in New York. He has done “collaborations”—the industry euphemism for upscale advertisements—with Bulgari, Dior, and Hennessy. He has permanent installations in Boston’s Logan Airport, and in JPMorgan Chase’s gargantuan new headquarters in midtown Manhattan. And in 2022, amidst a blowout year-long show, his “Unsupervised: Machine Hallucinations” became the first generative-A.I. art work—and one of the first linked to non-fungible tokens—to enter the MOMA collection, and with it the canon of contemporary art.

数据是阿纳多尔的起点与终点,是他的媒介,也是他的缪斯。他经常处理大型档案,或者像大型自然模型那样,处理由史密森尼学会(Smithsonian Institution)、伦敦自然历史博物馆等机构,以及他自己在世界各地雨林中的研究人员所收集的科学数据。「可视化」这个词用来形容阿纳多尔所进行的操作未免太过谦逊,尽管近年来他的作品异常引人注目。他的艺术曾点亮拉斯维加斯的球体馆(Sphere)、洛杉矶格莱美奖的舞台,以及达沃斯世界经济论坛和纽约联合国总部的厅堂。他与宝格丽(Bulgari)、迪奥(Dior)和轩尼诗(Hennessy)进行过「合作」——这是业界对高端广告的委婉说法。他在波士顿洛根机场和摩根大通(JPMorgan Chase)位于曼哈顿中城的巨型新总部都设有永久装置。而在2022年,在一场持续整年、盛况空前的展览中,他的作品《无监督:机器幻觉》(Unsupervised: Machine Hallucinations)成为首件进入纽约现代艺术博物馆(MOMA)馆藏的生成式人工智能艺术作品——也是首批与非同质化代币(non-fungible tokens)关联的作品之一——并由此进入了当代艺术的经典序列。

MOMA felt very far away as the images around us gave way to a “Matrix”-like environment of grids strobing to a driving beat. Anadol was describing how measurements from various rain forests were informing what I was seeing, but I was struggling to absorb our surroundings. “It’s overwhelming,” I told Anadol. “The room is seven hundred and twenty million pixels,” he replied, by way of explanation. The sensory overload, he added, was meant to evoke “the sublime power of being connected to incredibly vast data sets,” which Anadol views as the “new language of humanity.”

当我们周围的影像让位于一个类似《黑客帝国》的、网格随强劲节拍频闪的环境时,MOMA 的感觉变得非常遥远。阿纳多尔正在描述来自各个雨林的测量数据如何影响我所看到的一切,但我却难以消化周遭的景象。「这太震撼了,」我对阿纳多尔说。「这个展厅有七亿两千万像素,」他解释道。他补充说,这种感官超载意在唤起「与极其庞大的数据集相连所带来的崇高力量」,阿纳多尔将此视为「人类的新语言」。
Figure stands amidst a large room full of projected visuals

A view of Dataland’s first exhibition, “Machine Dreams: Rainforest.” Anadol believes art critical of A.I. is too “predictable.”

数据之地首展「机器之梦:雨林」的一瞥。阿纳多尔认为,批判人工智能的艺术过于「可预测」。
数据之地首展“机器之梦:雨林”一瞥。阿纳多尔认为,批判人工智能的艺术过于“可预测”。
Art work © 2026 Refik Anadol Studio / Courtesy Refik Anadol Studio

After about fifteen minutes, the surging graphics and music stopped, and we were in near-darkness. As strings started to play Mahler’s Third Symphony, the smell of wet earth rose from our diffusers, and a virtual rainstorm began. There was a delicate pulse in the background; Anadol explained that it was an average of our heartbeats, which, along with our galvanic response and body temperature, was being measured by our wristbands and processed by Dataland’s “Connectome,” a miniature data center with top-of-the-line chips provided by Nvidia. The storm built until the walls and floor were covered with pale gray particles, billowing like smoke, then tessellating in triangles. The model learns in subtle ways from each visitor, but your data is automatically deleted, though you can ask for a copy first. (Guests can also choose to skip the wristband and go through as “ghosts.”) “There is nothing, like, recorded here,” Anadol said. “It’s not a place for surveillance or whatever.”

大约十五分钟后,汹涌的图形和音乐戛然而止,我们陷入近乎黑暗之中。随着弦乐开始演奏马勒的第三交响曲,湿润泥土的气味从我们的扩散器中升起,一场虚拟的暴雨开始了。背景中传来一阵微妙的脉搏;阿纳多尔解释说,这是我们心率的平均值,连同我们的皮肤电反应和体温,正被我们的腕带测量,并由数据之地的「连接组」(Connectome)——一个配备英伟达(Nvidia)顶级芯片的微型数据中心——进行处理。暴雨逐渐增强,直到墙壁和地面被淡灰色的粒子覆盖,如烟雾般翻涌,然后镶嵌成三角形。模型会以微妙的方式从每位访客身上学习,但你的数据会被自动删除,不过你可以先索要一份副本。(访客也可以选择跳过腕带,以「幽灵」身份参观。)「这里没有任何东西被记录下来,」阿纳多尔说。「这不是一个用于监控或类似目的的地方。」

Eventually, as horns and strings held triumphal chords, everything dissolved into what Anadol calls data pigmentation: a roiling sea of polychrome spheres, like a ball pit come to life. Each represented a data point in the L.N.M., and all were somehow responding to our presence. “I love this feeling so much,” Anadol said. “For five thousand years, we have looked at art work and felt something. But can the art work feel us back?”

最终,当圆号与弦乐奏出凯旋的和弦时,一切都溶解为阿纳多尔所说的「数据色素」:一片由多色球体构成的翻腾海洋,就像一个活过来的球池。每个球体都代表大型自然模型中的一个数据点,并且所有球体都以某种方式回应着我们的存在。「我太喜欢这种感觉了,」阿纳多尔说。「五千年来,我们凝视艺术作品并有所感受。但艺术作品能否反过来感受我们呢?」

An energetic forty-year-old who wears all black all the time, and who rarely stops smiling beneath his round black-rimmed glasses, Anadol speaks less of problems than of possibilities. At his studio, in L.A.’s Frogtown neighborhood, some twenty data scientists, engineers, and architects work on dual-screen monitors, and occasionally walk on treadmills at standing desks. Anadol’s black Range Rover bears a license plate with an e-ink screen, which he programmed to read “Blade Runner,” beneath his plate number. (The film is Anadol’s most frequent cultural reference, though less for its dystopian plot than for what he calls its “vibe.”) He wears a black Oura ring on his right index finger and, sometimes, on his right wrist, a blue-bead bracelet made for him by members of the Yawanawá tribe in western Brazil, which he and Efsun Erkılıç—his creative and life partner, Dataland’s co-founder, and, as she put it, its “worst critic”—have visited routinely for immersion in the rain forest, along with the occasional trip on a local variant of ayahuasca. Through N.F.T. sales, the couple has raised millions of dollars for the tribe.

阿纳多尔四十岁,精力充沛,永远一身黑衣,圆框黑边眼镜下几乎总是挂着微笑。他谈论可能性远多于谈论问题。在他位于洛杉矶蛙镇(Frogtown)的工作室里,大约二十名数据科学家、工程师和建筑师在双屏显示器前工作,偶尔会在站立式办公桌的跑步机上走一走。阿纳多尔的黑色路虎车挂着一块电子墨水屏车牌,他将其编程为在车牌号下方显示「银翼杀手」。(这部电影是阿纳多尔最常引用的文化参照,不过更多是因为他所说的「氛围」,而非其反乌托邦情节。)他右手食指上戴着一枚黑色乌拉(Oura)戒指,有时右手腕上会戴一串蓝珠手链,那是巴西西部雅瓦纳瓦(Yawanawá)部落的成员为他制作的。他和埃夫松·埃尔克勒奇(Efsun Erkılıç)——他的创作与生活伴侣、数据之地的联合创始人,以及用她的话说,其「最严厉的批评者」——定期前往那里,沉浸在雨林中,偶尔还会体验当地一种死藤水的变体。通过非同质化代币的销售,这对伴侣已为该部落筹集了数百万美元。

Anadol studied photography and design at Bilgi University, in Istanbul, and earned a second M.F.A. at U.C.L.A., where his teachers included the noted media artist Casey Reas. But his formal education was supplemented, and subsidized, by tech companies. A Microsoft award covered some of his early research expenses, and in 2016 he was a fellow at Google’s Artists and Machine Intelligence program. There he worked with Mike Tyka, who had helped develop DeepDream, a program that used neural networks, trained on millions of images, to generate psychedelic patterns. “That moment to me is so important,” Anadol said. “That concept of, if a machine can learn, can it dream?” Ever since, Anadol has spoken of putting a “camera in the mind of the machine” to see what it “hallucinates” when trained on the large bodies of data—always used with permission, often by invitation—that he and his team meticulously prepare. In addition to wall- and even room-size “data sculptures” made with, say, wind data or tourist photos of New York or brain-wave measurements or digital archives of Renaissance art, he also produces dynamic “data paintings” that might be displayed in a museum or purchased by a collector.

阿纳多尔在伊斯坦布尔的比尔基大学(Bilgi University)学习摄影与设计,并在加州大学洛杉矶分校(U.C.L.A.)获得了第二个艺术硕士学位,师从著名媒体艺术家凯西·瑞斯(Casey Reas)等人。但他的正规教育得到了科技公司的补充和资助。微软的一项奖项支付了他早期的部分研究费用,2016年,他成为谷歌「艺术家与机器智能」(Artists and Machine Intelligence)项目的研究员。在那里,他与迈克·泰卡(Mike Tyka)合作,后者曾参与开发 DeepDream,一个利用神经网络、在数百万张图像上训练以生成迷幻图案的程序。「那一刻对我来说太重要了,」阿纳多尔说。「那个概念:如果机器能学习,它能做梦吗?」从那时起,阿纳多尔就一直在谈论将「一台摄像机放入机器的思维中」,看看当它在他和团队精心准备的大型数据集(始终经许可使用,通常是受邀使用)上训练时,会「幻觉」出什么。除了利用风的数据、纽约的游客照片、脑电波测量数据或文艺复兴艺术的数字档案等制作的、尺寸可达墙面甚至整个房间的「数据雕塑」外,他还创作动态的「数据绘画」,这些作品可能在博物馆展出,或被收藏家购买。

Anadol enjoyed an early triumph in 2018, when he lit up the silver sails of the Walt Disney Concert Hall, in Los Angeles, with projections developed using terabytes of recordings, program notes, and archival documents from the L.A. Philharmonic. But he wasn’t widely known until 2022, when “Unsupervised” arrived at MOMA. The show followed a series of digital art works which Anadol sold as N.F.T.s., and from which MOMA quietly earned a portion of the proceeds. Buoyed by the promise of digital art, and perhaps by its revenue, the museum displayed Anadol’s work on a twenty-four-foot-tall high-definition L.E.D. screen in the Agnes Gund Garden Lobby, where it was among the first things visitors saw.

阿纳多尔在2018年取得了一次早期成功,他用洛杉矶爱乐乐团数 TB 的录音、节目单和档案文件制作了投影,点亮了洛杉矶华特·迪士尼音乐厅的银色风帆。但他真正广为人知是在2022年,《无监督》来到 MOMA 之后。该展览之前,阿纳多尔以非同质化代币形式出售了一系列数字艺术作品,MOMA 悄悄从中获得了一部分收益。在数字艺术前景及其可能带来的收入的鼓舞下,博物馆将阿纳多尔的作品展示在艾格尼丝·冈德花园大厅(Agnes Gund Garden Lobby)一块二十四英尺高的高清 LED 屏幕上,那里是访客最先看到的东西之一。

To make “Unsupervised,” Anadol trained an A.I. on more than two hundred years’ worth of art from MOMA’s collection: not just paintings and sculptures but video games, films, and design objects, along with their associated metadata. Then his team had the model identify patterns beyond human-imposed categories like “Abstract Expressionism,” say, or “nineteenth century.” Technically, what appeared onscreen were visions from the “mind” of the model as it moved through its map of the collection, and as it responded to site-specific information like the weather and sounds made by visitors. Visually, the result was a restlessly fluctuating Technicolor whorl. Abstractions that looked like scribbled pastels or Helen Frankenthaler oils or the iTunes visualizer would give way to churning data pigment, sloshing in a square frame. Soothing ambient music played in the background.

为了创作《无监督》,阿纳多尔用 MOMA 馆藏中超过两百年的艺术作品训练了一个人工智能:不仅是绘画和雕塑,还有电子游戏、电影和设计物品,以及它们相关的元数据。然后,他的团队让模型识别出超越人为分类(比如「抽象表现主义」或「十九世纪」)的模式。从技术上讲,屏幕上出现的是模型在其馆藏地图中移动时,以及响应特定场域信息(如天气和访客发出的声音)时,其「思维」中的景象。视觉上,结果是一个永不停歇、变幻莫测的彩色漩涡。看起来像潦草粉彩画、海伦·弗兰肯瑟勒(Helen Frankenthaler)的油画或 iTunes 可视化效果的抽象图形,会让位于在方形画框中翻腾搅动的数据色素。背景中播放着舒缓的氛围音乐。

“Refik devised different ways of essentially diagramming, visualizing, and then even animating the history of modern art,” Michelle Kuo, one of the curators who organized the exhibition, told me. She compared it to the Surrealists and their “automatic” art, the whimsical painting machines made by the Swiss sculptor Jean Tinguely, and the collaboration between engineers from Bell Labs and Robert Rauschenberg.

「雷菲克设计了不同的方法,从根本上图解、可视化,甚至动画化现代艺术史,」策划该展览的策展人之一米歇尔·郭(Michelle Kuo)告诉我。她将其与超现实主义者及其「自动」艺术、瑞士雕塑家让·丁格利(Jean Tinguely)制作的异想天开的绘画机器,以及贝尔实验室工程师与罗伯特·劳森伯格(Robert Rauschenberg)的合作相提并论。

“Unsupervised” was wildly popular and eminently Instagrammable. Glenn Lowry, the former head of MOMA, credits the work with helping rebuild attendance after the pandemic, and saw it as a mise en abyme for the museum. “We acquire works of art and then we often sell those works of art in order to acquire new works of art,” Lowry told me. “The collection is in a constant state of metabolism and ‘Unsupervised’ is a metabolic project.” It was eventually metabolized itself: toward the close of the show, MOMA announced that it had accepted “Unsupervised” as a donation; one of the piece’s former owners, a crypto investor and digital-art collector, now sits on the museum’s board.

《无监督》极受欢迎,非常适合在 Instagram 上分享。MOMA 前馆长格伦·洛瑞(Glenn Lowry)认为这件作品帮助博物馆在疫情后重建了参观人数,并将其视为博物馆的一个「镜渊」(mise en abyme)。「我们收购艺术品,然后常常出售这些艺术品以收购新的艺术品,」洛瑞告诉我。「馆藏处于一种持续的新陈代谢状态,而《无监督》正是一个新陈代谢的项目。」它最终自身也被代谢了:展览接近尾声时,MOMA 宣布接受《无监督》作为捐赠;这件作品的前任所有者之一,一位加密货币投资者和数字艺术收藏家,现在是博物馆的董事会成员。

The critical reaction was more ambivalent. Lloyd Wise, in Artforum, noted how the piece evoked “the broader modernist tendency toward noncomposition,” and Sebastian Smee, in the Washington Post, called it an “early masterpiece of AI-generated art.” Jerry Saltz, in New York, dismissed the work as a “pointless museum mediocrity,” “some cross between relaxation exercise and euphoric TED Talk and NSA levels of data mining.” Ben Davis, in perhaps the most substantial treatment, wrote in Artnet that “Unsupervised” presented MOMA’s collection as “just a bunch of random visual tics to be permuted, rather than an archive of symbol-making practices with social meanings.” He suggested that Anadol’s “purely decorative, cheerleader-ish style of A.I. art”—as opposed to the social and political analyses of A.I. offered by the likes of Trevor Paglen and Hito Steyerl—is one of the reasons he has received so much support from tech giants.

评论界的反应则更为矛盾。劳埃德·怀斯(Lloyd Wise)在《艺术论坛》(Artforum)上指出,这件作品如何唤起了「更广泛的现代主义非构图倾向」,而塞巴斯蒂安·斯密(Sebastian Smee)在《华盛顿邮报》上称其为「人工智能生成艺术的早期杰作」。杰瑞·萨尔茨(Jerry Saltz)在《纽约》杂志上则斥之为「毫无意义的博物馆平庸之作」,「某种介于放松练习、亢奋的 TED 演讲和国安局级别的数据挖掘之间的东西」。本·戴维斯(Ben Davis)或许给出了最实质性的论述,他在 Artnet 上写道,《无监督》将 MOMA 的馆藏呈现为「只是一堆随机的视觉特征,可以被排列组合,而非一个具有社会意义的符号创作实践的档案」。他指出,阿纳多尔那种「纯粹装饰性、啦啦队长式的人工智能艺术风格」——与特雷弗·帕格伦(Trevor Paglen)和希托·施泰尔(Hito Steyerl)等人对人工智能的社会政治分析形成对比——是他获得科技巨头如此多支持的原因之一。

It takes a certain chutzpah to use the most consequential technology of the age—a technology which makes many deeply uneasy—without somehow bringing those problems into the work. But for Anadol, art critical of A.I. is important yet “predictable,” he told me. “I don’t think there’s anything unusual about being there,” he said. “What is unusual to me is being here.”

使用这个时代最具影响力的技术——一项让许多人深感不安的技术——却不以某种方式将这些问题带入作品,这需要一定的胆量。但对阿纳多尔来说,批判人工智能的艺术虽然重要,却「可预测」,他告诉我。「我不认为站在那个立场有什么不寻常的,」他说。「对我来说,不寻常的是站在这里。」

Anadol refers to his collectors and patrons, like Nvidia’s Jensen Huang, as “pioneers.” “Good or bad for humanity, I don’t know, but they are pioneers of their field,” he told me. “And I think that needs to be matched by another breakthrough. It needs another pioneership.” He drew a comparison to the Renaissance, when an artist might be offered a chance to paint a chapel ceiling. “Jump cut, I feel, like, ‘Hey, Refik, here’s a G.P.U. we are building. We believe this will change the world, and we believe that you can use this for good.’ ”

阿纳多尔将他的收藏家和赞助人,如英伟达的黄仁勋,称为「先驱」。「对人类是好是坏,我不知道,但他们是各自领域的先驱,」他告诉我。「我认为这需要另一个突破来匹配。这需要另一种先驱精神。」他将其与文艺复兴时期相提并论,那时艺术家可能会获得绘制教堂穹顶的机会。「跳切到现在,我感觉就像是,『嘿,雷菲克,这是我们正在打造的 GPU。我们相信这会改变世界,我们相信你可以用它来做善事。』」

In that case, I asked, Jensen Huang would represent the Medici?

那样的话,我问,黄仁勋就代表了美第奇家族?

“Can be,” he replied. “I mean, we can be critical about those giants, and what they do for societies and governments. Again, my other colleagues are very good at that. But what I’m trying to understand is, what is this really doing to the medium? What is it doing to the painting, to the brush, to the canvas?”

「可以是,」他回答。「我的意思是,我们可以批判那些巨头,以及他们对社会和政府所做的事情。再说一次,我的其他同事在这方面做得很好。但我试图理解的是,这到底对媒介做了什么?它对绘画、对画笔、对画布做了什么?」

When it comes to technology, Anadol’s preoccupations are ultracontemporary, but the way he talks about his role as an artist has more in common with the raw ambition of early modern painters and sculptors, or indeed of the tech founders he counts as friends and customers, than with the language of critique preferred by his peers. And yet this disconnect is partly why “Unsupervised” was a landmark. The piece made a broader public think, and worry, about how A.I. would change culture, and how it would bring new audiences, new money, and new ethical quandaries into the art world. And it marked the arrival of a new art-world character: a dreamer articulating a future for creativity on the stages of TED (where he has spoken twice) and the Nvidia annual conference (where Anadol presented work made with the L.N.M.), and now marshalling enormous resources to found a museum from scratch. Anadol’s positivity is not unheard of in his field: Andy Warhol, another innovative pop artist accused of being superficial and overly commercial, shaped his persona around “liking” everything. But, unlike Warhol, it’s hard to read Anadol or his work as ironic. I always had the sense that he meant exactly what he said.

在技术方面,阿纳多尔的关注点极其当代,但他谈论自己作为艺术家角色的方式,更接近早期现代画家和雕塑家那种原始的雄心,甚至更接近他视为朋友和客户的科技创始人的雄心,而非同行们偏爱的批判语言。然而,这种脱节正是《无监督》成为里程碑的部分原因。这件作品让更广泛的公众开始思考和担忧人工智能将如何改变文化,以及它将如何为艺术界带来新的观众、新的资金和新的伦理困境。它也标志着一个新的艺术界角色的登场:一个在 TED 舞台(他在那里演讲过两次)和英伟达年度大会(他在那里展示了用大型自然模型创作的作品)上阐述创造力未来的梦想家,现在又调动巨大资源,从零开始创办一座博物馆。阿纳多尔的积极乐观在他的领域并非闻所未闻:安迪·沃霍尔,另一位被指责肤浅和过度商业化的创新波普艺术家,将自己的人设塑造为「喜欢」一切。但是,与沃霍尔不同,很难将阿纳多尔或他的作品解读为讽刺。我总感觉,他说的就是他真实所想。

“Now we are entering the place for discovery, the place for sharing,” Anadol told me, as we plunged deeper into Dataland. We’d left the ball pit and gone through a dark hallway to the Latent Gallery, a narrow rectangular space in which various wall-sized screens offered different ways of interacting with the L.N.M. One lets you use a tablet for “machine drawing,” squiggling with your finger to provoke a response from the model. I selected “Blue Milk Mushroom” from a list and drew a mushroom shape; a hyper-realistic fungus with three caps appeared, surrounded by a pixie dust that conveyed, in a visual language I must have learned from video games, life and freshness. “The biomes are inspiring the drawing techniques,” Anadol said. Another station featured a “UMAP Browser,” which lets you navigate through the model’s training data and presents little factoids about various natural specimens.

「现在我们进入的是探索之地,分享之地,」当我们深入数据之地时,阿纳多尔告诉我。我们离开了球池,穿过一条黑暗的走廊,来到「潜在画廊」(Latent Gallery),这是一个狭窄的长方形空间,其中几面墙大小的屏幕提供了与大型自然模型互动的不同方式。一个屏幕让你用平板电脑进行「机器绘画」,用手指涂鸦来激发模型的回应。我从列表中选择「蓝牛奶蘑菇」,画了一个蘑菇形状;一个带有三个菌盖的超写实真菌出现了,周围环绕着一种精灵尘,用我必定是从电子游戏中学到的视觉语言传达出生命与新鲜感。「生物群系正在启发绘画技巧,」阿纳多尔说。另一个站点设有「UMAP 浏览器」,让你浏览模型的训练数据,并展示各种自然标本的小知识。

The next gallery—with screens on the walls, floor, and ceiling—contained the most transparently narrative part of “Machine Dreams”: an eight-minute journey through an idyllic forest world that resembles “Avatar.” First, you fall through a sort of wormhole teeming with blossoms and butterflies. Then a glass hummingbird—Anadol told me that the figure came to him in a dream, and was interpreted by a Yawanawá chief as sacred—leads us through a valley, following a stream that flows beneath our feet. We eventually arrive at a giant glowing tree, which opens yet another portal into a kind of cyber tunnel, its geometric forms flashing in blue, red, and purple. When we come out the other end, we hear a mournful birdsong. “At the deepest root of the wisdom tree’s memory is a single song,” the screen reads. “In 1987, the last Kauaʻi ʻōʻō bird sang to a partner that no longer existed. He left pauses in the dark for her to answer. We hold the silence for him.” In a show mostly devoid of explicit messages, the sentiment surprised me.

下一个展厅——墙壁、地板和天花板上都有屏幕——包含了「机器之梦」中最具透明叙事性的部分:一段八分钟的旅程,穿越一个类似《阿凡达》的田园诗般的森林世界。首先,你坠入一个布满花朵和蝴蝶的虫洞。然后,一只玻璃蜂鸟——阿纳多尔告诉我,这个形象是他在梦中见到的,被一位雅瓦纳瓦酋长解读为神圣——引领我们穿过一个山谷,沿着一条在我们脚下流淌的溪流前行。我们最终到达一棵巨大的发光树,它打开了另一个通往某种赛博隧道的入口,其几何形状闪烁着蓝、红、紫三色光。当我们从另一端出来时,听到一声哀伤的鸟鸣。「在智慧树记忆的最深处,只有一首歌,」屏幕上写道。「1987年,最后一只考艾岛管舌鸟(Kauaʻi ʻōʻō)向一个已不存在的伴侣歌唱。他在黑暗中留下停顿,等待她的回应。我们为他守着这片寂静。」在一个几乎不含明确信息的展览中,这种情感让我感到惊讶。

The bird’s plangent call echoed among other nighttime sounds as we processed to the final room, a high-ceilinged Sanctuary. Our biometric data appeared on the far wall—heart rate, time spent in the museum, “emotional temperature.” “This piece is a real-time reflection of all we collected and shared and felt,” Anadol said. I noticed the machine judged me “contemplative.” “So we will uplift the energy!” Anadol said with a laugh. On a towering screen appeared an enormous data painting: first the familiar multicolored spheres, and then marvellous garlands of bright little flowers, swirling as a Yawanawá song played. “We want to let the sanctuary heal us,” Anadol said. On our way out of the gallery, we passed through the Design Store, where you can buy T-shirts generated from your data. Those who purchase the most expensive tickets (for a hundred and twenty-nine dollars; the starting price is forty-nine) get what Anadol calls “edible data sculptures,” rain-forest-flavored chocolate formulated with the A.I.’s help.

那只鸟凄切的鸣叫在其他夜声的映衬下回荡,我们列队走向最后一个房间,一个天花板很高的「圣所」(Sanctuary)。我们的生物识别数据出现在远处的墙上——心率、在博物馆停留的时间、「情感温度」。「这件作品是我们收集、分享和感受的一切的实时反映,」阿纳多尔说。我注意到机器判定我为「沉思中」。「那么我们来提振一下能量!」阿纳多尔笑着说。在一面高耸的屏幕上,出现了一幅巨大的数据绘画:首先是熟悉的彩色球体,然后是奇妙的花环,由明亮的小花组成,随着一首雅瓦纳瓦歌曲的播放而旋转。「我们希望让圣所治愈我们,」阿纳多尔说。在离开画廊的路上,我们经过了设计商店,在那里你可以买到根据你的数据生成的 T 恤。购买最贵门票的人(票价一百二十九美元;起步价四十九美元)会得到阿纳多尔所说的「可食用数据雕塑」,即借助人工智能调配出的雨林风味巧克力。
Figure stands backed by a moving visual

In one gallery, The Sanctuary, an A.I.-generated art work responds to a visitor’s biometric data.

在一个名为「圣所」的展厅中,一件人工智能生成的艺术作品回应着访客的生物识别数据。
在一个名为“圣所”的展厅中,一件人工智能生成的艺术作品会对访客的生物识别数据作出回应。
Art work © 2026 Refik Anadol Studio / Courtesy Refik Anadol Studio

Anadol reassured me that the L.N.M.’s training data was ethically sourced, and that the Google cloud-computing facility running the museum mainly uses renewable power. But it wasn’t until after I’d left that I realized that climate change, habitat destruction, and all other human impacts on the environment go unmentioned in “Machine Dreams.” The conservation status of animals and environments is not included in the UMAP-browser interface, and the Kauaʻi ʻōʻō bird’s extinction is mourned almost as a mythical event, dated but nonetheless out of time; no cause is indicated.

阿纳多尔向我保证,大型自然模型的训练数据来源合乎伦理,运行博物馆的谷歌云计算设施主要使用可再生能源。但直到我离开后,我才意识到,「机器之梦」中完全没有提及气候变化、栖息地破坏以及所有其他人类对环境的影响。UMAP 浏览器界面中没有包含动物和环境的保护状况,考艾岛管舌鸟的灭绝几乎被哀悼为一个神话事件,有日期却仿佛超脱于时间之外;没有指明任何原因。

The A.I. itself is presented similarly, so heavily aestheticized as to seem like a force of nature without a history, only awesome power. I tried hard to make sense of the various explanatory screens at Dataland, but found them more symbolic than helpful: the charts and graphs were dense and often labelled with acronyms; some showed scrolling lines of code, meaningful only to a programmer. (Visitors are now able to pose questions to a chatbot on their phones.) Instead of glimpsing into the “black box” of the model, I felt as if I were staring at its outside painted in Day-Glo. The machine is watching you, learning from you, but you never know exactly how.

人工智能本身也被以类似的方式呈现,被高度审美化,以至于看起来像是一种没有历史、只有令人敬畏的力量的自然之力。我努力试图理解数据之地里各种解释性屏幕的内容,但发现它们象征意义大于实际帮助:图表和图形密集,且常常标有缩写;有些显示滚动的代码行,只对程序员有意义。(访客现在可以用手机向聊天机器人提问。)我非但没有窥见模型的「黑箱」,反而感觉自己像是在盯着它涂满了荧光漆的外壳。机器在注视你,向你学习,但你永远无法确切知道它是如何做到的。

For that reason, I think, Anadol’s quieter abstractions provide the most meaningful moments in “Machine Dreams”: the ghostly rain that moves to the Mahler, the data pigment that churns and cascades, the final monumental living painting that dances on the wall. Such work moves beyond spectacle—the kind supplied by “immersive experiences” like TeamLAB, or indeed Disneyland—to get at something like sense. No matter what data you feed into an A.I. model, it always remains abstract. Technical distinctions aside, the L.N.M. doesn’t see the rain forest any differently than “Unsupervised” saw MOMA’s collection. Nature is just more grist for the mill, and we are, too.

正因如此,我认为,阿纳多尔那些更安静的抽象作品提供了「机器之梦」中最有意义的时刻:随着马勒音乐飘动的幽灵般的雨,翻腾倾泻的数据色素,以及在墙上舞动的最终那幅不朽的活绘画。这样的作品超越了奇观——那种由 TeamLAB 或迪士尼乐园等「沉浸式体验」所提供的奇观——触及了某种类似感觉的东西。无论你向人工智能模型输入什么数据,它始终是抽象的。撇开技术差异不谈,大型自然模型看待雨林的方式,与《无监督》看待 MOMA 馆藏的方式并无不同。自然不过是磨坊里的又一批谷物,我们也是。

Anadol told me that he hopes Dataland gets the public to “think about the fears and concerns of humanity, to feel and hear things that are inaudible in everyday life.” He compared himself to James Turrell, doing with information what Turrell did with light. “I’m trying to find the same feelings with data in places that don’t exist yet,” he said. Whether that place seems like a pleasure-dome or a panopticon is up to you.

阿纳多尔告诉我,他希望数据之地能让公众「思考人类的恐惧与担忧,感受和聆听日常生活中听不到的东西」。他把自己比作詹姆斯·特瑞尔(James Turrell),用信息做特瑞尔用光做的事。「我试图在尚不存在的地方,用数据找到同样的感觉,」他说。那个地方在你看来是欢乐宫还是全景监狱,取决于你自己。

The day after visiting Dataland, I drove about forty miles east to Pomona College, where Turrell’s “Dividing the Light” sits in a small grove of palms in the middle of campus. I got there around half past seven, and joined a few others beneath the structure’s canopy: an off-white frame, lofted high on skinny columns, that opens onto the sky. Around eight, hidden lights began to bathe the canopy in a purple glow. Some of the people around me spoke quietly, but mainly we sat in silence. For the next hour or so, the colors of the canopy shifted, slowly but perceptibly, as the sky clung to its blue and then, after lingering in the inkiest navy for just a few precious minutes, seemed to give it all away. Somehow, with nothing but a square hole, Turrell pulls the sky downward, defamiliarizing the most familiar sight there is, making you look with loving care at the thing that’s too big, too far away to love. There’s nothing hidden from you. The magic is that there is no magic. If I weren’t telling you now, no one would know that I was ever there. ♦

参观完数据之地的第二天,我驱车向东大约四十英里,来到波莫纳学院(Pomona College),特瑞尔的《分割光明》(Dividing the Light)就坐落在校园中央的一小片棕榈树林中。我大约七点半到达,和其他几个人一起坐在结构的天篷下:一个灰白色的框架,高高架在细长的柱子上,向天空敞开。八点左右,隐藏的灯光开始给天篷沐浴上一层紫色的光晕。我周围有些人轻声交谈,但我们大多安静地坐着。在接下来的大约一个小时里,天篷的颜色缓慢但可感知地变化着,天空则紧抓着它的蓝色不放,然后,在最深沉的墨蓝色中徘徊了宝贵的几分钟后,似乎将一切全然交出。不知怎的,仅凭一个方形的洞,特瑞尔就将天空拉近了,使这最熟悉的景象变得陌生,让你带着爱意与关切,凝视那过于宏大、过于遥远而无法去爱的事物。没有什么是对你隐藏的。魔法就在于没有魔法。如果我现在不告诉你,没人会知道我曾在那里。 ♦

2 free previews left — register to get 30 translations a month