可控之梦 Operational, yet Oneiric｜卡西· 瑞斯 Casey Reas

Cassy Reas，Untitled Film Still 4-19，2019/2023，影像装置

今天的大模型正在疾速证明机器在图像生成的精度和准确性方面的指数级提升，用似假还真的产物疯狂地感染着整个互联网，这些当下的神经网络被不断添加了更精密的可控性，直到它们有一天完成「所令即所得」的许诺。而与此同时，神经网络不再那么像「梦」——「梦一般」（oneiric）并不在于图像本身的光怪陆离或扑朔朦胧，而在于：你无法真正操纵它，而它的降临所触发的情感亦是不可预期的，甚至让人困扰、惶惑或莫名欣喜的。机器学习不断被修剪出的、更可被预期和判断的形状，和潜在空间运算本身的「梦」属性，似乎正在被拉向更为遥远的两极。机器在今天更被期待于生产真实，或者至少在真实的渐近线上滑动，而不被鼓励继续生产梦境。或者更准确地说，机器所生产的「真实性」和「梦」似乎无法纠缠，前者是重要的介入世界治理的砝码，而后者过于飘渺无依。

Cassy Reas，Untitled Film Still 4-20，2019/2023，影像装置

更久之前，罗兰·巴特在《走出电影院》[1]一文里以极度现象学的方式描述了梦境如何近似于电影的底层媒介（我们通过梦境而理解电影），尽管从技法的层面，电影可以不断生产真实的镜像，但从经验层面，我们又是被它催眠的，「在这个不透明的立方体中……」进入一场「昏暗的幻梦」。亚历山大·加洛韦（Alexander Galloway）在《界面效应》(The Interface Effect）[2]里试图描述计算机的媒介特征时，也曾回到电影。在他看来，计算机「实现的是一种实践而不是一种存在，一种效果而不是一种对象」，故而它不像电影那样是本体论（尤其是现象学）的。计算用逻辑关系和规则去简化和模拟了本体论的维度，故而，「存在」成为了计算机运作的对象而非体验（这在计算机术语中也被称为「面向对象的编程」），加洛韦故而认为：既然在计算机这里，「存在」仅仅是对象，那「体验」则需要被从他处寻找。这或许可以解释为何「真实性/梦」在电影里可以彼此逼近，纠缠，互为彼此，而在普遍意义上的机器学习里却是分离的——即使那些现象学意义上的，梦一般的感受曾经浮出水面，也迅速湮灭在了参数、逻辑和网络的运作里。即使神经网络寻找数据规律的模式本质上极其像梦，这并不意味着它提供「梦的经验」。

Cassy Reas，Untitled Film Still 5-15，2019/2023，影像装置

在展览「可控之梦」里，算法艺术的先驱实践者卡西· 瑞斯（Casey Reas）选择展出他两个系列的作品，《无名电影剧照》和《Network E》。如瑞斯长期以来的创作，这两组作品都由算法生成，没有开头，也没有结尾。《无名电影剧照》来自瑞斯针对生成对抗网络（GAN）所展开的，为期三年多的实验「压缩电影」（Compressed Cinema），它是对肯·雅各布斯1969年电影《吹笛人的儿子》的反转：在这个系列中，每一部作为源头素材的长片被机器提取和浓缩为不到十分钟。事实上，GAN完全可以生成高保真的、摄影级的图像，但《无名电影剧照》却似乎有意地驶向了它的反面——计算机并没有复刻电影的「真实感」，而是捕捉了它的「梦一般」：模糊的脸，扭曲的符号，不明确的转折，和无边际的混沌。用列夫·曼诺维奇（Lev Manovich）的观点来说，这似乎符合一种计算机特有的关于「变形」的美学[3]，亦即，图像在没有电影意义上的切割或溶解的情况下，自动生长并扭曲成另一个图像。然而，跟去尝试定义这种算法媒介特殊性（algorithm specific）的美学比起来，似乎这些「无名电影」所释放出的「梦一般」气息更值得徜徉。这种气息与电影之梦很显然有所不同，但它是否仍然可以是「梦」？

Cassy Reas，Untitled Film Still 4-20，2019/2023，影像装置

在瑞斯的Github主页上，一直有一个名为「作品」（the Work）的页面，在这个页面上，瑞斯询问着对于一个算法艺术家来说，「作品」的本质是什么——软件，软件背后的代码，都不是「作品」，「作品」是界定系统的规则。对任何一个以代码作为工具的艺术家来说，规则的可控性始终是不可或缺的。「规则」的说法其实是对代码本质主义的松动（规则不是代码，但可以被代码传递），它潜在构建了不同媒介之间的通道。在这个展览的语境里，梦不是浮在表层的审美或感官体验——否则，我们可以轻易地说黑盒子更能造梦，因为不可逃逸，或者光怪陆离的生成图像就是梦，因为与清醒时所见毫无关联。梦更像是我们或许在从电影到人工智能，再到将来可能出现的计算技术形态中，都仍然眷恋的一种经验，这种经验足够底层，以至于我们可以暂时越过媒介特殊性，和不同技术之间的张力，去尝试捕获它。在《界面效应》中，加洛韦提出过关于「伊利斯」（Iris）和「赫尔墨斯」（Hermes）的比喻，他们所传递的是同样性质的神谕（规则），而作为媒介，伊利斯是「内在的」（immanence），赫尔墨斯是「表达的」（expression）。「梦」更接近这里的神谕或规则，而伊利斯、赫尔墨斯或者其他的转译区别在于以隐含还是明朗的，以忠实还是演绎的方式中介（mediate）了梦本身。在这个意义上，机器学习可以创造出可控之梦。

Cassy Reas，Untitled Film Still 5-19，2019/2023，影像装置

在走廊尽头的LED屏幕上，极为细微的白光从屏幕的深处亮起，一串一串点状绽放，直到屏幕被这些蒲公英状、或病毒状、或无声烟火形状、或神经元状的白色物质填满——是的，我正在尝试解这个由算法控制的深梦。在对既有的真实语法进行无尽复制和再生产之外，也许机器学习的分岔路上，还有一条或者很多条小径，它们不断化为梦境，却同样由「真实世界」的全部数据所承托。

Nearly ten years ago, Alexander Mordvintsev used a neural network algorithm to generate an uncanny image as if seen through the eyes of a computer. He uploaded the photo at around 2 a.m., and what followed was a series of cognitive avalanches: neural networks could observe patterns imperceptible to humans or even create creatures that defy human classification. This project from ten years ago was given a very subtle name: Deep Dream.

Foundational models nowadays are rapidly demonstrating the exponential improvement in precision in image generation, flooding the internet with hyper-realistic generative contents. These contemporary neural networks are being continuously enhanced with more refined operationality, driving ever closer to fulfilling the promise of 'you get what you prompted.' However, at the same time, neural networks are becoming less like dreams. The oneiric quality lies not in the fantastical or elusive nature of the images generated, but in the fact that you cannot truly control the generative process - only triggering emotions that are unpredictable, often leaving one confused, disturbed, or inexplicably delighted. Machine learning is being increasingly customised towards more predictable and understandable states, while the oneiric quality of latent space seems to be positioned at the other pole of operationality. Machines are more expected to produce reality, or at least slide along the asymptote of reality, rather than continue generating dreams. The reality and dream produced by machines seem to be diverging, with the former becoming an essential tool for world governance, while the latter is seen as too ephemeral and ungrounded.

Casey Reas, Untitled Film Still 5-16 ,2019/2023, 影像装置

Roland Barthes, in his essay 'Leaving the Movie Theater,' [1] described in an extremely phenomenological manner how dreams can be seen as a medium or substrate of cinema (we understand cinema through the lens of dreams). Although, from a technical standpoint, cinema can continuously mirror reality, from an experiential standpoint, we are hypnotised by it when 'entering a dim reverie' in this 'opaque cube…' Alexander Galloway, in The Interface Effect [2], also revisited cinema when trying to describe the medium specificity of the computer. In his view, computers 'instantiates a practice not a presence, an effect not an object,' and therefore, they are not ontological (especially not phenomenological) like cinema. Computation simplifies and simulates ontological dimensions through logical operations and rules, making being an object of computation rather than an experience (in computing terms, this is also known as object-oriented programming). Galloway thus argues that since 'being' is merely computation's object, we ought to look elsewhere to try to understand its experience. This might explain why reality/dream can converge, intertwine, and mutually influence each other in cinema, but remainseparate in the broader context of machine learning—even if phenomenological, dream-like sensations occasionally surface, they are quickly drowned in the relentless operations of parameters and networks. Even though the pattern-seeking nature of neural networks is inherently oneiric, this does not mean they provide a dream experience.

In the exhibition Operational, yet Oneiric, algorithmic art pioneer Casey Reas chose to display two series of his works, Untitled Film Stills Series and Network E. As with much of Casey Reas's long-standing practice, both sets of works are generated by algorithms, with no beginning and no end. The two works both reflect Reas's long-standing practice ofusing code as a medium and exploring the aesthetic experience of code.

Casey Reas, Untitled Film Still 5-21 ,2019/2023, 影像装置

The Untitled Film Stills Series is the result of a three-year experiment titled 'Compressed Cinema', which explores Generative Adversarial Networks (GANs). This project serves as a reverse take on Ken Jacobs's 1969 film The Piper's Son. In this series, each feature-length film used as source material is extracted and condensed by the machine into less than ten minutes. Although GANs are fully capable of generating high-fidelity, photo-realistic images, Untitled Film Stills Series seems to deliberately steer in the opposite direction—the computer does not replicate the realism of cinema but instead captures its oneiric quality: blurred faces, distorted symbols, ambiguous transitions, and boundless morphing. In Lev Manovich's terms, this aligns with a specific aesthetic of 'morphing'[3], a technique facilitated by the computer, making 'montage no longer central or even necessary, as one image grows and warps into another without a cut or even a dissolve in the cinematic sense'. However, rather than attempting to define this algorithm-specific aesthetic, it seems also worthwhile to immerse oneself in the oneiric atmosphere that these non-existing movies exude. This atmosphere is clearly different from the dreamlike quality of cinema, but can it still be considered a dream?

Casey Reas, Untitled 2 kiss me ,2019/2023, 影像装置截帧

On Casey Reas's GitHub page, there has long been a section titled 'the Work,' where Reas questions the essence of work for an algorithmic artist. Neither software nor the code behind it constitutes the work; rather, the work is defined by the system's rules. For any artist using code as a tool, the operationality of these rules is essential. The notion of rules subtly loosens the essentialist grip on code (rules are not code, but they can be conveyed through code), potentially creating pathways between different media. In the context of this exhibition, dreams are not merely superficial aesthetics or sensory experiences—otherwise, we could easily argue that the black box is more capable of generating dreams because it is immersive and inescapable, or that bizarre images generated by foundational models are dreams because they bear no relation to consciousness. Instead, dreams resemblean experience that we may continue to long for across various forms of technology, from cinema to artificial intelligence, and possibly even in future computational forms. This experience is foundational enough that we can momentarily bypass the specificity of the mediums and the tensions between different technologies in an attempt to capture it - the dream experience. Back to The Interface Effect, Galloway uses the metaphors of Iris and Hermes, who deliver oracles of the same nature (rules). As media, Iris is immanent, while Hermes is expressive.  If dreams are these oracles or rules, and the distinction between Iris, Hermes, or any other form of medium/technical translation lies in how they mediate the dream itself—whether implicitly or explicitly, faithfully or interpretively. In this sense, machine learning can create an operational dream. 

At the end of the corridor, on an LED screen, an extremely subtle white light begins to glow from the depths of the screen, blooming in small clusters until the screen is filled with these dandelion-like, virus-like, silent firework like, or neuron-like white structures—yes, I am trying to decipher this deep dream controlled by an algorithm - the Network E. Beyond the endless replication and reproduction of existing real-world syntax, perhaps there is a fork in the path of machine learning, or perhaps there are many paths that continually transform into dreamscapes, yet are equally supportedby the entirety of real-world data, its operational, yet oneiric. 

卡西·瑞斯（1972年生）是一位美国艺术家和教育家，现居洛杉矶。他通过当代软件的视角进行创作，作品涵盖从小型纸本作品到城市规模的装置艺术。2001年，他共同创立了Processing，这是一种为视觉艺术设计的开源编程语言和环境。他还共同创建了Feral File，一个用于展示和收藏数字艺术的在线平台。

瑞斯的软件、版画和装置艺术曾在美国、欧洲和亚洲的多个个展和群展中展出，并被多家博物馆永久收藏，包括乔治·蓬皮杜艺术中心、惠特尼美国艺术博物馆和旧金山现代艺术博物馆等。他近期的个展包括在洛杉矶郡艺术博物馆（Los Angeles County Museum of Art）举办的《一个空房间》（An Empty Room，2023年）；在柏林数字艺术博物馆（Digital Art Museum，DAM）举办的《幻化地形》（Conjured Terrain，2023年）；以及在纽约bitforms画廊举办的《不存在的其他形式》（It Doesn’t Exist (In Any Other Form)，2023年）等。

瑞斯现任教于加州大学洛杉矶分校（UCLA）的设计媒体艺术系，同时共同领导“社会软件”（Social Software）项目，旨在研究通过软件进行创作的社会影响和可能性。他拥有麻省理工学院（MIT）的媒体艺术与科学硕士学位，以及辛辛那提大学设计、建筑、艺术与规划学院的学士学位。

龙星如，策展人，写作者，博古睿学者，业余无线电操作员。策划展览包括《撒谎的索菲亚和嘲讽的艾莉克莎》，首届北京艺术双年展科技艺术板块《大地热流：回到太阳时间的访客》，《时间的幼虫》（博古睿研究中心，上海纽约大学）等。研究发表于瓦尔堡研究所，伦敦大学学院高等研究院，ZKM媒体艺术中心“艺术与人工智能”会议，ISEA电子艺术研讨会等，兼职任教于中央圣马丁艺术学院空间叙事专业。个人网站：irislong.xyz

Iris Long is a writer, curator, a Berggruen Fellow and an amateur radio operater. She has curated/co-curated exhibitions around art, science and technology, such as  “Lying Sophia and Mocking Alexa”, the art&tech sector of the inaugural Beijing Art Biennial, “Earth Heat Flow: the Visitor Who Returns to Solar Time”, “The Larva of Time”(Berggruen Institute, Institute of Contemporary Arts at NYU Shanghai). Her research has been presented in “Space in Time: From the Heavens to Outer Space”(Warburg Institute), UCL Institute of Advanced Studies, “Art and Artificial Intelligence”(Open Conference, ZKM), ISEA and so on. She teaches part time at Central Saint Martins, University of Arts, London. More info: irislong.xyz

泥轩 ｜ MUD Gallery

上海市普陀区莫干山路50号6号楼5楼

5 / F, Building 6, 50 Moganshan Road, Putuo District, Shanghai

周二至周六 Tuesday to Saturday 11:00 - 17:30

+ 86-18217215543

小红书: 泥轩 MUD Gallery      

Instagram: mud.gallery      

微信公众号：MUD Gallery