River Swimming in a Salmon

In the new cosmic era of “post-pandemic” 2022, an AI image generator was given the prompt “salmon swimming in a river.” It produced several puzzling images of similar composition, each of which shows what is unquestionably salmon in a stream of water. However, none of the results matched the familiar image of a legendary piscine hero swimming full-bodied against the current. Instead, they all depicted a fraction of a salmon, a fillet of salmon, its pink flesh ready to be seared or made into sashimi. In AI’s defense, the algorithm did not interpret the prompt incorrectly. All the right elements are there: this is not a pig in a volcano. Yet it all appears wrong to human eyes.

We might say that what went wrong was the montage. As a cinematic device, montage is not only that which connects discrete units—multiple images, signs, or scenes—but that which generates meaning out of their conjuncture and sets these elements in motion. Exhibition-making shares many similarities with the principle of montage.1 It is also a quintessential feature of our image-ridden era.

In the early twentieth century, Soviet film theorists and practitioners such as Sergei Eisenstein and Lev Kuleshov, as well as German art historians and cultural theorists like Aby Warburg and Walter Benjamin, settled on montage as a means for understanding the collapse of the old world and the imminent arrival of a new one.2 So what might AI-generated montage—as exemplified by this at-once bizarre, funny, and disturbing image-montage of salmon fillets in a river—tell us about where we are and where we are headed?

Upon seeing the AI-generated salmon images, most people would burst out laughing. A salmon fillet swimming in a river is very funny. But why? It is not just because such a scenario is realistically improbable. Few would laugh in the same way if the generator responded to the prompt “salmon dancing in a sushi restaurant” with an image of a fillet doing the twist against a background of oversized “floating world” prints. Nor is it only humorous because human intelligence seems in this instance to surpass artificial intelligence, proving it to be mere artificial stupidity. Instead, in revealing something more ambiguous and disturbing, the image provokes the laughter of relief. Is it a dark, Freudian prophecy?

Hui Tao, The Fall (The Legend of the White Snake), 2023 (new version). Silica gel, fiberglass, metal, paint. Width: 398 cm; tail: 65 × 95 × 22 cm. Courtesy of the artist. Photo: Power Station of Art.

Think of Ferdinand Zecca’s flying machine patrolling the sky over Paris before the invention of the spaceship, or Salvador Dalí’s Lobster Telephone (1938). Or go back to the dawn of cinema and recall Jacques Aumont’s account of early critics’ distress at montage: “Anything can come after anything else … from a horse rider to a girl, a fly to an elephant, the North Pole to the Sahara Desert.”3 Considering Dalí’s (in)famous quip—“When I ask for a grilled lobster in a restaurant, I am never served a cooked telephone”—what would he make of what AI has served us?4 With its salmon fillet swimming in a river, has AI-the-artist achieved Dalí-esque levels of surrealist art-making and nonsensical posturing? Even better: Has it mastered the estrangement effect which, according to Russian formalists, is what makes art artistic?

What about AI-as-curator? Taking things apart and out of their contexts is a feature of what Anselm Franke calls the “objectifying, immobilizing, mummifying device” of the modern museum.5 The salmon fillet in the river might thus be typical of the disenchanted, mechanistic, categorically hierarchical order of things that the imperialist institution supports. Rather than producing anything revolutionary, AI-the-curator reinforces—and might even push further—a modern/colonial logic of border-making and confinement. Rather than imprisoning living creatures within artificial boundaries, it cuts them to pieces. As Los Tigres del Norte sing of the US-Mexico border: “Yo no crucé la frontera, la frontera me cruzó” (I didn’t cross the border, the border crossed me).6

The challenges of today’s curatorial practices are vividly incarnated (pun intended) in this image: not only is the salmon taken out of its habitat, it is also cut up, objectified, immobilized, and mummified in its artificial pink-orange glow. As an institution of categorization and knowledge production, the modern museum is closer to scientific reasoning than artistic creativity or iconological generativity. By “scientific reasoning,” I mean the process of taking things apart and investigating their nature, asking, “What is it?” Art and religion (and hopefully curation) instead put things together and ask, “What does it mean?” Montage opens the question out similarly: What meaning might the ensemble generate?

Let us dwell further on AI’s equation of salmon with a fillet. The algorithm that produced the fillet images had been fed a fragmented dataset in which the bourgeois vision of the world is overrepresented to the extent that it becomes the “full picture.” (Algorithms are the new organizers of the capitalist cosmos.) In this context, a living salmon’s automated reduction to a ready-to-eat salmon fillet might exemplify the effects of labor alienation. While most readers can perfectly picture a salmon fillet with its pink-orange color and oval shape, they seem less likely to immediately identify a salmon in its unsliced entirety.

A fillet as the primary visual signifier of “salmon” might tell us more about ecologically alienated capitalist humanity than any fantastical speculation on the “AI apocalypse.” The dataset that prioritizes the fillet over the fish reflects the overrepresentation of a certain “ethno-class,” in Sylvia Wynter’s term.7 In fact, no one knows better what a live salmon looks like and how it behaves than the fishermen and the factory workers who actually do the farming and filleting of this globalized commodity. The visual joke, then, is on the global bourgeoisie.

Zhang Wenxin, Notes of the Hollow series, 2021–present. Photograph, stainless steel mirror, 3D print. Overall dimensions variable. Courtesy of the artist. Photo: Power Station of Art.

Montage conjoins discrete elements (which could be understood as always already a montage) and sets them in motion. One element is missing in both the AI-generated image and in our analysis of the original prompt: the verb “swimming.” Of course, the act of “swimming” cannot be shown but only inferred in a still image—for example, by a twisted fin or a splash of water. And while it may seem absurd that a slice of salmon fillet should be pictured in a stream of water, it is not inconceivable: had this image been constructed by a (human) artist, we might read it as a satire on consumer alienation. Much more troubling is the implication that the connector “swimming” was willfully dismissed in the image generator’s interpretation: only a living creature could be swimming.

What happens when the dynamic connector—the single word “swimming”—is highlighted and reintroduced into the sentence? The ability to fill in gaps is the basic cognitive skill through which montage operates: the combination of “salmon” and “river” requires a verb to make sense of it (“jumping,” “flying,” “plunging,” or the more poetic “playing” or “dancing”). The phrase could even be reversed as “river swimming in a salmon.” As visual syntax, montage connects myriad forms and images (“cosmos”) and sets them in motion (“cinema”). The myriad forms are not only infinite but infinitely editable. Proto-cinematic attempts to recode the cosmos—from reading the fate of an individual in the stars to consulting the heavens to know when to till the land—recognize this editability. Since the “dawn of everything,” a variety of cultural practices in the spheres of religion, economy, and agriculture have recoded the cosmos in this way.8 In this sense, montage as a cinematic technique has a much longer history than the modern technology of filmmaking. These successive “montages” have fundamentally changed the way the world looks and works.

At the same time, the AI-generated salmon images do not truly form a montage but rather a démontage, a dissection: they disassemble an assemblage, stop a motion, take things out of context. Maybe rapidly evolving AI has not only caught up with but surpassed human intelligence, even in its image-making capacity. In this sense, perhaps the salmon fillet in the river was AI’s flippant commentary on consumer capitalism. Speculating on what AI has been and could become is beyond the scope of this essay, but considering these images as démontage might indicate how our contemporary society is at the same time pervasively connected and profoundly segregated and fragmented. The modern/colonial museum and progressive politics alike fall prey to decontextualization, reductionism, and essentialized (self-)identification.

In the 14th Shanghai Biennale, entitled “Cosmos Cinema,” montage is understood both as a device to observe the world (“cosmos”) and as an organizing principle—an onto-epistemology that sets the world in motion and re/organizes it (“cinema”). This exhibition-as-film is narrated in nine successive “palaces” (九宫图). The physical arrangement of “palaces” refers to a mythic Chinese cosmological diagram laid out in the manner of a three-by-three Rubik’s Cube that structures heaven and earth, time and space, and yinyang, creating complex dynamics among them.9 This premodern, montage-like structure of nine palaces is still in use today in a variety of practices that map out the myriad connections between the mundane world and the workings of the cosmos.10

Kidlat Tahimik, Perfumed Nightmare, film still, 1977, 93 minutes. Courtesy of the artist.

Cui Jie, (from left to right) International Space Station; The Rainfall Pavilion and the Lakeside Pavilion; Base Station #2, 2019. Acrylic and spray paint on canvas. Courtesy of the artist and Pilar Corrias, London. Photo: Power Station of Art.

Mao Chenyu, The Cosmists of the Ximao Household, 2004–23. 4K video, color, Qing Dynasty wooden furniture, glass, pine resin, 9:27 minutes. Courtesy of the artist. Photo: Power Station of Art.

Zijie, SHENZHI LEE on the science island: Project Chinese word processing, on writing, power, capital and technolog, 2020–present. Multimedia installation. Dimensions variable. Courtesy of the artist. Photo: Power Station of Art.

Zhou Xiaohu, Ship of Fools, 2018. Courtesy of the artist. Photo: Power Station of Art.

Marina Camargo, Distance within Space, 2023. Installation with light, projection, and cutout brass. Dimensions variable. Courtesy of the artist. Photo: Power Station of Art.

Elda Cerrato and Luciano Zubillaga, Okidanokh, 1964–2022. Single-channel video, 8:09 minutes. Elda Cerrato Archive (ECET).

Elda Cerrato and Luciano Zubillaga, Ocultamientos IV, 2023. Print on fabric, 10 pieces. Each 300 × 110 cm. Elda Cerrato Archive (ECET). Photo: Power Station of Art.

Kidlat Tahimik, Perfumed Nightmare, film still, 1977, 93 minutes. Courtesy of the artist.

“Cosmos Cinema” ends with “brave wind and rain,” a “palace” that could have included the upstream-beating salmon—if not its aquacultured simulacrum. “Brave wind and rain” (栉风沐雨), an expression taken from the Zhuangzi, one of Taoism’s foundational books, refers to the story of Yu (大禹), a semi-mythical figure who manages to tame a devastating recurrent flood. This story is entangled with the river diagram hetu and the Luo River Chart that gave rise to the nine palaces. Yu is said to work so diligently in his efforts against the flood that he refuses go home. His refusal constitutes a break with the “homecoming rule” hailed through the generations as a Confucian virtue. Instead, Yu combs his hair with the wind and showers in the rain.11 The expression “brave wind and rain” thus came into use to praise diligent labor, like that of the peasant cosmists in artist Mao Chenyu’s video series Ximaojia Universe (2004–). The peasants observe the stars, decode and recode the wind and water of the nearby mountain and river, cultivate the land, grow grains, and sustain life’s flourishing.

The wonder of taming a flood could have been achieved by any anonymous person, or more likely, a collective of people. But stories and myths of bravery and diligence tend to foreground a singular, heroic figure like Yu. The legacy of capitalist modernity’s death cult, which is bringing organized life on earth to the brink of annihilation, is to seduce us into hoping for a heroic figure to lead an exodus: a supreme leader, a philosopher-prophet, a visionary business tycoon, a genius artist … even a charismatic curator. Walter Benjamin links this auto-instituting hero mythopoeia to the conditions of art in the age of its technological reproducibility. In a footnote in his well-known essay, he states that “the crisis of democracies can be understood as a crisis in the conditions governing the public representation of the political man … This results in a new form of selection—selection before an apparatus—from which the champion, the star, and the dictator emerge as victors.”12

“Cosmos Cinema” is a collective endeavor and a collective speculation on where we have been, who we are, and what we might become. It intentionally decentralizes the production of meaning through a loose deployment of cinematic and proto-cinematic means of montage. The nine “palaces”—thematic clusters—into which the exhibition is arranged might be understood as a “director’s cut,” offering one of many possible edits of these thematic units. Structuring the exhibition’s meaning-making mechanism through this at-once cosmic and (proto-)cinematic device foregrounds associative freedom and invites visitors to reshuffle, re-edit, and ultimately re-curate the exhibition. This arrangement aims to ensure that no authority can dictate a singular narrative about or impose a fixed meaning onto this cosmos-cinematic exploration. What does it mean when the AI salmon fillet encounters in the river a dragon-horse carrying a divinatory cosmological diagram? After all, the origin of the nine-palaces cosmography—often used as a divination tool—is a river: in the mythic image of the Yellow River hetu (河图) carried by the dragon-horse (龙马). However, there’s no salmon there to speak of.