Rethinking World Literature in the Age of AI Peina Zhuang, Péter Hajdu Sichuan University, College of Literature and Journalism, 24 First South Section, First Ring Road, 610064 Chengdu, China https://orcid.org/0000-0003-3278-4009 alison19831208@163.com Shenzhen University, School of Foreign Languages, 3688 Nanhai Avenue, Nanshan District, 518060 Shenzhen, China / HUN-REN Research Center for the Humanities, Tóth Kálmán Street 4, 1097 Budapest, Hungary https://orcid.org/0000-0002-3623-1578 pethajdu@gmail.com The debate about world literature has never stopped since the term was first introduced in a conversation between Goethe and his secretary almost two centuries ago, and it has intensified in the age of AI, which is bringing about changes in almost every aspect of society, including world literature. Specifically, while AI brings enormous benefits to the dissemination and translation of world literature, it also creates challenges and problems. Based on the analysis of the translation of literary texts across different cultures, this paper argues firstly that AI may contradict David Damrosch’s claim that “world literature gains through translation,” as world literature can in fact “lose” something, e.g., literariness, as the examples selected in this paper show, and secondly that AI can also provide a “biased” picture of theories or works from different cultures. We should therefore be wary of these two risks, otherwise we could be misled in this “technological wave.” Keywords: literature and AI / AI-generated poetry / translation / literariness / ChatGPT / Lem, Stanisław 183 Primerjalna književnost (Ljubljana) 47.3 (2024) The term “world literature” made its “official” debut in Goethe’s fa- mous conversation with his young disciple Johann Peter Eckermann in 1827, after he read the translation of a Chinese novel. Ever since, the discussion on world literature has never stopped, from Goethe, Marx and Engels, all the way down to contemporary scholars, includ- ing David Damrosch, Haun Saussy, Ning Wang, and others who have contributed in various ways to this theoretical topic from different perspectives and backgrounds. Each stage of the changing conception and theorization of world literature is concomitant with the previous PKn, letnik 47, št. 3, Ljubljana, november 2024 184 industrialization, economic capitalization, and globalization. The de- velopment of world literature as circulation has never been indepen- dent from changes of technology and media that naturally determine the accessibility of all literatures. The drop of book prices around 1800 encouraged international exchange in literature. When the press be- came a mass medium in continental Europe (but also on other conti- nents) in the second half of the nineteenth century, publishing litera- ture on a daily basis, the huge demand for literature created a massive market for translated novels, and the new literary agencies were by no means restricted by borders of national markets or languages. In our days, the invention and development of AI, such as ChatGPT, might create a new media context for world literature, which invites us to revisit the old notion. In this paper, we analyze how world lit- erature comes to be influenced by this “technological wave,” as mani- fested in the aspects of literary production and translation. We argue that despite the unprecedented “accessibility” and “availability” that AI offers people in reading, circulating and translating world literature, it is rather “biased” in presenting theories or works from different cul- tures, and sometimes fails to retain the “foreignness” either in literary creation or translation. In this sense, AI may unexpectedly run counter to the notion that David Damrosch enshrines in the concept of world literature as “writing that gains in translation” (Damrosch 281) since it might lead to the circulation of a homogenized world literature, which might filter out the foreignness of the source text. Before the widespread use and application of large language models (LLMs), human beings used to be the subjects that created world lit- erature. The invention of LLMs creates or can probably create a differ- ent situation, at least in some scholars’ view. ChatGPT, as an artificial intelligence language model with a powerful knowledge base, has the ability to write academic papers, as well as to create poetry, prose, and novels. Thus, as Yingying Zhang suggests, “in terms of future devel- opment, the birth of ChatGPT is not only a technological and cul- tural phenomenon, but also a literary phenomenon. Many people have already started ChatGPT’s game-like literary attempt, which will bring about a milestone technological revolution” (Y. Zhang 1). Following this vein, scholars argue that today’s “game-like literary attempt” may one day turn into a reality of world literature; in contrast, our com- parative study of literary works created by ChatGPT and those by real humans proves that this will not happen immediately. Writers invest their own experience and imagination in literary cre- ation, resulting in various types of representation and style. The growth Peina Zhuang, Péter Hajdu: Rethinking World Literature in the Age of AI 185 of ChatGPT suggests that the subject of literary creation may shift from humans to machines. Works created by ChatGPT are rich and diverse with sound external forms, such as rhyme in poetry, language with a sense of musical beauty, and novels with characters and plots. It can produce works with a certain aesthetic quality, which raises the alarming possibility of replacing human literary creation. However, form is not the only standard in judging the quality of literary works. Chinese philosopher Baihua Zong (1897–1986) once stated that when people appreciate art such as literature, they often “tend to lean towards the content, setting, story, and acts of life in the art, rather than artistic ‘form’ alone” (Zong 31). To study literature is to study human beings. AI-generated poetry became a literary topic with Polish sci-fi writer Stanisław Lem (1921–2006). He can rightly be regarded as a world- literature figure, whose books have been translated into more than 40 languages, and on whose 1961 novel Andrei Tarkovsky based his clas- sic film Solaris (1972). In his 1965 short story cycle Cyberiada (The Cyberiad, English translation 1974), exclusively populated by robots, constructor Trurl builds the electronic bard (Elektrybałt), a machine designed to create poetry. We will return to its poems later, but it is helpful to see already here how Trurl sees the problem of writing a program to create poetry: The program found in the head of an average poet, after all, was written by the poet’s civilization, and that civilization was in turn programmed by the civilization that preceded it, and so on to the very Dawn of Time, when those bits of information that concerned the poet-to-be were still swirling about in the primordial chaos of the cosmic deep. Hence in order to program a poetry machine, one would first have to repeat the entire Universe from the begin- ning—or at least a good piece of it. (Lem, Cyberiad 43–44) Lem appears as a true structuralist or even post-structuralist critic here, one who is maybe too quick to disregard the individuality of an author and ascribes the creative process to the cultural codes and the semio- sphere. Trurl repeats most of the Universe to program his bard, and this is probably the reason why it becomes the greatest poet ever. It is only a small exaggeration to say that ChatGPT’s access to a huge part of humankind’s textual heritage makes it similar to Trurl’s electronic bard, yet it still lacks human experiences and emotions, with profound ideological connotations, which literature often reflects. The data-based working mechanism means that “the ability of ChatGPT’s literary production mainly lies in its mastery and applica- tion of literary forms, but not the capability to create content to be PKn, letnik 47, št. 3, Ljubljana, november 2024 186 represented. … ChatGPT’s production lacks not only emotions, but also subjectivity in intervening in the real world” (S. Zhang 44). The fundamental difference between the process of creating literature used by ChatGPT and by humans lies in that the former lacks real feelings, emotions, experience, and understanding of life, without which all the knowledge and texts of human history become “flattened” in the pro- cess unless a human subject intervenes. Even for this result we had to ask AI to improve the “form” after inputting “prompts” several times, and the result tends to be much less satisfactory from the viewpoint of form in languages other than English. AI will possibly be able to model human subjectivity in the future and produce high quality literature, but world literature might need something else, which we will call universality. Ning Wang differen- tiated between “world literature in general and world literatures in particular, the former referring to a universal criterion for evaluating literature of the greatest world significance, the latter to the different representations, including translations of literatures from all countries” (N. Wang, “‘Weltliteratur’” 297). Wang’s concept of “world literature in general” implies a canonical approach, but it refers to the canonical ways of appreciating and selecting literary works rather than a con- crete recommended set of canonized works. While we doubt that there exists “a universal criterion for evaluating literature,” we accept that a literary work needs to contain something universal to be included in the category of the “general” as opposed to the particular, which is the all-encompassing category of every piece of literature written any- time, anywhere and in any language. This universal, however, must be expressed in a particular literariness.1 Literary works create “worlds” different from the geographical one, and “there are a great many imagi- native projections of possible and conceivable worlds; and people spend most of their lives in those latter worlds” (Saussy 293).2 Any great piece of literature creates its own imaginary world, which is both embedded 1 For the universal as opposed to the national, see Ursa. For further insights about literature and universality, see Duhan et al. 2 Haun Saussy uses the concept of “world” in its Heideggerian meaning (Hei- degger 170). This concept was consciously and explicitly “vulgarized” to adapt it to the postcolonial field by Gayatri Chakravorty Spivak (Spivak 260). Since then, “worlding” rather means getting embedded in the globalized world system than the individual literary work’s creation of an experience in which we can feel at home. Eric Hayot discusses the applicability of the Heideggerian concept in the third chapter of his book On Literary Worlds, but otherwise sticks to the postcolonial meaning, as does Pheng Cheah. For the history of this transition, see Juvan 16–19. Peina Zhuang, Péter Hajdu: Rethinking World Literature in the Age of AI 187 in and yet independent from the traditions of certain regions, cultures and social circles. David Damrosch’s definition of world literature as “writing that gains in translation” (Damrosch 281) appears so original and insightful precisely because it turns upside-down the truism that texts tend to lose a lot in translation. Gaining something, however, does not exclude losing something else. And gaining does not necessar- ily follow from the intrinsic merits of the given writing: it might hap- pen accidentally, and it would be that accident which made something world literature. However, the dialectics of familiarity and alienness described by Schleiermacher must be at work here too: the pure univer- sality of a literary work might no longer excite when it travels to a new place, while pure alienness makes it incomprehensible (Schleiermacher 5–158)—unless it gains some universality on the way. Therefore, from the viewpoint of intrinsic merits, Marshall Brown’s comment seems valid, namely that “world literature … is not writing that gains in translation, but writing that retains its alienness even in the original” (Brown 364). And it is this foreignness that current AI literary produc- tion still seems to lack. Let us return to Lem’s electronic bard. At the beginning, Trurl’s friend Klapaucius is rather skeptical about AI-generated poetry (espe- cially because the electronic bard’s first poetic attempts are rhythmic- rhyming gibberish)3 and tests the bard by giving it a complicated assignment to write poetry. We could not resist the temptation of giv- ing ChatGTP the same assignment, in three languages. Lem’s transla- tors, of course, did not translate the assignment literally, but obviously created their poems first and adapted the assignment to the already made poem, and it is probable that Lem proceeded similarly. This is the prompt for the poem in the Polish source text: Niech ułoży wiersz o cyberotyce! … Żeby tam było najwyżej sześć linijek, a w nich o miłości i o zdradzie, o muzyce, o Murzynach, o wyższych sferach, o nieszczęściu, o kazirodztwie, do rymu i żeby wszystkie słowa były tylko na literę c!! (Lem, Cyberiada 45) Write a poem about cyberoticism! … There should be at most six lines, about love and betrayal, about music, about Negroes, about high society, about mis- fortune, about incest, with rhymes, and all the words should only start with the letter c!! (our trans.) 3 It is difficult to evaluate the poetic qualities of gibberish, but in Beatrix Murányi’s Hungarian translation the electronic bard’s first attempts show some poetic strength. They are meaningless, but in a strange dadaist way they sound like poetry, suggestive of some abject topic. PKn, letnik 47, št. 3, Ljubljana, november 2024 188 When we gave these instructions to ChatGTP, it generated the follow- ing poem: Czuły czar zdrady czai się w cieniu nocy, Cyganie śpiewają o miłości, marzeniach rozmów, Czułość, cicha carmina, w serce wprowadza młodość, Cierpienie, chwilowe chwile, ciche żale snów. Czuję, cichutko czuję, ciosy czasu ciche, Ciemność czaruje, czystość chwiejna, czarny chwilowy chichot. The tender spell of betrayal lurks in the shadows of the night, Gypsies sing of love, dreams of conversation, Tenderness, silent carmina, in the heart brings youth, Suffering, momentary moments, silent griefs of dreams. I feel, quietly feel, the silent blows of time, Darkness enchants, shaky purity, black momentary giggle. What catches the eye immediately is that the alliteration is quite in- complete. Fourteen of the thirty-nine words of the poem (36%) do not start with the letter c. Only two lines do actually rhyme. Such a verse game is unlikely to produce great poetry, but the paratactic series of short clauses with very few verbs and little cohesion also indicate poor aesthetic value. A comparison with Lem’s own poem in the source text demonstrates how faulty these features are in the AI-generated poem: Cyprian cyberotoman, cynik, ceniąc czule Czarnej córy cesarskiej cud ciemnego ciała, Ciągle cytrą czarował. Czerwienała cała, Cicha, co dzień czekała, cierpiała, czuwała… …Cyprian ciotkę całuje, cisnąwszy czarnulę!! (Lem, Cyberiada 45) Cyprian the cyberotomaniac, cynic, cherishing tenderly The emperor’s black daughter’s miracle of dark flesh, Chanted his magic with a zither. She was turning red, Silent, every day she waited, suffered, stayed awake… …Cyprian kisses his aunt, rejecting the black woman!! (our trans.) All but one word begin with the letter c (co dzień, “every day,” can be regarded as a single syntactic unite, though), all the lines rhyme, and the three sentences narrate a complete story of betrayal. We have to admit that this was ChatGTP’s only attempt and we did not make it revise and develop the poem. After a while it could improve, but Lem’s electronic bard did not have a chance for revision either. Peina Zhuang, Péter Hajdu: Rethinking World Literature in the Age of AI 189 We performed the same experiment with the poems embedded in The Cyberiad’s Hungarian and English translations, and the results were similar. Here are the Hungarian descriptions of the task and human and AI-generated poems: Írjon kiberotikus verset! … Legfeljebb öt sor legyen, szóljon szerelemről, árulásról és halálról, a néger kérdésről és a nimfomániáról, legyen benne a bonyolult női lélek extrém konfliktushelyzetben bekövetkező meghasonlásának ábrázolása, a középkori feudális viszonyok és erkölcsök maró bírálata, rímeljen, és minden szó k betűvel kezdődjön! (Lem, Kiberiáda 150) Kóbor kaffer kószál királylány kertjében. Királylány kacéran kacsint kéjvágyó kedvében: Kapj karodba, kaffer! Király kinéz, kiált: Katonák! Kürtszó, kivégzés. Királylány kacag kuszán. Kegyetlen kor! Kicsapongó, koronás kurtizán! (150) Write a cyberotic poem! … No more than five lines, about love, betrayal and death, about the Negro question and nymphomania; it should represent the complexity of the female psyche in extreme conflict, also containing harsh cri- tique of medieval feudalism and morality, rhyming, and all words beginning with the letter k! (our trans.) A stray kaffir roams in a princess’s garden. The princess winks flirtatiously in her lustful mood: “Give me a hug, kaffir!” The king looks out, cries: “Soldiers! Tantara, execution.” The princess laughs confusedly. Cruel age! A promiscuous, crowned courtesan! (our trans.) The alliterations are perfect, the rhyming is not bad, the narrative is a bit lapidary, but not without some good phrasing. Here is what ChatGTP could create following the same instructions: Keserű szóval suttog szívem, kín, Középkor királynője, koronás királynők, Kígyóként kúszik, kacér, kozmetikázó kéj, Kapzsi, kötött köntös, kert végén kárhozat, Kedves kéjutazás, kétely közt kárhozat. With bitter words my heart whispers, anguish, Queen of the Middle Ages, queens with crowns. A coquettish, cosmetic lust crawls like a serpent; A greedy, knitted robe, damnation at the end of the garden, A lovely pleasure cruise, damnation in doubts. PKn, letnik 47, št. 3, Ljubljana, november 2024 190 Four words do not begin with the letter k; the only “rhyme” in the poem is a repetition of the same word at the ends of two consecutive lines; the poem does not have any characteristic rhythm which would really make it verse. However, its most faulty feature is the lack of any development of a theme: the paratactic list of unconnected clauses— some of them (such as the greedy robe) are utter nonsense—make it rather a list of k-words than a poem. The English translation by Michael Kandel set the task this way: “Have it compose a poem—a poem about a haircut! But lofty, noble, tragic, timeless, full of love, treachery, retribution, quiet heroism in the face of certain doom! Six lines, cleverly rhymed, and every word beginning with the letter s!!” (Lem, Cyberiad 50). Love, treachery and alliteration are the features retained from the Polish ST, while replac- ing Lem’s neologism cyberoticism as the required main feature of the poem with the motif of haircut, suggestive of a translator who had a particular topic in mind. The poem in the English translation explains a Biblical theme, i.e., a particular cultural reference to the probably most famous haircut ever: Seduced, shaggy Samson snored. She scissored short. Sorely shorn, Soon shackled slave, Samson sighed, Silently scheming, Sightlessly seeking Some savage, spectacular suicide. (51) Perfect alliteration, some good rhymes, and a concise retelling of the myth from the Judges. It is a small wonder that ChatGTP could not think of Samson’s story full of love, treachery and heroism. This is what it created following the same instructions: Silken strands surrender, shears swiftly swoon, Sculpting strands, sovereign, secrets soon. Serrated scissors sing, severing strife, Sacrificing serenity, silent sacrifice. Sorrowful strands fall, shorn, serene, Symphony of shearing, sublime, unseen. ChatGTP is obviously a better poet in English than in Polish or Hungarian, because it has access to much more data (text corpora) in English, which is problematic, as we will discuss later. Only one or two words break the alliteration, the rhymes are good, and some expressions Peina Zhuang, Péter Hajdu: Rethinking World Literature in the Age of AI 191 show some poetic quality. However, the same problem of the simple paratactic listing of clauses that do not develop a coherent content char- acterizes this assignment too. In Lem’s story, the electronic bard drives many poets to suicide through their desperation about achieving anything similarly great to what it could create, and after Trurl tries to isolate it in an asteroid, it broadcasts its poetry and disturbs the traffic because many spaceships’ crew members are “seized with severe attacks of esthetic ecstasy” (Lem, Cyberiad 56). At the moment, as it seems, we do not have to worry about such severe consequences of AI-generated poetry. In our other experiment we typed the words “spring, bloom, birds, lamb, countryside, daisy” into ChatGPT and asked for the composi- tion of a poem on spring. It created the following two poems: Spring Spring is here, the season of new birth The earth awakes from its long winter sleep The flowers bloom in every hue and shade The birds sing sweetly in the trees and fields The lamb is born, a symbol of pure joy It frolics in the meadow with its kin It follows its mother with trust and love It grazes on the grass and daisies white The countryside is filled with life and light The sun shines brightly in the clear blue sky The breeze is gentle and the air is fresh The stream flows softly with a soothing sound Spring is here, the season of new hope It brings us beauty, peace and happiness It reminds us of the wonder of creation It fills our hearts with gratitude and praise Spring Spring has arrived, the world awakes Beneath the blue, the earth now shakes Buds burst open, petals unfurl The promise of life, anew, unfurled PKn, letnik 47, št. 3, Ljubljana, november 2024 192 Beneath the sun, the fields turn green The countryside a joyous scene Birds return from distant lands Their song echoing through the strands Lambs gambol through the fields so gay Their carefree spirit on display And in the midst of all this cheer The daisy blooms, so pure and dear Spring has come, and with it hope A chance for life to start and cope With every bloom and every bird A promise of joy, now seen and heard. The quality is definitely above that of an average college student. The poems make good use of diverse images, sometimes in rhyming schemes, as in the second poem. The atmosphere of the first is livelier, that of the second more peaceful. Both evoke the beauty, joy and hope spring brings to people, but they still exhibit stylistic and structural differ- ences from poems by “real” poets such as Thomas Nashe (“Spring, the Sweet Spring,” 1600), William Wordsworth (“Lines Written in Early Spring,” 1798), Gerard Manley Hopkins (“Spring,” 1877), Christina Rossetti (“Spring Quiet,” 1862), Robert Frost (“Spring Pools,” 1916), and Edna St. Vincent Millay (“Spring,” 1921). The repetitiveness in diction, such as “awake,” “hope” and so on, and the fixed structure with a strict rhyming scheme stand out in both poems, displaying stark contrast with the intentional choice of the imi- tative words such as “jug-jug,” “pu-we,” and “to-witta-woo” in Thomas Nashe’s poem, which gives a lively and vivid picture of spring. Anyone with an aesthetic taste can strongly feel the machine-trace in both poems, lacking the human aura or energy and having no foreignness but rather a hackneyed one on spring. Human poets tend to surprise readers with some unexpected and, well, not mechanical features. For instance, the diction in the poem by Thomas Nashe is not that complicated, but one can strongly feel the lively tone of spring, and can even sing the poem due to its use of sound words and the reference to dancing maidens and frisking lambs. The mastery of sprung rhyme by Hopkins, the pondering on human’s des- tiny and its relation with nature in Wordsworth’s poem, the integration of aesthetics and religion by Rossetti and Frost, and the deep symbolic narrative by Millay are nowhere to be seen in the machine compositions. Peina Zhuang, Péter Hajdu: Rethinking World Literature in the Age of AI 193 Although those poets tend to use simple straightforward words rather than obscure, erudite ones to convey what they want to express, they still manage to create some unique literariness or alienness. St. Vincent Millay’s line “Not only under ground are the brains of men / Eaten by maggots” can be a case in point. The expression is close to every- day usage despite the iambs, but the content comes as a shock in the middle of a Spring-poem. It diverts the readers from the lively spring to a deeper inquiry, as to the meaning of spring or life, since “Beauty is not enough” (St. Vincent Millay). At this point we can recall the Russian formalists’ concept of “defamiliarization” as the essence of literature, since all the poems mentioned show a stark contrast to the too familiar tone of the machine-created poems. As Victor Shklovsky put it in 1917, “the technique of art is to make the objects ‘unfamiliar’, to make forms difficult, to increase the difficulty and length of perception because the process of perception is an aesthetic end in itself and must be prolonged. Art is a way of experiencing the artfulness of an object; the object itself is not important” (Shklovsky 12). “Defamiliarization” in the poem, or more broadly, in literature, adds a sense of heaviness, complexity and strong aftertaste to the poem, specially created to free the perception from “automatism.” When we contrasted AI- and human-composed poetry in the previ- ous passages, we disregarded the fact that human “intervention” per- meates the whole process of AI literary production, starting from data input. AI must be “initiated” in composition instead of starting itself automatically and habitually. Long before the debut of ChatGPT, Italo Calvino foresaw that machines could create through their own learning and progress. He regarded literature as a machine using language, but he emphasized that only a writer can make it work, since the machine “would not work without the spasms of an I who is immersed in his- torical time, without his reaction, his crazy happiness, and his anger of hitting the wall with his head” (Calvino 206). The sense of time, soci- ality, and emotionality of human writers are still beyond ChatGPT’s reach. However, ChatGPT’s unique creative approach can bring us a unique aesthetic perception, which we can use to further enrich our imagination and understanding of the human creation itself. Another high expectation about ChatGPT’s influence on world lit- erature lies in the prospect of “re-making” world literature due to the changes that befall the circulation mode of literary works, one aspect of which is the wide access to the above “peripheral” works and theo- ries. The emergence of ChatGPT has broken down the barriers and limitations of literary communication in terms of time, space, and PKn, letnik 47, št. 3, Ljubljana, november 2024 194 language. Written in the most widely used language in the world, works in English often have an advantage in disseminating and cir- culating easier worldwide. Works in languages other than English are often more restricted in circulation, which makes their position in world literature disadvantaged. The center becomes more “centered” in the long run, but both the universality and relativity of world literature should be emphasized, since “without the former, literary works of any countries could be regarded as world literature, and without the latter, world literature will become all the more Eurocentric or Westcentric” (N. Wang, “‘Weltliteratur’” 298). AI is expected to act against such “centrism” through its incommensurable access to information. The actual situation is the opposite of this expectation, however. AI seems to strengthen instead of weakening the dominant position of central literatures as the answers which AI gave us when we asked about theories of world literature show. The first list of representative theorists of world literature ChatGPT gave us did not include any Chinese and very few Asian names (not even those who, although born in Asia, were educated and were working in the US, like Edward Said and Gayatri Spivak). When we narrowed down the question to Chinese scholars, the answer was still disappointing, with outdated and irrelevant infor- mation. ChatGPT mostly mentioned premodern belletristic writers, with the very few exceptions of Qian Zongshu (1910–1998), a novel- ist and scholar with significant contribution to comparative literature, and David Der-wei Wang (1954), who actually is a scholar of modern and contemporary Chinese literature. The only really relevant hit was Ning Wang (although with an incorrect birthdate), probably due to his extensive publications in English, while those whose achievement, in the Chinese context, appears equally or even more important in world literature studies, such as Weigui Fang or Daiyun Yue, did not make it onto the list. This demonstrates the dominance of English-related information in AI tools. The renewal of world literature theories at each stage, as history shows, helps re-think, re-define and re-compose the concept of world literature, transforming it into a dynamic and lively system. The inclu- sion into AI of theories elaborated by eastern scholars, such as those from Asia and more specifically, China, could entail the inclusion into world literature of literary works from various cultural backgrounds, thus de- centralizing the discourse in the long run, since such theories are more keenly aware of civilizational differences. Only with the contribution of literatures written in languages other than English, can world literature appear as a balanced system. Moreover, if one asks ChatGPT to name Peina Zhuang, Péter Hajdu: Rethinking World Literature in the Age of AI 195 the representative works of world literature for different periods, it pro- duces roughly the same answer, thus running counter to the dynamic and generative feature of world literature, since the basis it uses is the data input into itself until now, lacking in a diachronic investigation into world literature. World literature thus becomes something rather fixed either in its inclusion or the representative theories or scholars. ChatGPT can also lead to a gradual leveling of literary works through being insensitive (or too sensitive) to commercial aspects. With the use of AI such as ChatGPT, a variety of literary works come into the readers’ view in an endless stream, accelerating the breakdown of the boundary between elite and popular literary works. This could be regarded as the triumph of postmodernism, which is typically char- acterized by “the supersession of everything outside of commercial cul- ture, its absorption of all forms of art high and low, along with image production itself” (Jameson 135). On the other hand, it could be inter- preted as decomposing the classic. Postmodernism might have aban- doned the concept of the classic, i.e., works that have eternal potentials of reinterpretation or canonical works with high aesthetic and educa- tional value. Postmodernism’s blurring of elite and popular culture, and its challenge to the canons, do not mean embracing commercial- ism without any reservation. To use a crucial concept of Hans-Georg Gadamer (Gadamer 31–37), ChatGPT does not have taste, and since aesthetic taste judgements, despite being intersubjective and impecca- ble, are incomprehensible in discursive ways, it is questionable whether it can ever be trained to develop one. One might expect AI to make the “footwork” needed for what Franco Moretti called distant reading. When he described how he can read peripheral literatures distantly without knowing their language, he was actually suggesting the reading of literary histories and compendia written by native scholars in the languages of the central literary cul- tures (Moretti 59–60). In this regard, ChatGPT might have the advan- tage of being able to scan the academic discourse of peripheral litera- tures in their native language as well. However, the commercial bias of many search engines and the overwhelming presence of mass culture in search hits easily lower the threshold for the literary. ChatGPT tends to promote the “world literature” circulation of lowbrow mass products. Although “the whole world may be searching for a young voice that not only conforms to the most typical characteristics of the internet in today’s era, but also generates a sense of continuity” (He 2), not only AI-produced literature but also AI selection of representative literary works by human writers tends to become standardized and simplified. PKn, letnik 47, št. 3, Ljubljana, november 2024 196 If one asks ChatGPT in English about a literary culture of a differ- ent language, it will predominantly search the Internet in English for answers, which makes it much less useful in creating a new, less biased consciousness of world literature. Although one might hope that AI as a translation tool can also help one access various literary cultures, the potential of AI transla- tion is also overestimated. Translation stands as the unavoidable issue in any discourse on world literature. Goethe’s much quoted formula- tion about the coming of age of world literature was triggered by his reading of a Chinese novel in translation. Although David Damrosch, who defined world literature as circulation, emphasized that circula- tion beyond the original context can happen both in translation and in the original language, the statement that “world literature is writ- ing that gains in translation” (Damrosch 281), seems to discard the latter option as marginal. Although in the current world literature, which shows more and more traces of globalization, reading litera- ture in English is becoming a habit of many people in many non- English-speaking cultures, translation still seems the more important medium of circulation beyond the original context, and thus “world literature also implies the meaning of ‘transnational’ or ‘translational’” (N. Wang, “‘Weltliteratur’” 297). English translation can also medi- ate between two minor literary cultures. Without an intermediary of translation, some literary works can only be in a state of “death” or “marginalization” in other cultures and literary traditions (N. Wang, “World Literature” 24). Literary translation appears to be more impor- tant in the age of AI, representing the courage and confidence of vari- ous literatures to go global, and also the broad-mindedness of world literature. Due to artificial intelligence, world literature will have more diverse kinds of translators, from experts and scholars proficient in foreign languages to underpaid professional translators working under time pressure and to machine algorithms. Translation by AI has continuously broken through technologi- cal bottlenecks and has made significant progress in terms of transla- tion accuracy thanks to systems such as GoogleTranslate and DeepL. Intelligent translation is not only a requirement of an average user, but is also important for cultural strategies of various countries, since “national translation technology capability is an objective require- ment for countries around the world in the era of artificial intelligence. Countries with a sense of responsibility should attach importance to national translation technology capability and promote global tech- nological development” (H. Wang 43). The development of LLMs Peina Zhuang, Péter Hajdu: Rethinking World Literature in the Age of AI 197 such as ChatGPT has greatly improved the convenience, automa- tion, and accuracy of translation technology, thus further promoting the formation of a technological community with a shared future for humanity. ChatGPT relies on its own knowledge base and corpus to translate between various natural languages, which has great potential in literary dissemination. Despite the progress AI made in translation, one cannot stay blind to its deficiencies, especially in the field of literary translation, which typically still relies on the work of professionals, experts and scholars, or what is called academic translation, making us wary of absolute translation technicism. J. Hillis Miller announced almost ten years ago three important challenges to the new World Literature, the challenge of translation being the first, and this is still so today. The imaginary world in which a poet lives is “structured differently, and the transla- tor’s job, already hard enough, is complicated by the duty to render this difference” (Saussy 293). We expect no less of a poetic translation than that it should create an autonomous literary world. Translation of world literature involves heterogeneous cultures and civilizations, which puts the translator’s task beyond the already challenging requirements of accurate understanding and rendition. Already Aristotle thought that poets’ excellence could be measured by their metaphors, but “[a]t present, machine translation and translation technology cannot effectively handle the translation of metaphorical words, and the degree of accurate inference of metaphorical meaning is very low” (Li 25). The problem of translating metaphors, however, cannot be limited to poetry, since they play a crucial role in the lan- guage of most literary artforms, as well as in philosophy or language in general (see Ricoeur). ChatGPT has progressed far beyond previous machine translations, but it still fails to fully meet the requirements of literary works, as can be demonstrated with the English translation of the highly complex literary and philosophical term “yi” (义), a central notion for Mozi. ChatGPT renders it as “righteousness” or “righteous- ness and loyalty,” offering a brief extra explanation of its meanings in Chinese classics. But “yi” signifies far more than just “righteousness” or “loyalty” for Mozi. It can also mean “benevolence/benefits-oriented,” “ethics of conduct” and so on, depending on the context. Obviously, ChatGPT here still tends to offer a literal translation of this culturally loaded term “yi,” regardless of its multiple implications, which consti- tutes a central feature in classical Chinese literature and philosophy, where the meaning of a word relies mostly on the context and one’s individual insight rather than the literal sense. PKn, letnik 47, št. 3, Ljubljana, november 2024 198 The deficiencies of ChatGPT’s translation became more obvi- ous after some experimentation which we discuss below. We asked ChatGPT to translate the classical Chinese essay “Wenfu” (文赋, “Exposition on the Genre Fu”) by Lu Ji (261–303) into English. Since the essay was a prestigious and sometimes central artform of classical Chinese literature, especially during Western Jin Dynasty (266–317) and thereafter, this choice is a legitimate challenge to literary transla- tion skills of AI. One of the first surprising results was that the quality of transla- tion by ChatGPT is unstable. The first two or three paragraphs were translated rather well, even better than that by DeepL, with, for example, such idiomatic diction as “brilliance of literati, subtleties and extol,” and a balanced and succinct syntactic structure, such as “sighing over their passing, gazing at myriad things and contemplating their diversity.” However, this standard does not last; later the trans- lation becomes more and more simplified, tending to frequently use “subject+verb+and” structure; for instance, the last paragraph of the essay is translated into a six “it+verb” sentence: “It spans, it observes, it rescues, it spreads, it absorbs, and it encompasses.” If we were to restart ChatGPT every time after it has finished two or three paragraphs, the result would be more satisfying, but this is time-consuming process, which still cannot handle the difficulty of translating culture-loaded terms and concepts, such as the above mentioned “yi.” Another shortcoming appears to be ChatGPT’s failure in demon- strating or presenting the aesthetic feature and flavor in literary transla- tion. Let us take its translation of Robert Frost’s poem “Spring Pools” for an example: Spring Pools These pools that, though in forests, still reflect The total sky almost without defect, And like the flowers beside them, chill and shiver, Will like the flowers beside them soon be gone, And yet not out by any brook or river, But up by roots to bring dark foliage on. The trees that have it in their pent-up buds To darken nature and be summer woods Let them think twice before they use their powers To blot out and drink up and sweep away These flowery waters and these watery flowers From snow that melted only yesterday. (Frost) Peina Zhuang, Péter Hajdu: Rethinking World Literature in the Age of AI 199 Version 1 by ChatGPT 这些池塘，即使在森林中，仍然倒映出 几乎没有瑕疵的整个天空， 像它们旁边的花朵一样，寒冷而颤抖， 就像旁边的花朵一样，它们很快就会消失， 但并非在任何小溪或河边， 而是通过根部将黑暗的叶子引上来。 那些在它们封闭的芽中拥有的树木 让大自然变暗并成为夏季的树林 – 在它们动用力量之前，让它们三思 不要抹去、喝掉和扫走 这些花朵的水和这些水中的花朵 就像昨天刚刚融化的雪一样。 Version 2 by ChatGPT 此池，纵在深林，犹能倒映天穹无瑕， 花儿相依，凛冽颤抖，终将如同傍花之物消逝。 却非溪水江河之畔，而是根脚悄然将幽暗羽翼铺展。 那些树木，藏匿着春日之力，欲使大自然黯然褪色成夏林， 愿它们三思，切勿滥用力量，将花海澄净之水， 以及水中的芬芳花朵一扫而空，如同昨日融化的白雪。 We get two versions of the translation, the second one more literary, but the flavor and significance conveyed are rather similar to the first, and the rhyming scheme of the source text is also lost. What mat- ters more is the losing of “depth” the poem wants to express. Robert Frost is good at using colloquial language, familiar rhythms, common metaphors and symbolic techniques in daily life in depicting the New England countryside. His record of the overlooked rural scenery often contains profound, symbolic, and even metaphysical meanings, going beyond the personal experience and linking together the inner world, nature, and the entire universe, while explaining the values on which his profound religious beliefs are based. A good translation should en- deavor to represent all this, though translators might render the various features differently. The use of “在它们动用力量之前” and “切勿滥 用力量” in the ChatGPT’s translation is rather awkward, compared with “请让它们一思再思” and “它们不妨多多体贴” by a human translator (Frost). The use of “请” and “不妨” conveys the poet’s la- menting on the quick passing away or disappearing of the beauty in PKn, letnik 47, št. 3, Ljubljana, november 2024 200 nature, so the trees in the poem should “think twice” before draining the pools that were formed by yesterday’s melting snow water, hop- ing to prolong their beauty as much as possible and also indicating a deep sense of humanism and humor towards life by the varied extent of such “appealing” on the trees in “请” (“please”), “让” (“let”) and “ 不妨” (“might as well”). This is a testimony to the diversity of human translation, compared with the roughly same translation by ChatGPT, though varied a little bit in diction after inserting a new prompt but still lacking the atmosphere of a poem. What is more, not only Damrosch believes that world literature is “writing that gains in translation” (Damrosch 281) but Walter Benjamin also assumed that a translated work can, in a foreign context, gain “continuing life” or “afterlife” (Benjamin 153). This is far from being the case with the translation by ChatGPT; it might even be the opposite. With AI we can check the process of a poem being trans- lated again and again rather quickly. Let us take the following lines of “Spring Pools” as an example: “Let them think twice before they use their powers / To blot out and drink up and sweep away.” If one asks ChatGPT to translate the Chinese version back into English, then from English into Chinese and again from Chinese into English, one can find the “final” English version: “Let them reconsider before they unleash their might / To blot out and consume and sweep away” and “Before they unleash their might, let them ponder / To blot out and consume and sweep away.” The replacement of “think twice” by “reconsider” and “ponder” plus the use of “consume” does not conform to the original lively and poetic atmosphere, creating a rather awkward style. The back-and-forth translation experiment makes the poem go far away from the origi- nal without much literariness left. The use of ChatGPT ironically runs counter to the expectation both Damrosch and Benjamin invest in the role of translation in literature. We “lose” things, mainly “literariness,” instead of gaining them, turning a literary work or part of the work into a flattened piece of composition. Undoubtedly, LLMs like ChatGPT are making day-by-day break- throughs, serving as a new platform for the system of world literature. AI allows machines to connect with world literature based on the estab- lished cognition of humans. The convenience offered by this technol- ogy cannot guarantee a problem-free approach to world literature, which makes us wonder whether LLMs are a blessing or a curse for world literature. It undeniably brings great opportunities one might be happy to seize, but also some challenges, both to world literature and Peina Zhuang, Péter Hajdu: Rethinking World Literature in the Age of AI 201 the humanities more generally. The formulation of patent law of works created by AI can be a complicated problem, as well as judging if a work is wholly created and owned by AI. Despite the positive role AI plays in world literature, one still needs to stay vigilant against the poten- tial problems it could cause in the creation, translation and circulation of world literature, for instance, retaining of “alienness” of the source text and the various “biases” integrated into this technology, as shown by different representations of the same writer, especially those writ- ers from developing countries, in different AIs developed in different regions. Only by being aware of this can we use the technology to our advantage in advancing the values cherished both in cultural relativ- ism and cultural universalism of world literature. AI will choose, sieve and represent based on the cognition system of its designers; therefore, it may further promote the popularity of literatures from dominant cultures, going contrary to the original intention of Goethe’s concep- tion of world literature. As we have tried to argue, we need a balance between technology and humanity, as well as center and periphery in order to use AI for a “better” world literature. WORKS CITED Benjamin, Walter. “The Translator’s Task.” Translated by Steven Rendall. TTR, vol. 10, no. 2, 1997, pp. 151–165. Brown, Marshall. “Encountering the World.” Neohelicon, vol. 38, no. 2, 2011, pp. 349–365. Calvino, Italo. Una pietra sopra: discorsi di letteratura e società. Turin, Einaudi, 1980. Cheah, Pheng. What Is a World? On Postcolonial Literature as World Literature. Duke University Press, 2016. Damrosch, David. What Is World Literature? Princeton University Press, 2003. Duhan, Alice, et al., editors. Literature and the Work of Universality. De Gruyter, 2024. Frost, Robert. “Spring Pools.” Poem Hunter, 3 Jan. 2003, https://www.poemhunter. com/poem/spring-pools/. Gadamer, Hans-Georg. Truth and Method. Translated by Joel Weinsheimer and Donald G. Marshall, Continuum, 2004. Hayot, Eric. On Literary Worlds. Oxford University Press, 2012. He, Jing. “ChatGPT时代，翻译如何创造新的语言和新的文学？” (“How to Create a New Language and Literature by Translation in the Age of ChatGPT?”). Wenxue Bao, 30 Mar. 2023, p. 2. Heidegger, Martin. Basic Writings: From Being and Time (1927) to The Task of Thinking (1964). Edited by David Farrell Krell, HarperCollins, 1993. Jameson, Fredric. The Cultural Turn: Selected Writings on the Postmodern, 1983–1998. Verso, 1998. Juvan, Marko. “Wordliness, Worlds, and Worlding of Literature.” Metacritic Journal for Comparative Studies and Theory, vol. 8, no. 1, 2022, pp. 5–22. PKn, letnik 47, št. 3, Ljubljana, november 2024 202 Lem, Stanisław. Cyberiada. Krakow, Wydawnictwo Literackie, 2002. Lem, Stanisław. The Cyberiad: Fables for the Cybernetic Age. Translated by Michael Kandel, Seabury Press, 1974. Lem, Stanisław. Kiberiáda. Translated by Beatrix Murányi, Budapest, Európa, 1971. Li, Yingjun. “机器翻译与翻译技术研究的现状与展望: 伯纳德·马克·沙特尔沃 思访谈录” (“The Present and Future of Machine Translation and Translation Technology: An Interview with Dr. Mark Shuttleworth”). Chinese Science and Technology Translators Journal, no. 1, 2014, pp. 24–27. Moretti, Franco. “Conjectures on World Literature.” New Left Review, no. 1, 2000, pp. 54–68. Ricoeur, Paul. The Rule of Metaphor: Multi-Disciplinary Studies of the Creation of Meaning in Language. Translated by Robert Czerny et al., Routledge and Kegan Paul, 1986. Saussy, Haun. “The Dimensionality of World Literature.” Neohelicon, vol. 38, no. 2, 2011, pp. 289–294. Schleiermacher, Friedrich. Hermeneutics and Criticism and Other Writings. Edited by Andrew Bowie, Cambridge University Press, 1998. Shklovsky, Viktor. “Art as Technique.” Russian Formalist Criticism: Four Essays, edited and translated by Lee T. Lemon and Marion J. Reis, University of Nebraska Press, 1965, pp. 3–24. Spivak, Gayatri Chakravorty. “Three Women’s Texts and a Critique of Imperialism.” Critical Inquiry, vol. 12, no. 1, 1985, pp. 243–261. St. Vincent Millay, Edna. “Spring.” Poetry Foundation, 13 Apr. 2017, https://www. poetryfoundation.org/poems/44728/spring-56d223f01f86e. Ursa, Mihaela. “Universality as Invariability in Comparative Literature: Towards an Integrative Theory of Cultural Contact.” Primerjalna književnost, vol. 37, no. 3, 2014, pp. 151–163. Wang, Huashu. “国家翻译技术能力研究：概念内涵、要素分析和主要特征” (“National Translation Technology Capacity: Concept, Elements and Main Features”). Chinese Translators Journal, no. 2, 2023, pp. 35–43. Wang, Ning. “‘Weltliteratur’: From a Utopian Imagination to Diversified Forms of World Literatures.” Neohelicon, vol. 38, no. 2, 2011, pp. 295–306. Wang, Ning. “世界文学与翻译” (“World Literature and Translation”). Literature and Art Studies, no. 3, 2009, pp. 23–31. Zhang, Sheng. “亚里士多德的梦魇与萨特的愤怒” (“Aristotle’s Nightmare and Sartre’s Fury: The Impact of ChatGPT on Literary Creation”). Studies in Culture and Art, no. 3, 2023, pp. 44–48. Zhang, Yingying. “人工智能时代, 技术和人文应相互照亮” (“The Mutual Illumination of Technology and Humanities in the Era of Artificial Intelligence”). Wenxue Bao, 18 June 2023, p. 1. Zong, Baihua. 宗白华全集 (第2卷) (A Complete Anthology of Zong Baihua, Vol. 2). Hefei, Anhui Education Press, 2008. Peina Zhuang, Péter Hajdu: Rethinking World Literature in the Age of AI 203 Premišljevanje svetovne književnosti v dobi umetne inteligence Ključne besede: literatura in umetna inteligenca / poezija / literarno prevajanje / literarnost / ChatGPT / Lem, Stanisław Odkar se je pred skoraj dvema stoletjema v pogovoru med Goethejem in nje- govim tajnikom prvič pojavil izraz »svetovna književnost«, se razprava o njej ni nikoli ustavila, ampak se je v dobi umetne inteligence, ki prinaša spremembe na skoraj vseh področjih družbe, tudi v svetovni književnosti, le še okrepila. Natančneje, čeprav umetna inteligenca prinaša izjemne prednosti za obtok in prevajanje svetovne književnosti, odpira tudi mnoge izzive in težave. Na podlagi analize prevodov literarnih besedil med različnimi kulturami članek trdi dvoje: prvič, UI lahko nasprotuje trditvi Davida Damroscha, da »svetovna književnost pridobiva s prevajanjem«, saj ta lahko, kot kažejo izbrani primeri v prispevku, zaradi UI tudi »izgublja«, na primer svojo literarnost; drugič, UI lahko ustvarja »pristransko« sliko o teoretskih ali literarnih delih iz različnih kultur. Prav zato moramo biti pozorni, da nas ti dve nevarnosti »tehnološkega vala« ne bi zavedli. 1.01 Izvirni znanstveni članek / Original scientific article UDK 82.0-1:004.8 DOI: https://doi.org/10.3986/pkn.v47.i3.10