AI advances online literature, but has limitations


A snapshot of ChatGPT generated poem Photo: Yang Xue/CSST

When ChatGPT entered the world of literary creation, discussions about AI’s use in online literature started to heat up. Technological advancements push AI-generated creation from possibility to inevitability, but whether it can earn respect for its creative achievements remains to be seen.

Technological empowerment

The horizons AI opens for online literary creation are endless. Whether discussing preexisting apps such as “Xiaoice,” “Jiuge,” “Vivi,” and other poetry-writing programs or the currently trending ChatGPT, online literary creation has traditionally had a closely intertwined relationship with technology. Online literature has seemingly found a new creative tool in wave of cutting-edge AI-based technology. While AI wasn’t made for online literature, its functional extensions bring considerable “literary added value” to online creation.

First, it is now a common practice in AI-based creation to maximize and optimize the function of intelligent programs as tools for literary creation. For instance, when we use ChatGPT in fiction writing, we just need to enter the title, genre, theme ideas, storylines, main characters, as well as details, pace, style, chapters, word count, and other relevant factors, and it promptly generates the text for a novel. Additionally, one can utilize features like creativity generators and story plot generators for inspiration and fresh ideas, saving time and energy. Current popular AI writing software like ChatGPT, Claude2, and Yuewen Miaobi all possess robust writing assistance functions.

Although intelligent writing software is a great help in the creation process, achieving successful outputs are almost impossible upon the first attempt. The quality can be inconsistent and often requires several rounds of optimizing prompts to help the AI fully comprehend the writer’s intent. Subsequent time must be spent polishing manually, making modifications and improvements to achieve the best results. In most cases, AI serves as an “intelligent language assistant” and cannot work independently, so that in the end the quality of human-machine interactive writing depends on the literary competence and creative abilities of the program operator.

Web novels are often lengthy, sometimes reaching millions of words. Writers can establish the overall framework and worldview themselves and allow AI to assist with detailed descriptions. This approach simplifies the writing process significantly, as writers only need to assess and adjust the logical flow, transitioning authors from a purely creative role to a hybrid creative-editorial role.

Current AI writing assistants all fall under the category of Artificial Intelligence Generated Content (AIGC), whose creative effectiveness relies on a fusion of extensive literary data pre-training models, multimodal technologies, and generation algorithms, enabling these applications to swiftly produce text. However, to make the works vivid, interesting, and of reliable quality, it’s essential to add a high-level layer of human-machine interactions, resulting in what is known as the “Pygmalion Effect” — you get what you expect. If you provide ChatGPT with appropriate themes or keywords, it generates text that meets your requirements. The quality of questions determines the quality of the text produced, creating a direct correlation between the creator and the machine.

Second, facilitated by novel artificial intelligence technologies, new multimedia narration forms have emerged in online literature. These advancements may potentially alter the structural forms of literary mediums, shifting towards a form of multimedia storytelling that combines images, audio, and text, no longer confined to pure textual narratives. Multimedia and hypertext creation that are powered by digital technologies are crucial to set online literature apart from traditional literature. In Western countries, online literature includes experimental works in multimedia formats and hypertext linking. When online literature first appeared in China, there had been some influential multimedia and hypertext works such as Ha Ha, University!, Shaking Life, and Christmas Eve Subway.

Due to the high technological requirements of such hypertext creations and challenges monetizing them, there was a shift towards unadorned text and lengthy stylized novels, forming a unique “Chinese model” for online literature. The advancement of AI, particularly the deep involvement of powerful technologies like ChatGPT-4 and Midjourney, which offer robust graphic, mapping, audio, and video generation capabilities, has made multimedia storytelling and text-image combinations effortless. Now, multimedia storytelling is becoming dominant in the online literature market.

The incredible market performance of short video platforms such as Douyin and Kuaishou and the adaptation of online literary IP into multiple media formats demonstrates the advantages and income generating prospects of “text, graphics, audio, video, and gaming” offerings in the burgeoning “literature-art-entertainment-industry.” Tools like AI-powered Runway Gen-2 synthesizes videos and enables text-to-video, text+image to video, and image-to-video conversions. Runway’s editing tools facilitate image translation, segmentation, restoration, transformation, speech synthesis, speech recognition, and video synthesis, among other functionalities, which generate and edit visual content.

As these technologies progress and become more widely used, multimedia works may become the mainstream form of online literature. Online novelists will then adopt a dual identity, as writers and audiovisual content creators, while literature websites transform into comprehensive platforms for “literature originals+audiovisual and gaming operations.” In May of this year, Yuewen Group established a new division, aiming to enhance IP incubation and ecosystem efficiency, promote a “multi-modal content—platform” operational chain, and explore AI applications across the entire industry chain, from assisting content creation to IP incubation and multi-modal development and experiences. This upgrade to AIGC-enabled original multi-modal content platforms is designed to adapt to the changes brought by AI-enabled literature.

Third, the ecosystem for online literature is expected to transform. Chinese online literature has undergone various phases of development, from the “hipster style” in the 1990s PC era, the slump following the dot-com bubble burst around the turn of the millennium, the commercial explosion after the establishment of the “Qidian” model in 2003, to the 2010 shift towards mobile platforms and the rapid spread of social media catalyzed by the rollout of 4G and 5G technology from 2013 onwards. Technological advancements have shaped the landscape of Chinese online literature, progressively transforming its ecology. With the surge of AI technologies, such as ChatGPT, might AI spark a new evolutionary phase in online literary ecology? Early signs suggest the answer is yes.

Compared to PC, mobile platforms, VR, AR, MR, and even blockchain and metaverse technologies, the unique advantage of AI in online literature lies in its role not only as a medium or carrier, but also as a creative tool and content producer. With constant algorithmic innovations and as pre-trained models and multi-modal development mature, AIGC’s technological capabilities have undergone significant qualitative changes. AI is likely to gain universal and robust literary production and dissemination capabilities. It will not only change the way literature is created and exists but will also alter the entire landscape of online literature.

Three boundaries

Though AI has brought new possibilities to literary creation, AI is fundamentally different from human creation and has three clear boundaries restricting its creative process.

The first is the boundary of life experience. Currently, AI exists in a weak early stage, lacking self-awareness and life experience. However, literary creation is rooted in self-awareness painstakingly earned through life experiences. For instance, the hardships Kafka experienced by “being an honest man in a dishonest world” gave birth to his works The Metamorphosis, The Castle, and The Trial. Cao Xueqin, facing poverty and “having to borrow to provide daily meals for the whole family,” wrote The Dream of Red Mansions, where every word gives the impression it was soaked in blood. All these famous works testify to the importance of emotional experiences in literary creation. The data generated by AI erases human “physical experiences,” subverts the traditional concept of literary creation where the body is used as a pen and the soul as paper, it breaks the organic connection between literary work and the author’s social, historical, and life experiences, inevitably weakening the existential significance of literature.

For web novels, technological tools may assist novelists as they fulfilling readers’ expectations for popular elements like golden heroes, girl-next-door characters, out-of-the-box ideas, or mind-opening plot twists. However, solely relying on this technology will not spawn a generation of famous online writers, nor will it create a new type of online literature that has never appeared before. It can only rearrange stories within existing genres such as fantasy, martial arts, historical time-travel, or sci-fi romance. Lacking the life experiences essential for literary creation and devoid of the vivid perceptions of human struggle, AI cannot overcome this boundary in literary creation. This gap will inevitably result in a disconnect between technology and literature.

The second boundary is emotional resonance. Literature, including online literature, is an art form fueled by human creativity and emotional investment. While AI can mimic various literary styles and create different themes, it follows preset rules and patterns, lacking self-awareness and subjective feelings. Therefore, AI fails to express genuine emotions or personally resonate with what it has produced. If you ask AI to write a poem praising the motherland, the text which will naturally appear may meet the requirements in terms of content, but it would be a stretch to call it a “poem.” AI is challenged not only by the abundance of language without rhyme, but more so in its lack of genuine emotions. Thus the prompt fell flat and it lacked the essence of poetry, failing to evoke any emotional resonance.

In comparison, read Ai Qing’s striking lines from “I Love This Land.” It lacks extravagant language and vivid metaphor, yet it possesses an artistic power that moves hearts. Poetry and literature require genuine emotions. AI, like ChatGPT, operates by using artificial neural networks and natural language processing for language model training. It answers user queries or generates meaningful dialogue based on different application scenarios. It employs search algorithms to string words, sentences, and paragraphs together, selecting the best words to enhance the quality and fluency of dialogue. However, this kind of literary creation lacks emotion, the love or hate deeply rooted in the human world.

The third boundary is cognitive limitation. AI primarily possesses computational intelligence and, to some extent, perceptual and cognitive abilities. However, while AI’s cognitive capabilities might be robust and remarkable, its thought process is often singular, narrow, and highly specialized. AlphaGo surpasses all top human players in the Go game, yet it cannot compose poetry. Intelligent navigation systems can find precise locations in complex spaces but lack expertise in medical diagnosis. Expertise in specific domains is the forte of intelligent machines, but it’s also their limitation.

The cognitive appreciation of literature demands a holistic and systemic understanding. It requires recognition of society, nature, and human beings themselves. The values shaping literary production cannot be replaced by computational language training models. Literature relies on life experiences, judgments of right and wrong, ideals, emotions, and complex realms. Intuition and sensitivity, which drive the innovation of literature, are not strong suits for AI.

If one acknowledges that online literary creation is a form of value inscription, whether produced by writers or generated by AI, if a literary work lacks human values, those bits and pixels displayed on electronic screens become a meaningless arrangement of symbols. AI empirically connects symbols, learning experiential connections between words from its human language repository, simulating human neural networks to create probabilistic combinations. However, it doesn’t actually comprehend the meaning within these symbols, it lacks a real understanding of each word’s significance, and cannot establish causal relationships between things. Therefore, it fails to transcend the boundary of value cognition. AI has had a significant impact on society, but it will not replace literary creation in the foreseeable future. As one web novelist put it, “By the time it does so, most occupations in society might already have disappeared.”


Ouyang Youquan is a professor from the School of Humanities and the president of the Institute for Online Literature at Central South University.

Editor:Yu Hui

Copyright©2023 CSSN All Rights Reserved

Copyright©2023 CSSN All Rights Reserved