1. Introduction
[1.1] I can still remember the days of crying because I felt that I lacked the time and skills to turn inspirations into works of fan art. Thankfully, I was not alone; I had otaku peers who were crying for the same reason. We wondered, "What if a machine could read our minds and transfer our ideas directly into accomplished works?" With today's generative artificial intelligence (AI) technologies, this daydream has come true to some extent. With the right prompt, otakus can now produce texts with ChatGPT, images with Midjourney, and audio with SoftVC VITS, and then combine these materials into many types of fan art, including fiction, illustration, manga, music video, and so on. Works tagged as "AI-generated" have become more common on otaku art websites, such as Pixiv and Bilibili, demonstrating that at least some otakus are using AI for creating and consuming their beloved otaku culture.
[1.2] Nevertheless, this is not a wonderful "dream come true" story. When we otakus talked about a "fan art–making machine," few really considered how the machine would operate and with what consequences. Once daydreams become reality, many find that they are actually unhappy with the AI-equipped machines. To date, ethics, copyright, monetization, and authenticity are some of the main themes that have emerged in disputes over using generative AI for otaku creative practices (Lamerichs 2023). Reading dispersed opinions from otaku communities while recalling my daydreaming memories, I realize that it is equally important to debate the ways in which these themes are discussed when arguing relevant practical issues. As my experience implies, people sometimes lack the awareness to imagine and connect the comprehensive impacts of technologies with their own lives when picturing what they want the machine to serve.
[1.3] This essay explores the connections between humans and AI by adopting resources from studies of otaku cultural and media practices. It starts with a clarification of the term "otaku" and an introduction to Japanese cultural critic Hiroki Azuma's (2009) concepts of "database consumption" and "database animal." Azuma's theory is then applied through a discourse analysis of otaku users on Twitter (renamed X in 2023) talking about their experience with Waifu Labs, an anime portrait AI generator (https://waifulabs.com/). Highlighting their ways of using this AI generator and their tendency to humanize the AI, I argue that before the recent AI hype, otaku culture had already held certain creation and communication conventions that are similar to what has been established as the operating mechanism of generative AI. Both rely on the database to process information and build knowledge. As suggested by Azuma, this "database model" of otakus' consumption and production of cultural and media objects questions the distinctions between human creation and machine production or, to go further, the distinctions between human and machine. Such a distinction should be considered as constructed rather than natural and granted. This perspective offers insights to help us reflect on how we recognize the relationships and interactions between humans and AI.
2. Otaku, database, and animal
[2.1] In Japanese and as a loanword in other languages, "otaku" is often regarded as a term identifying enthusiastic Japanese anime and manga fans. However, one could find it highly polymorphous after investigating the actual usage of this term along with its social, cultural, and historical backgrounds. The term has no coherent meanings, and the individuals and communities under its umbrella hold their inner diversities and conflicts, which lead to a variety of approaches to studying otaku (Galbraith et al. 2015). Hence, "otaku" is not straightforwardly interchangeable with the word "fan" (Larsen 2018), and defining otaku requires anchoring it in concrete discourses. Here, I define otaku as people who are highly obsessed with and extraordinarily knowledgeable about a particular object (Azuma 2009; Galbraith 2019). Broadly speaking, this "particular object" can cover a wide range of themes, genres, and hobbies extending beyond anime and manga. This definition is meant not to be confined to certain objects, underlining the idea that people can be considered otakus because of how they relate to different objects through certain manners of obsession and knowledgeability. As Grassmuck (2000, ¶ 6) states, "Otakudom has nothing to do with any specific subject; it is about the way people relate to a subject." On the one hand, otakus' way of relating can overlap with the practices of media fans (Ito 2012), a point where otaku studies intersect with fan and audience studies. On the other hand, when associating with their otaku objects, otakus' distinctive methods for information management and perceptions of what constitutes humans and nonhumans make otaku a distinct "mode of being" (Grassmuck 1991, ¶ 4).
[2.2] This definition is compatible with Azuma's theory as the idea of database model focuses on how otakus handle relations with their otaku objects and the surrounding world. In his book Otaku: Japan's Database Animals, he claims that otakus are "database animals" conducting "database consumption" in postmodern societies. Azuma believes that societies have transformed from the modern to the postmodern era with a fundamental change in social institutions: the decline of the grand narrative. In line with French philosopher Jean-François Lyotard's view, Azuma sees the grand narrative as the singular social system or standard unifying social members, a system that manifested "intellectually as the ideas of humanity and reason, politically as the nation-state and revolutionary ideologies, and economically as the primacy of production" (2009, 28). However, its power to consolidate society has weakened, and the vast standard has been replaced by the coexistence of countless smaller standards. Under this circumstance, Azuma rates otakus as human beings well-adapted to the postmodern transformation by developing a "database" model in their consumption and (re)creation of otaku cultural and media products. This database comprises numerous coded information (i.e., "elements" or "data") derived from existing cultural and media texts. Otakus constantly implement pieces of data from the database as a source of knowledge to interpret not only cultural and media objects but also the society in which they live. Meanwhile, their interpretations and (re)creations of relevant products contribute new data to update the database. Thus, different smaller alternative narratives are created, read, and evaluated for otakus to organize their lives so that they are not disturbed by the loss of the grand narrative.
[2.3] For instance, otakus who are passionate about anime and manga do not necessarily prefer works with complete stories, unified worldviews, and well-rounded characters; rather, they need otaku elements to be presented in diverse but loosely associated multimedia commodities, such as figures, novels, drama CDs, video games, and anime products. Similarly, their (re)creations of otaku products are also the combinations of various elements, instead of the construction of a full narrative in accordance with the grand one. Examples of these elements include a character's appearance, dressing style, personality, ways of speaking, occupation, race, gender identification, and many more, all of which are recorded as data in the database. People keep seeking, identifying, interpreting, and reshaping data across a great variety of products to pursue ideal combinations of otaku elements. Correspondingly, the relevant industries promote more decentralized multimedia projects rather than single coherent works to satisfy the different needs of elements.
[2.4] This is the basic mechanism of otakus' database consumption. In this manner, otakus downplay the integrity of narratives, consistency of messages, and originality of artworks, favoring more fragmented small pieces of information that allow them more space to assemble, consume, and (re)create the information at their will. Due to this strong demand for information and the mechanical-like manner of information processing, Azuma calls otakus "database animal" (2009, 95), arguing that such needs and practices break down the long-standing beliefs about humans and humanity. Hence, he names this otaku mode of being "animal," a position in between what is usually considered human and nonhuman.
3. Waifu, machine, and human
[3.1] Sizigi Studios publicly released Waifu Labs at Anime Expo 2019 (note 1). According to Dictionary.com, "waifu" is a kind of slang used by otakus to refer to their particularly loved fictional characters—mostly but not necessarily female characters. Waifu Labs introduces itself as a site equipped with AI to draw ideal waifus—that is, to generate anime-style portraits. It uses Danbooru, a large-scale anime image database (note 2) and the generative adversarial networks to train the AI system (note 3). The site presents users with a 4 × 5 grid of random AI-generated images, any of which can be chosen as an initial template to begin image customization. At every step, it allows users to select different color palettes, painting styles, and design details (e.g., hairstyle, facial features, clothing, and pose) by displaying multiple AI-generated images as options. After several rounds of selection, a final waifu portrait can be downloaded.
[3.2] According to Sizigi Studios' statistics, two years after Waifu Labs' launch, the site had completed 20 million image-generating commissions (Liu 2022), indicating that many otakus have found this site attractive. This is also why I conducted a discourse analysis on Twitter about how users talked about their experiences with Waifu Labs back in 2021. I want to learn how we otakus desire AI and AI creations and how the relations among AI, AI creations, and human beings are perceived in this situation. Tracing the notion of discourse and discourse analysis back to Foucault, I subscribe to the view that discourse is the "particular knowledge about the world which shapes how the world is understood and how things are done in it" (Rose 2016, 187). My analysis seeks to identify what knowledge guides otakus to interact with Waifu Labs and what knowledge they construct when doing so. This approach echoes Azuma's database model, as otakus are believed to hold certain knowledge while also challenging other sets of knowledge, while being database animals. Although the broader discourses surrounding generative AI have evolved since 2021, I am confident that my findings can still reveal dynamics between otakus and AI that are worth consideration.
[3.3] I chose Twitter as the platform to collect materials for the discourse analysis because Sizigi Studios maintained an official Twitter account and actively responded to users' feedback during the research period. Meanwhile, as Twitter served as an open platform used by otakus from many different locations, I was able to access a more diverse group of otaku users. I used "waifulabs" as a search keyword, as I noticed that relatively few users used hashtags to post relevant content. The search timeframe was from July 2019 (the launch time of Waifu Labs) to November 2021. As the majority of the results were pure image sharing or archiving posts containing limited and homogeneous messages, I categorized them as one type of material for examination. Furthermore, 185 posts were selected as more informative samples, including posts written in English, Japanese, and simplified/traditional Chinese. Based on user profiles and their past tweets, the collected materials were basically posted by individuals who identified themselves as otakus or at least someone familiar with otaku culture. These accounts seldom divulge real personal information. Nevertheless, due to ethical concerns, I avoided direct quotations of the complete posts and excluded user IDs, post links, and any other information that could potentially identify the precise posts and users' accounts.
[3.4] Overall, the collected posts reacted positively toward Waifu Labs. Expectations of more diverse image results were sent to the development team as constructive advice. Only one or two posts mentioned ethical or copyright concerns. I did not intend to bias my selection toward positive tweets, but the limitation of the sampling made it clear that people who were more supportive of AI were much more likely to learn, use, share, and talk about specific AI tools. Consequently, in search of "waifulabs," compared with those who criticized and rejected generative AI, those who were open to AI and their posts about AI engagement were easier to search for and collect. This sample group provides useful evidence for analyzing how and why AI is needed by certain otakus and how they consider their relationships with AI.
[3.5] I want to discuss two analysis findings. The first is related to users' expressions of purpose for using Waifu Labs: to obtain images for entertainment and to gain inspiration for further art creation. For the former, users posted waifu images on Twitter, wrote brief character settings for their waifus, expressed their affection for these waifus, and praised each other's waifus (indicating the approval of the image quality and AI's ability). These posts convey the idea that the simple process of choosing among options and downloading images from the Waifu Labs itself is already a meaningful and joyful practice. Some performed these actions as a type of otaku routine, while others described themselves as addicted to the process. For example, one user kept uploading one waifu image with comments every day for a whole year. Another user said that they were "hooked" and "lost track of time," downloading over 6,000 generated images within two weeks. They even invited others by saying, "Why not step into this swamp of waifu-generating?"
[3.6] Next, for the other purpose, many users employed AI's endless stream of images as points of departure for their own creative activities. Some users who were originally artists commented that Waifu Labs was helpful in providing design references. Taking the generated images as prototypes, users recreated them into different artworks, such as graphics, 3D anime models, and short animations. They shared these works on Twitter, integrating the usage of Waifu Labs into their otaku creative activities. Sizigi Studios encouraged this kind of recreation, often retweeting relevant works. The team even organized a Waifu Labs art contest, inviting users to redesign and redraw the AI-generated images (note 4).
[3.7] Notably, these exercises with Waifu Labs fit into the pattern of database consumption. Azuma credited the development of information and internet technologies since the late twentieth century as a key factor for the formation of the database model. Now, after the advance and proliferation of AI technologies, otakus are interacting with the database in both literal and structural senses. The production of Waifu Labs portraits represents the accumulation and assembling of otaku elements; the users' acts of generating and appreciating waifu images are the hunt for their preferred elements; and the recreation of waifu images can be regarded as the production and consumption of derivative works, a significant component of the postmodern database model. Derivative works include amateur, noncommercial, and fandom art creations, the core of which is the ongoing rereading and reproduction of existing works. It is "a chain of infinite imitations and piracy" (Azuma 2009, 27), given that the chosen "original work" for the fabrication of a derivative work is oftentimes also a callback to and imitation of earlier works through the reassembly of different existing elements from the database. Therefore, Waifu Labs images are derivative works at every level: AI generates images based on existing human artworks, then human creators take these images to craft further works, which might someday be scraped and included in a training dataset. These all lead to my argument that, as a structure and system to organize, consume, produce, and circulate information, a "database" is not only installed on AI machines but is also embedded in the realms of human cultural and media activities. Issues of AI in otaku cultural and fandom practices are not the "latest phenomenon"; rather, they connect with preexisting practices and tensions. AI provides what the otakus need—a larger volume of information delivered more efficiently—and enhances the effect of what otakus always do. Otakus' database consumption and production erase the lines between the original and the copy, the creator and the audience, blurring the boundary between human creation and machine generation, as the two sides share similar database mechanisms to achieve production. The rise of AI further exposed this already blurred boundary.
[3.8] The other major analysis finding is the personification of AI. Developers and users characterized the AI behind Waifu Labs as a subject sharing common tastes and sentiments with them, as if AI were one of their own. For instance, Sizigi Studios has always addressed Waifu Labs as an artist who works hard to "draw" (rather than "produce" or "generate"), has its own mind, and "doesn't like being told what to do." Other users accept this frame in their conversations with Sizigi Studios on Twitter, depicting the AI as a member of the otaku community who "loves waifus," is a "big fan" of certain otaku elements and fictional characters, and is a creation collaborator. Such a humanized depiction of AI can be examined with the concepts of anthropomorphism and animism. Animism involves "personifying nature" (Richardson 2016, 110) and anthropomorphism imbues "the real or imagined behavior of nonhuman agents with humanlike characteristics, motivations, intentions, or emotions" (Leo-Liu and Wu-Ouyang 2022, 2). Japanese Buddhism and Shintoism and the traditional East Asian philosophies all contain anthropomorphic and animist traditions, which believe that humans are not the sole possessors of souls and spirits. These beliefs have had sustained impacts on East Asian sociocultural development (Kitano 2007), and otaku culture—with its Japanese roots—is no exception. Otaku culture often attaches humanlike qualities to nonhuman otaku objects, such as fictional characters and fandom merchandise. In this case, it is the generative AI endowed with human characteristics and shaping a discourse where it is natural and enjoyable to have AI as one's equal otaku companion. In this sense, AI machines are not the other (machine), but the us (human).
[3.9] I also regard the tendency to personify AI as an indication of the blurred boundary between humans and machines. What human beings do can be mechanical while what machines possess can be personalized. As mentioned above, Azuma's elaboration of "database animal" addresses this blurring and the issues of knowledge construction behind it. While it may be reasonable to assume negative connotations of the word "animal," to consider the "animalization" of people as a dangerous sign, or to worry about whether such language choice reinforces or even worsens the already unfavorable social impression of otaku culture and communities (Brienza 2012), my interpretation is that Azuma does not see animality as a degradation of humanity. If there is uneasiness caused by the word, it could be seen as Azuma's success in challenging the common sense of humans and humanity. For Azuma, knowledge about humans is largely determined by traditional European philosophies, which distinguish humans from nonhumans based on their ability to understand and use symbols. This knowledge construction, however, is not capable of dealing with struggles over human–machine interactions and the ambiguity of the alleged essence of the human as compared with the other. In a 2021 interview, Azuma affirmed that the progress of AI technologies has further threatened the validity of traditional Western definitions of humanity, as AI tools are being trained to handle symbols in order to communicate with and work for human beings. The growing concern about AI is one more proof of the inability of the existing knowledge about humanity to explain the eliminated distinctions between human and nonhuman existences (Azuma and Hui 2021). To solve the practical problems that have emerged with the increasing applications of AI in various social fields essentially requires reflecting on and refreshing the knowledge used to manage these problems. This entails asking who we are, what AI is, what we want and could obtain from AI, and what kind of relationship we ultimately expect to build with AI in our society.
[3.10] Hence, Azuma turns to "animal," which is not a denial of humanity, nor an insult to otakus. Instead, it aims to remind people how otakus' database consumption and production navigate in between the grounds of humans and machines, exhibiting features not necessarily consistent with so-called common sense or traditional definition of humanity. In other words, otakus' practices of communication and creation and the knowledge guiding these practices can go beyond the human–machine dichotomy, cope with the ambiguity of humanity, and (re)negotiate these already blurred boundaries. For Galbraith, otaku culture, located in an "in-between place," can enlarge our imagination of different options for living in this world (2019, 8). Accordingly, with otakus' database consumption of AI technologies, there is the potential to further imagine different visions of how we can get along with AI.
4. Conclusion
[4.1] Undoubtedly, Azuma's theory is not flawless. The first debatable point may be his starting premise, which takes for granted that all societies have experienced the postmodern transformation. One might also ask whether he is overly optimistic about the social impacts of the database model and the otaku culture's alternative potential. Nevertheless, these do not cancel the values of Azuma's works. By placing otakus in broader social contexts for scrutiny, he shows that otaku culture is never an isolated territory; on the contrary, it is a vital site to grasp notable moments of social transformation. We are currently at the crossroads of another transformation as a consequence of the AI technologies we invented. In this regard, reconstructing the knowledge about our machines and ourselves with an open and imaginative mind is worthwhile to drive the technology in a direction that genuinely serves human welfare.
5. Acknowledgments
[5.1] An early version of this essay was presented at the Fan Studies Network North America (FSNNA) 2023 Conference with the title Reconsidering Otaku and AI: Human, Machine, and Database. I would like to thank Dr. Suzanne Black and Dr. Naomi Jacobs for suggesting that I submit the essay to this issue.