AI and the Unimaginable Couples: Chinese Men and White Women

My first realization of the potential of GenAI in relation to academic research took place during one of our university Friday Seminars – academic presentations on different research topics followed by a Q&A session. The speaker shared with us his research on cancer patients in India. Due to privacy concerns, he had come up with an ingenious system to convey the setting of his fieldwork without showing us actual photographs: images generated by Artificial Intelligence (AI). One recent afternoon, as I was thinking about my own research project, those AI images suddenly popped up into my mind. 

For the last few years, I have been conducting research about transnational and interracial romantic relationships in Hong Kong. Specifically, I focus on the lives, stories, and experiences of couples composed of a Chinese man and white woman.  The fact that these couples are relatively “unusual” or “uncommon” was one of the factors pushing me towards this direction when I had to change my original doctoral project due to the Covid-19 pandemic. As I read through the many publications on transnational relationships and “interracial” intimacies, I soon realized that not much had been written on such pairings. Conversely, scholars have produced all kinds of interesting books and articles on Asian women and white (or Western) men. 3 E.g., Karen Kelsky, Women on the Verge: Japanese Women, Western Dreams (Durham and London: Duke University Press, 2001); Gary P. Leupp, Interracial Intimacy in Japan: Western Men and Japanese Woman 1543-1900 (London and New York: Continuum, 2003); Nicole Constable, Romance on a Global Stage: Pen Pals, Virtual Ethnography, and “Mail-Order” Marriages (Berkeley: University of California Press, 2003); Sealing Cheng, On the Move for Love: Migrant Entertainers and the U.S. Military in South Korea (Philadelphia: University of Pennsylvania Press, 2010); Heidi Hoefinger, Sex, Money & Love in Cambodia: Professional Girlfriends and Transactional Relationships (New York: Routledge, 2013); Patcharin Lapanun, Love, Money and Obligation (Singapore: NUS Press, 2019); Rebecca Forgash, Intimacy across the Fencelines: Sex, Marriage, and the U.S. Military in Okinawa (Ithaca and London: Cornell University Press, 2020); Monica Liu, Seeking Western Men: Email-Order Brides under China’s Global Rise (Stanford, California: Stanford University Press, 2023). Such scholarship, focusing on relationships which follow mainstream desires and patterns of attraction, has been extremely useful to understand how intimate relationships reproduce structural inequalities, but can also be potentially transformative. In the case of relationships between Chinese men and white women, they certainly constitute a “minority” in demographic terms, especially in Hong Kong, where since the British colonial period most interracial relationships were composed of local women and white men, through prostitution or the so-called “protected women” arrangement. 4 Philip Howell, “Race, Space and the Regulation of Prostitution in Colonial Hong Kong,” Urban History 31, no. 2 (2004): 229–48; Tony Sweeting, “Hong Kong Eurasians,” Journal of the Royal Asiatic Society Hong Kong Branch 55, no. 2015 (2015): 83–113. Even on a more global scale, such relationships are still relatively “uncommon” due to the construction of white women as inaccessible for “colored men” 5 Ruth Frankenberg, White Women, Race Matters: The Social Construction of Whiteness (London: Routledge, 1993); Ann Laura Stoler, Carnal Knowledge and Imperial Power: Race and the Intimate in Colonial Rule (Berkeley, Los Angeles, London: University of California Press, 2010). and the gradual transformation of Asian men into feminized or asexual beings, through restrictive immigration and anti-miscegenation laws, negative propaganda, and stereotypical media representations. 6 See e.g., Peggy Pascoe, “Race, Gender, and Intercultural Relations: The Case of Interracial Marriage,” Frontiers: A Journal of Women Studies 12, no. 1 (1991): 5–18; Elise Chenier, “Sex, Intimacy, and Desire among Men of Chinese Heritage and Women of Non-Asian Heritage in Toronto, 1910-1950,” Urban History Review 42, no. 2 (2015): 29–43; Kate Bagnall, “Golden Shadows on A White Land: An Exploration of the Lives of White Women Who Partnered Chinese Men and Their Children in Southern Australia, 1855—1915” (Ph.D. diss., University of Sydney, 2006); Emma Jinhua Teng, Eurasian: Mixed Identities in the United States, China, and Hong Kong, 1842-1943 (Berkeley: University of California Press, 2013); Jachinson Chan, Chinese American Masculinities: From Fu Manchu to Bruce Lee, Asian Americans. (New York and London: Routledge, 2001); Travis S.K. Kong, “Chinese Male Bodies: A Transnational Study of Masculinity and Sexuality,” in Routledge Handbook of Body Studies (London and New York: Routledge, 2012), 289–306. Nevertheless, it is precisely because of these factors and their “unconventionality” that these relationships also warrant the attention of scholars. Moreover, the topic I am working on is even more timely in light of the changing geopolitical, socioeconomic and cultural context of a “rising Asia” and “declining West,” with scholars demonstrating the declining currency of whiteness in East Asia, the emergence of an attractive form of masculinity embodied by Asian business elites, and the growing attractiveness of cultural phenomena like K-culture for Western audiences. 7 Kam Louie, “Popular Culture and Masculinity Ideals in East Asia, with Special Reference to China” 71, no. 4 (2012): 929–43; Kimberly Kay Hoang, Dealing in Desire: Asian Ascendancy, Western Decline, and the Hidden Currencies of Global Sex Work (Oakland: University of California Press, 2015); Liu, Seeking Western Men: Email-Order Brides under China’s Global Rise; Dredge Byung’chu Kang, “Eastern Orientations: Thai Middle-Class Gay Desire for ‘White Asians,’” Culture, Theory and Critique 58, no. 2 (2017): 182–208; Shanshan Lan, Willy Sier, and Aldina Camenisch, “Precarious Whiteness in Pandemic Times in China,” Asian Anthropology 21, no. 3 (2022): 161–70; Min Joo Lee, “Desiring Asian Masculinities through Hallyu Tourism,” in The Rise of K-Dramas: Essays on Korean Television and Its Global Consumption, ed. JaeYoon Park and Ann-Gee Lee (Jefferson, North Carolina: McFarland, 2019). Thus, my research addresses the interrelated questions of how global and local socioeconomic changes transform intimate desires, and of how intersectionalities of race, gender, and class are reconfigured in these romantic relationships and their interactions with family, friends, and society at large. 

Due to the meagre size of my friendship circle in Hong Kong and the scattered nature of the group I wanted to research, I decided to search for potential participants by approaching couples on the street, learning many insights about public visibility, racialization, and “proportions.” What I mean with the last term is that it was relatively less common to see a Chinese man-white woman couple compared to two Chinese individuals or other white-Chinese interracial couples (same sex or heterosexual). Still, I interacted with over 60 couples (interviewing about 35 of them) and saw at least as many from a distance, especially after the fieldwork phase of my study was over. This configuration resonated with many of my interlocutors, who explained how in the past (and to some extent still in the present) it was “rare” or “uncommon” to see couples like them, but it was gradually (and hopefully) becoming more common. For some, this exceptionality was almost a source of pride, proof of their cosmopolitanism and open-mindedness, evidence of more equal relationships (compared to what they saw as colonial legacies), and setting an example for others, representing a hope for change.  

Once the fieldwork part of my project was over, I began analyzing my data and making sense of it through writing. One of the dilemmas I faced was that I had taken photos during fieldwork with and of couples, but using those photos would mean that identities could not remain confidential. At the same time, because the couples I was researching were so invisible in media representation – it is still uncommon to see a Chinese/Asian man paired up with a white woman in films and TV series – and drew so many surprised stares in public, I felt that it was important to include some visual images of these “unusual” couples. That’s how I remembered those AI generated images. Perhaps GenAI could be the solution to my predicament.

So, I began playing around with an app called Poe on my phone, which has different AI bots to choose from, some of which are free. I downloaded it because ChatGPT was not directly accessible in Hong Kong. I found a couple of image generators and typed in my first prompt. The first bot was called StableDiffusionXL. My prompt was: “A romantic couple with a Chinese man and a white woman” [Fig. 1]. The result was confusing to say the least. The image depicted what looked like a cartoon-ified wealthy Asian family: the husband on the left, wearing a suit with a visibly tanned face; a young girl in the middle, presumably their daughter; and the wife on the right side, with a paler skin color and her hair in a bun. Perhaps the bot had interpreted “whiteness” as the skin color of the wife, as she does indeed look fairly white. But why the child?

I tried again and again using different terms. I decided to change the term “white” into something seemingly more specific and “scientific,” even though it has been criticized as an even more problematic term: “Caucasian woman – no kids”; “Caucasian woman (white, blonde, Western).” Nothing. All of the images continued to depict only couples with two Asian individuals. The bots continued to struggle with my prompts, seeming incapable of generating images of the sort of interracial couples with whom I work.

I redoubled my efforts, trying something more specific: “A Chinese man kissing a Western white Caucasian woman” [Fig. 2]. Here, I tried lumping together all the terms that were used in the literature and by my interlocutors to describe “white women”: Western, white, and Caucasian. Yet, the woman in the resulting image was still very much an Asian woman, albeit with lighter skin color than the partner she was passionately kissing.

Then I had an idea. What if I tried the opposite? Perhaps this could give me an idea of whether the image generator worked or not. My prompt was: “Chinese woman with Western man” [Fig. 3]. Here again, the image depicted a visibly lighter-skinned Asian woman dressed in Chinese garb with a darker-skinned Asian man. The result was also not what I had expected. But then again “Western” could be an ambiguous term, interpreted as a man from Western China perhaps. Moreover, this time I had not inputted the adjective “white” for the woman, but she was still very pale.

Regardless, I decided to give it another try: “Chinese woman with white Caucasian man” [Fig. 4]. This time the image generator got it perfectly right. The man was tall, blonde, and pale-looking, the stereotypical white man that most people in China imagine when they think of foreigners. The woman next to him was a more petite Asian woman, and a traditional Chinese house stands in the background.

That worked within just two trials. Maybe I had found the formula. I decided to use the same exact prompt in reverse. Maybe the secret for success was to combine white and Caucasian: “Chinese man with white Caucasian woman” [Fig. 5]. Still nothing. Even after five or six more attempts, I continued to receive images of two Asian partners.

As I grew increasingly frustrated, I hypothesized that maybe something was just wrong with this particular AI model. I decided to try another image generator offered in the Poe app: Playground-v2.5. This time, based on my only “successful” attempt with the first bot, I began with the opposite combination: “Romantic interracial couple: white man, Chinese woman” [Fig. 6]. The prompt yielded an image that was immediately right, depicting a DiCaprio lookalike beside a Chinese woman wearing an upper garment with a “mandarin collar” and with her hair tied up in an elegant bun.

Once again, I inverted the prompt to seek the couple I was interested in: “Romantic interracial couple: white woman, Chinese man” [Fig. 7]. In this image, the woman was simply a variation of the previous one, while the man was indeed Chinese-looking. Clearly, something was wrong here, as two exactly symmetrical opposite combinations produced strikingly different results. This could not possibly be a coincidence, as it had been the case for both of the GenAI bots I had used. Moreover, the bug reflected the problems of media representation, where it is quite frequent to see a white protagonist engaged in some sort of intimate relationship with an Asian woman, whereas the opposite is far less common. Similarly, when couples such as the ones the bot could not seem to conceive are seen in public, their intimacy is not always acknowledged as a romantic involvement. 8 Constable, Romance on a Global Stage: Pen Pals, Virtual Ethnography, and “Mail-Order” Marriages, 170; Jocelyn Eikenburg, “‘Is He Your Foreign Exchange Student?’ When You’re a White Woman Who Looks Older than Her Asian Husband,” Speaking of China, 2014, https://www.speakingofchina.com/china-articles/foreign-exchange-student-youre-white-woman-looks-older-asian-husband/. While in Hong Kong this might be less the case than elsewhere, such as in mainland China, some of my interlocutors did report similar incidents.

I kept trying with various prompts, and after six more attempts, I finally got something acceptable: “Romantic interracial couple: white woman, Chinese man – no Chinese woman” [Fig. 8]. At last, the woman looked like she could be a Western-white-Caucasian woman, and her partner indeed seemed Chinese. With both dressed elegantly, this was the symmetrically opposite image to the first one that this bot had generated.

By this point, I had already realized that this AI image-generating strategy was probably not going to work for my research. However, at this time I was also in the midst of writing my dissertation chapter on media representation and its impact on dating practices. As Gina Marchetti explains, Hollywood typically represents Asian men as evil or asexual, whereas Asian females are depicted as “sexually available to the white hero.” 9 Gina Marchetti, Romance and the “Yellow Peril”: Race, Sex, and Discursive Strategies in Hollywood Fiction (Berkeley: University of California Press, 1993), 2. Far from solely remaining on-screen fantasies, these representations profoundly shape actual dating practices. As one of my interlocutors explained, as an Asian man you have to work significantly harder to have success on the dating market. Because the impact of these tools extend way beyond the frustrations of a researcher interacting with a screen, we must ask ourselves: Why is it then that GenAI struggles so much to generate the image of a Chinese man paired with a white woman? And how does this blockage reflect on our current society and dating lives?

The problem is not limited to using free (perhaps low-quality) image generators. Soon after my little experiment, I read about other tech news outlets identifying similar challenges and biases when using GenAI. Particularly, someone writing for The Verge ran into the same issue I had encountered using Meta’s AI image generator, Imagine. 10 https://www.theverge.com/2024/4/3/24120029/instagram-meta-ai-sticker-generator-asian-people-racism The tool kept representing couples with two Asian partners instead of an Asian male and white female as requested by the author’s prompt, quite ironical for a tool called “Imagine.” The news soon caught the attention of other outlets, who also noticed that Meta’s own CEO, Mark Zuckerberg, is married to a woman of East Asian heritage. 11 https://www.businessinsider.com/zuckerbergs-wife-chinese-american-metas-ai-image-generator-cant-cope-2024-4

However, we cannot simply blame Zuckerberg for this. While AI is an amazing technology and a promising tool that could make our lives easier, it comes with many problems and dangers not reducible to any particular CEO. The main problem with this technology is that AI models are fed and trained by humans, mainly middle-class white men. Therefore, as numerous observers have pointed out, AI technology works to reproduce and reinforce race, gender, and class biases and inequalities. In the realm of representation, despite partly successful calls for increased representation of diverse pairings in popular culture – which have gained some traction through (limited) inclusivity in casting and filmmaking, and phenomena like fan fiction and the rise of K-culture – AI biases reproduce the unimaginability of certain relationships. Just like movie productions or TV series, the challenges that people encounter with producing images representing less mainstream couples through GenAI contribute to rendering such couples invisible while also shaping dating practices and desires. The call for more accurate and equitable representations, then, cannot be limited to the big screens; it must also include these new technologies, and it must start now that they are beginning to gain popularity with the public. Today, the question that anthropologist Nicole Constable asked back in 2003 in her book Romance on a Global Stage 12 Constable, Nicole. 2003. Romance on a Global Stage: Pen Pals, Virtual Ethnography, and “Mail Order” Marriages. Berkley: University of California Press. – “Why and for whom is it unimaginable?” (“it” being, in that study, American women looking for Asian husbands abroad) – has a new answer: it is unimaginable for AI because of its problematic training. 

Gabriella Angelini is a PhD candidate in Anthropology at the Chinese University of Hong Kong. Her research interests focus on transnational intimacies and the intersectionalities of race, gender, and class, examined through feminist and postcolonial theories. She also uses critical autoethnography as a complement to the traditional methods of anthropological inquiry. Email: g.angelini@link.cuhk.edu.hk