Who is 'Asian'? Challenging Essentialist Categorization Through Scientific Communication.

Abstract

Research on racial categorization has largely treated 'Asian' as a monolithic category, despite substantial heterogeneity in national origins, phenotypic features, and cultural practices within this population. This dissertation examines how the category 'Asian' is represented in human cognition, artificial intelligence systems, and psychological research practices, with particular attention to systematic bias toward East Asian prototypes that renders South Asian, Southeast Asian, and Central Asian populations less visible. Study 1 investigated how differences in (i) categorization instructions and (ii) response categories provided would affect face-sorting behavior with 139 participants across three independent tasks. When provided with disaggregated Asian subgroup categories (East Asian, Southeast Asian, South Asian, Central Asian), participants recognized a substantially broader range of individuals as belonging to some Asian subgroup compared to conventional binary Asian/non-Asian choices, with increases of up to 46 percentage points for faces with the lowest baseline categorization proportions. Study 2 examined whether AI image generators exhibit similar biases. Forty images generated using prompts with varying diversity reminders were categorized by 191 participants. Results revealed that AI systems default to East Asian representations, with substantial between-platform differences. Critically, diversity-focused prompts did not significantly alter AI outputs, suggesting surface-level prompt engineering cannot override deeply embedded patterns in training data. Study 3 conducted a scoping review of 408 studies across eight psychology journals. Among 91 studies mentioning Asian participants, 89.0% used only the broad label 'Asian' without ethnic disaggregation, and 58.8% failed to specify whether identity was self-reported or researcher-assigned. Study 4 examined how 336 participants perceived the accuracy of different identity measurement formats. Pan-ethnic umbrella terms received the lowest accuracy ratings, while formats allowing participants to select self-descriptive labels received higher ratings. Collectively, these findings reveal systematic biases in how 'Asian' identity is operationalized across racial categorization behavior, AI systems, and research practices – suggesting that categorization patterns reproduce essentialist boundaries historically encoded in legal and bureaucratic definitions, and that these patterns can be challenged through methodological transparency and disaggregated measurement approaches.