Algorithmic Bias in the Calligraphy Education Evaluation System

Abstract

This study examines the algorithmic biases embedded in AI-driven calligraphy education evaluation systems and their profound implications for cultural and educational equity. By analyzing existing commercial systems (e.g., RJET Xiumo, Shanghai Writing Proficiency Test Digitalization System), the research identifies three core dimensions of bias: (1) Technical-mechanical reductionism, where rigid metrics (e.g., stroke standardization, grid positioning) disregard the dynamic artistry and emotional expression central to calligraphy, such as misjudging masterpieces like Wang Xizhi’s "Lanting Xu" as flawed; (2) Data hegemony, reflected in skewed training datasets dominated by Tang regular script (67%) and modern works (91%), while marginalizing minority scripts (e.g., seal script, Dongba script) and historical styles, leading to inaccurate evaluations of non-mainstream calligraphy; and (3) Cultural violence, where algorithmic standardization imposes industrial logic on artistic practice, eroding aesthetic diversity and pedagogical autonomy. Empirical evidence reveals that these biases exacerbate educational inequalities—e.g., disadvantaging rural students and suppressing stylistic innovation—while reinforcing technocratic control over cultural valuation. The study calls for a human-centered redesign of AI systems, emphasizing data justice, algorithmic transparency, and symbiotic collaboration between technologists, educators, and cultural stakeholders to preserve calligraphy’s essence as a "heart art" in the digital age.