三輪 洋文

Detecting and addressing discrimination is one of the most crucial ways for ethnic and sexual minorities and women to fully integrate into the society. However, humans often fail to correctly judge what constitutes discrimination. The recent rapid development of artificial intelligence (AI) based on large language models (LLMs) has shown promise in assisting with this task, although its performance is understudied. This study investigates how humans and LLM-based AI, such as OpenAI’s ChatGPT, detect the concept of ‘discrimination’ and how their representation differs. Specifically, surveys were conducted asking humans (Japanese respondents) and ChatGPT (GPT-4) to evaluate the degree of discrimination in hypothetical unequal treatment scenarios presented to them. The scenarios varied in terms of targets, including their ethnicity, gender, and sexuality, and types of discrimination, such as those based on tastes, stereotypes, and statistics. The results show that ChatGPT generally classifies the scenarios as more discriminatory than humans do. However, ChatGPT also shares a tendency with humans to be more tolerant of unequal treatment based on ethnicity and gender compared to sexuality, and it is less likely to detect statistical discrimination than taste- or stereotype-based discrimination. Although LLM-based AI presents a potential tool for addressing discrimination and can offer temporary solutions, it may not fully capture all types of discrimination.

In coalition governments, while parties need to compromise with their partners to govern together, they also need to differentiate from the partners to please their supporters. How do coalition parties reconcile compromises with differentiation? We argue that coalition parties, especially junior partners, adjust their strategies to compromise or differentiate themselves from their partners depending on the level of public support for their partners. Parties tend to differentiate from partners when partners receive lower public support, and compromise with partners when they enjoy higher public support. To test this argument, we analyzed parliamentary speeches by legislators from a junior coalition partner in Japan. We applied a dynamic linear model to the results of machine learning text classification of legislators’ speeches. Our results show that as the senior party’s public support rate falls, the speech styles of legislators from the junior partner increasingly diverge from those of the senior partner’s legislators.

In Japan, the gender gap in attendance at 4-year universities is narrowing, yet significantly fewer women apply to selective colleges. A growing body of literature suggests that gendered expectations of educational returns and cultural stereotypes, most compellingly reflected in parental influence, are pivotal in shaping children’s college choices. However, the capacity to make causal claims remains limited in existing research, as observed gendered preferences among parents may merely echo earlier educational investments made before college applications. This study advances the field by leveraging a survey experiment approach. Our conjoint experiment, where respondents evaluate hypothetical college choices of fictitious children, yields several key findings. First, responses show no difference when the children, regardless of their gender, opt for selective colleges. Second, the respondents who expect higher returns from college education for men than for women tend to favor their fictitious son’s selective college applications. Third, fictitious daughters’ college choices are rated more favorably when they conform to societal gender stereotypes. These findings underscore the potential influence of recent admissions reforms in Japanese universities, including explicit quotas for female applicants, aimed at increasing their enrollment.

Voter preferences for a politician’s gender have been considered a potential factor of women’s underrepresentation in politics. Although recent studies demonstrate that voters do not generally dislike women candidates in elections, they may still doubt women’s competence and view persuasive behavior as counter-stereotypical for women. This may lead to the prediction that female politicians are less effective in persuading voters through their policy statements compared to men. However, evidence on this issue is scarce and mixed, with existing studies often having notable weaknesses in their experimental design. To address these problems, we carefully designed and conducted two preregistered survey experiments in the U.S. and Japan, where electoral rules are candidate-centered, and women’s underrepresentation is severe compared to other developed countries. We asked respondents how much they agreed with the policy statement of a fictitious politician whose gender was randomized in a mocked campaign website or a municipal council newsletter. We also varied the statement’s issue area and ideological position to examine whether persuasion is more effective when it aligns with gender stereotypes. The results showed no significant difference in policy persuasiveness between women and men politicians, and post hoc equivalence tests suggest the nonexistence of even minimal effects in both countries. Moreover, we found partial evidence that women politicians had an advantage when they claimed gender-stereotype statements in Japan, while the opposite was true in the U.S.

Japan is an intriguing case in the literature of judicial independence because of the Liberal Democratic Party's (LDP) long-lasting one-party dominance in government—which, in theory, leads to low independence. Although scholars have found plentiful anecdotal observations implying the LDP's judicial control, quantitative evidence remains scarce. Focusing on the Supreme Court as the first step, this study provides a new dataset on justices' ideal points through an extensive compilation of justices' voting data and application of the dynamic item response theory model. I validate the estimates' interpretability as ideological positions. The results present several novel findings: justices with a career-judge background are relatively conservative; the Chief Justice tends to be particularly conservative; the conservative camp has tended to hold on to a court majority since the 1960s; and the government's partisanship, to some degree, influences appointed justices' positions. These results reinforce scholars' views that Japan's court leans conservatively and aligns with the LDP, with implications for judicial independence in Japan.