RGK Faculty - Ji Ma

As Large Language Model (LLM)-based agents increasingly undertake real-world tasks and engage with human society, how well do we understand their behaviors? We (1) investigate how LLM agents' prosocial behaviors -- a fundamental social norm -- can be induced by different personas and benchmarked against human behaviors; and (2) introduce a behavioral and social science approach to evaluate LLM agents' decision-making. We explored how different personas and experimental framings affect these AI agents' altruistic behavior in dictator games and compared their behaviors within the same LLM family, across various families, and with human behaviors. The findings reveal substantial variations and inconsistencies among LLMs and notable differences compared to human behaviors. Merely assigning a human-like identity to LLMs does not produce human-like behaviors. Despite being trained on extensive human-generated data, these AI agents are unable to capture the internal processes of human decision-making. Their alignment with human is highly variable and dependent on specific model architectures and prompt formulations; even worse, such dependence does not follow a clear pattern. LLMs can be useful task-specific tools but are not yet intelligent human-like agents.

Language shapes diverse cultures and creates natural barriers between human societies. The landscape of nonprofit and philanthropic studies in non-English languages is barely charted, impeding the globalization of this research field. This project (1) describes the topics shared between English and Chinese scholarship on nonprofits and philanthropy and (2) explores why English scholarship is cited in Chinese journal articles from five aspects: rationale of scholarship, novelty, relevance, social network, and reputation. The English articles cited by Chinese scholars tend to: (1) focus on instrumentality but not expressive values, (2) develop rather than disrupt existing paradigms, and (3) be relevant to topics popular in the Chinese literature and have authors with Chinese scholarly connections. In general, Chinese scholars tend to cite English articles that are value-neutral, non-disruptive, and native. Theoretical and methodological implications for examining nonprofit studies in other languages are discussed.

This research developed a machine-learning classifier that reliably automates the coding process using the National Taxonomy of Exempt Entities as a schema and remapped the U.S. nonprofit sector.

This study considers the effects of government funding to nonprofits from a network perspective. By analyzing a novel, 12-year panel dataset from the People's Republic of China, I find no evidence that government funding to a nonprofit crowds out private donations to the same organization. However, I find a substantial crosswise crowding-in effect at the ego network level: an increase of one Chinese Yuan in government funding to a nonprofit's neighbor organizations in board interlocking network can increase the private giving to the nonprofit by 0.4 Chinese Yuan.