Literary characters and GPT-4: from William Shakespeare to Elena Ferrante.

This article examines the behavior and personality traits of 148 well-known literary fictional characters from the 17th to the 21st century by prompting the large language model GPT-4 to play the Dictator game, a behavioral economics experiment. The study finds a general decline in selfishness over time, with 50% of 17th-century characters’ decisions being selfish compared to 19% in the 21st century, and male characters exhibiting more selfishness than female characters, especially in earlier centuries. GPT-4-generated personality traits show a strong net positive valence overall, increasing notably in the 21st century, where traits like "empathetic," "fair," and "selfless" are most prominent, contrasting with earlier centuries’ emphasis on traits such as "manipulative," "ambitious," and "ruthless." While GPT-4’s modeled characters display human-like features such as altruism and sensitivity to zero-cost giving, the model diverges from human behavior in showing low sensitivity to relative ordinal position and price elasticity. This approach offers a novel computational method for analyzing literary characters and their evolution across time, with implications for both behavioral economics and literary studies.