World Learning

People around the world share broad experiential goals such as “relaxing” or “having fun,” but the concrete activities that realize these experiences vary widely by place. Existing tools like search engines, guidebooks, travel blogs, and large language models (LLMs) can answer questions such as “What do people do for fun in Japan?”, yet they typically present one country at a time and require users to already know what to ask. These answer-first interfaces make it hard to see how a single experience is realized differently across countries, or to surface surprising contrasts that might provoke deeper questions. This paper presents WorldLearning, an interactive system that uses an LLM and a world map to help users explore how local people in different countries typically have fun in their free time. Given a user-specified activity (for example, “going to karaoke”), WorldLearning uses an LLM as a scoring model to estimate how typical that activity is as something locals do for fun in each country, and visualizes the resulting scores as a choropleth map with ranking lists and follow-up explanation tools. The system is designed to transform interaction with LLMs from one-shot question–answer exchanges into a dialectical process in which visualized patterns generate new questions that users then probe further. A preliminary study with two participants engaging in guided think-aloud exploration suggests that WorldLearning can surface unexpected global patterns, prompt questions that users would not have asked on their own, and support sustained curiosity-driven inquiry about why everyday experiences are realized differently across countries.