Historical Shifts and Geographical Drifts: An Exploration of AI Embedding (2025-2026)
This project examined how AI systems represent the meaning of words across time and languages, and how those representations can be made easier for people to understand. AI converts words into numerical "embeddings" that capture meaning, but these systems often function as a black box. The project investigated how the meanings of AI-related and value-laden terms have changed over the past 30 years, whether concepts are represented consistently across different languages and cultures, and how to communicate these complex patterns through interactive visualizations.
The team developed new tools to analyze changes in word meaning over time and across languages. Team members fine-tuned AI models using 30 years of academic and newspaper text, creating visualizations that tracked how technical and value-related terms evolved. They also built a multilingual analysis pipeline that compared how concepts shifted across six languages using multiple AI models, revealing where meanings aligned or diverged across linguistic and cultural contexts. These methods produced both quantitative analyses and visual outputs that made complex AI representations easier to interpret.
To broaden public understanding of AI, the team translated its research into an interactive exhibit, "Lost in Embedding Space," at the Rubenstein Arts Center. The exhibit featured immersive installations that allowed visitors to explore how AI-related language has evolved over time and how concepts are represented differently across languages. By combining AI research with art, interactive technology and data visualization, the project made abstract ideas about AI more accessible while creating new tools for studying how meaning changes across time and cultures.
Timing
Fall 2025 – Spring 2026