Seeing Without Knowing: AI-Powered Insights from Public Spaces

A new case study details a privacy-preserving video analytics system that extracts valuable behavioral data from pose-based representations, offering a powerful alternative to traditional surveillance methods.
![The system models the dynamics of oral arguments by simulating justice responses-predicting each [latex]n^{th}[/latex] turn based on case facts, the legal question, prior conversational context from the preceding [latex]n-1[/latex] turns, and the identity of the speaking justice-through both prompt-based methods utilizing varied base models and agentic approaches leveraging tools like case docket searches and historical voting data, all evaluated by a framework assessing both the realism and pedagogical value of the simulations.](https://arxiv.org/html/2603.04718v1/2603.04718v1/x1.png)







