Reproduce “Position: Human-Robot Interaction in Embodied Intelligence Demands a Shift From Static Privacy Controls to Dynamic Learning” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Shuning Zhang, Hong Jia, Simin Li, Ting Dang, Yongquan `Owen' Hu, Xin Yi, Hewu Li
Embodied robots can compromise privacy by inferring sensitive facts from individually innocuous observations, even when raw data are protected. The paper frames this inferential privacy problem as a context-dependent learning and alignment task, proposing a Contextual Integrity- and CIRL-based framework in which agents learn personalized privacy costs from explicit instructions, corrections, and multimodal social cues.
Key insights
- Privacy protection must cover latent beliefs, reasoning, memory, and downstream behavior—not only sensor inputs or communicated data.
- A privacy norm can be acceptable in one context and inappropriate in another, so privacy alignment should be personalized, situational, and dynamically learned.
- The position paper identifies cross-context generalization, noisy feedback, conflicting multi-user preferences, changing norms, and machine unlearning as major unresolved challenges.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.