CONFIDANT: A Privacy Controller for Social Robots
summary
A privacy controller called CONFIDANT classifies conversation context and blocks sensitive robot disclosures using speechbrain for speaker ID, Google Speech-to-Text and Google Natural Language for transcription, topic and sentiment, sentence-BERT for semantic similarity, and Apriori association rules to map label combinations to low/moderate/high control. Trained from scenario labels (n=174) and evaluated on filmed Misty II interactions (n=95), the controller raises perceived privacy-awareness, trustworthiness and social-awareness versus a baseline and detects explicit secrecy phrases via semantic similarity.
AI-written summary · about
chapters (9)
- 0:00Introduction to Confidant
- 1:38Privacy challenges in social robots
- 2:24Design of Confidant privacy controller
- 3:51Machine learning models in Confidant
- 5:02Dataset collection for training
- 6:50User study and privacy scores
- 7:34Evaluating Confidant on Misty II
- 8:42Results and user trust impact
- 9:29Future improvements and conclusion
related talks
13mPrivacy Communication Patterns for Domestic Robots
10mPrivacy and Respectful Discourse in AI Chatbots
31mPanel: Trustworthy AI: Enhancing Human Agency and Machine Accountability in the Age of Agentic AI
25mPolisis: Automated Analysis and Presentation of Privacy Policies Using Deep Learning
01h 17mPrivacy and New Technologies
07mVision: Human-as-the-Unit Privacy Management with AI Agents
36mData privacy in an AI-driven world
46mAI That Reads Emotions: What It Means for Data Privacy, Consent, Your Teams