Embodied AI: A Cognitive Shift, Not a Cosmetic Change
summary
Argues that giving AI voice or a face flips a cognitive switch that changes user behavior and trust, not just appearance, and demonstrates three working apps: a live voice assistant embedded in slides, a real-time voice-to-voice translator, and two debating avatars that run on a cascaded pipeline (speech-to-text → LLM → text-to-speech) versus lower-latency native audio. Cites success with a healthcare voice agent involving 5,000+ physicians, warns of McDonald’s/IBM failure, shows safeguards in GitHub notebooks and notes running cost ~15¢/minute and target 200 ms latency.
AI-written summary · about
chapters (15)
- 0:00Introduction
- 1:05Humanizing AI inevitability
- 3:38Defining virtual embodiment
- 6:27Belief vs behavior in AI
- 9:32When anthropomorphism helps or harms learning
- 11:22Voice latency in AI
- 13:57Demo: embedded voice assistant
- 15:20Voice agent success/failure
- 17:27Safeguards in AI deployment
- 18:32Accessibility with voice/avatars
- 20:18Demo: English-Japanese translation
- 21:55Avatars and cognitive shift
- 26:19Avatar realism challenges
- 27:02Resources for voice/avatar systems
- 28:28Q&A
related talks
57mDesigning Human-AI Experience
16mWeak signals of a vocal revolution in UIs
55mDon’t Make Me Prompt: Intelligent Interfaces Users Actually Like
27mWhat's Next in Voice for Brand Leaders
50mDesigning Experiences For Voice Interactions
38mAI-Powered Conversational Interfaces
46mDesigning Artificial Intelligence
12mBuilding Voice AI for People Who Don’t Trust AI