Solving the Next Billion-People Privacy Problem
summary
Argues that virtual assistants risk becoming the next monopoly privacy sink and proposes an open, federated stack: an open-source assistant Almond, a high-level programming language ThingTalk, a shared API catalog Thingpedia, a semantic parser (LUINET) and a distributed protocol (DTP). Prototype runs on Android and web, achieves over 95% function identification and ~90% end-to-end program accuracy with simple quoting, handles 200 functions, a 200-person needs survey yields 220 sharing scenarios (85 unique) and shows 70% of constraints are enforceable today.
AI-written summary · about
chapters (16)
- 0:00Intro
- 1:26facebook
- 5:54What Can Academia Do? (2)
- 10:00Turn Virtual Assistants into Our Friend
- 12:26Key Technology Programming virtual assistants in natural language
- 14:23Natural Language Programming: Asthma Example
- 17:32Almond: 1st Programmable Virtual Assistant
- 19:34Thingpedia: Encyclopedia of Things
- 23:12Technical Challenges
- 25:41Almond: 1st Distributed Virtual Assistant
- 27:13Virtual Assistant: Programmable Sharing Agent
- 33:56TACL: Thing Talk Access Control Language
- 34:16TACL Parsing Accuracy
- 34:29Automatically Generated GUI
- 40:08User Study: "Sharing Without Passwords"
- 41:24Summary
related talks
31mAlmond: Crowdsourcing an Open, Programmable Virtual Assistant
30mHastening The Apocalypse: Self-replicating Personal AI Assistants
10mA Day in the Life: An Immersive Data Experience
23mPersonal OS: Build Your Army of 100 AI Interns
26mOpen AI Assistants
44mWell-Being, AI, and You
07mVOICE 2021 | Chris Butler, IPsoft
07mVision: Human-as-the-Unit Privacy Management with AI Agents