Virtual assistants, such as Alexa, Google Home, and Siri, are revolutionizing our digital life. In the future, they will provide us with a uniform, fully personalized, natural language interface to all our diverse data sources, web services and IoT devices. The virtual assistant will become a powerful platform as it sees all our personal data and has great influence over the services and vendors we use. We propose a collaborative research effort to develop open-source virtual assistant technology, in concert with a commercially viable distributed infrastructure that safeguards users' data privacy, supports interoperability, and promotes open competition. To jumpstart this effort, we have developed Almond, a working open-source prototype of distributed virtual assistants. Almond lets users use natural language to share their data, IoT, and services with fine-grain control, while preserving their privacy by keeping data on their own devices. Almond also lets users issue advanced commands that monitor real-time events and connect multiple services together. This research lays the groundwork for the creation of three key open, non-proprietary, collaborative virtual assistant resources: (1) Thingpedia, a repository of virtual assistant skills, (2) LUInet (Linguistic User Interface Network), a neural network that translates natural language into ThingTalk, a virtual assistant programming language, and (3) a Distributed ThingTalk Protocol, DTP, that supports sharing with privacy via cooperating virtual assistants.
