Virtual Information Officer (VIO)

Overview

The Virtual Information Officer (VIO) system is a machine-learning based agent that learns to understand and processes factual changes to a database. The principle interaction sequence consists of five steps: (i) a user submits an input free-text message describing the change to the database, (ii) a machine learning based analysis procedure that translates the message into a filled-in form update, (iii) inspection and correction of the proposed form update by user (or other human), (iv) execution of the update, and (v) a feedback loop that improves the machine learning based on the user's interaction with the form.

To ground our vision in a specific application, we have built an agent that functions as a webmaster assistant. For example, a user emails the request: "Change John Doe's home phone number to 800-555-1212" to the agent. The webmaster agent then replies with the biographical data form displaying information about John Doe with the new phone number already entered. The user then simply approves the change. This website operates with a VIO agent. The VIO design (i) allows users to express the updates they want to make in human terms (free text input expression of intent), (ii) reduces the hunt time people spend looking for the correct form, and (iii) allows users to quickly repair any inference errors the agent makes.

In addition, we have measured, in control laboratory experiments, user interactions with agents that demonstrates that the VIO is efficient (faster) and more effective (less errors made to site) than sending a request to a human webmaster. Finally, since the VIO observes the user's interaction, the machine learning performance of the analysis step improves over time.