Developing an International Space Station curriculum for the Bootstrapped Learning program

Abstract

DARPA's Bootstrapped Learning (BL) program is aimed at advancing the state of the art in instructable computing. Two objectives of this program are developing a general electronic student that makes use of machine learning algorithms to learn from the kind of focused instruction typically provided by a human teacher and creating a repository of automated curricula that can be taught to the student. This paper focuses on the second objective, describing a curriculum developed for the BL program to both instruct and test the student that places the electronic student (eStudent) in the role of an International Space Station (ISS) flight controller. The eStudent is taught how to detect and diagnose single-fault problems within the thermal control system of the ISS. During each lesson, the eStudent interacts with an ISS simulator to review alerts and access telemetry values. To obtain greater visibility into its diagnostic reasoning, the eStudent is trained to create an external representation of its reasoning about the current problem - a diagnostic rationale. This includes describing potential problems, hypothesizing possible events and states, positing possible causal explanations as rationale assertions, seeking evidence for or against these assertions, projecting possible risks, and using possible risks to focus attention when developing a rationale. In addition to describing the curriculum developed as part of the first year of the BL program, we also describe some of the future directions we will investigate as part of the second year. 1,2