PAL Boot Camp: Preparing Cognitive Assistants for Deployment

Abstract

Most visions for decision support and information technology anticipate the use of machine learning to enable software to respond to an adapting environment, including the ability to learn capabilities while on-the-job. Currently, systems and software engineering processes hinder employment of task learning technology, because the adaptation it provides runs counter to our notions of stability. Similarly, systems must typically demonstrate satisfaction of requirements before deployment, rather than learn tasks while on the job. This paper introduces new problems for the field of software engineering and discusses an approach for preparing cognitive systems for deployment. We describe one approach to a boot camp for cognitive systems and present the results of simulations of the boot camp. The results of our experiments provide thresholds and patterns for knowledge, and the requirement for specific patterns of human use of cognitive systems. These results are then used to infer requirements for a boot camp and measures for the prediction of successful employment of the assistant.