Explanatory Debugging: Supporting End-User Debugging of Machine-Learned Programs

Abstract

Many machine-learning algorithms learn rules of behavior from individual end users, such as taskoriented desktop organizers and handwriting recognizers. These rules form a "program" that tells the computer what to do when future inputs arrive. Little research has explored how an end user can debug these programs when they make mistakes. We present our progress toward enabling end users to debug these learned programs via a Natural Programming methodology. We began with a formative study exploring how users reason about and correct a text-classification program. From the results, we derived and prototyped a concept based on "explanatory debugging", then empirically evaluated it. Our results contribute methods for exposing a learned program's logic to end users and for eliciting user corrections to improve the program's predictions.

When users generate music playlists: When words leave off, music begins?

“…These concepts are then used as "natural" building blocks for new systems. The Natural Programming methodology [17] which investigates users' existing approaches to complete a task and ways of organising information by observing without influencing them about how the task should be done -has been used to design programming languages and systems, including interfaces that adapt themselves to user preferences [10]. We employed the Natural Programming methodology to understand what matters to end users in music playlists to inform future system design which encourages user involvement.…”

End-user interactions with intelligent and autonomous systems

“…Recent attempts have moved beyond explaining rule-based systems [25] toward more complex algorithms [20,26]. Examples of explanations for specific decisions include why… and why not… descriptions [14,16], visual depictions of the assistant's known correct predictions versus its known failures [29], confidence of the system in making predictions [13,19], and electronic "door tags" displaying predictions of worker interruptibility with the reasons (e.g., "talking detected" [31]). Recent work by Lim and Dey has resulted in a toolkit for applications to generate explanations for popular machine learning systems [17].…”

BeatBox

“…This allows interactive end-user creation of recognizers personalized or even tailored to the current task. For example, an end user might train a system to identify images related to a current concept of interest [5], apply customized qualitative codes in transcripts [10], automatically classify incoming email [16], or synthesize sound in response to a musician's gestures [4]. Due to the individualized nature of recognition in these types of applications, there is generally not pre-existing data available to train a classifier.…”