How can end users efficiently influence the predictions that machine learning systems make on their behalf? This paper presents Explanatory Debugging, an approach in which the system explains to users how it made each of its predictions, and the user then explains any necessary corrections back to the learning system. We present the principles underlying this approach and a prototype instantiating it. An empirical evaluation shows that Explanatory Debugging increased participants' understanding of the learning system by 52% and allowed participants to correct its mistakes up to twice as efficiently as participants using a traditional learning system.

Most programs today are written not by professional software developers, but by people with expertise in other domains working towards goals for which they need computational support. For example, a teacher might write a grading spreadsheet to save time grading, or an interaction designer might use an interface builder to test some user interface design ideas. Although these end-user programmers may not have the same goals as professional developers, they do face many of the same software engineering challenges, including understanding their requirements, as well as making decisions about design, reuse, integration, testing, and debugging. This article summarizes and classifies research on these activities, defining the area of End-User Software Engineering (EUSE) and related terminology. The article then discusses empirical research about end-user software engineering activities and the technologies designed to support them. The article also addresses several crosscutting issues in the design of EUSE tools, including the roles of risk, reward, and domain complexity, and self-efficacy in the design of EUSE tools and the potential of educating users about software engineering principles.

Abstract-Research is emerging on how end users can correct mistakes their intelligent agents make, but before users can correctly "debug" an intelligent agent, they need some degree of understanding of how it works. In this paper we consider ways intelligent agents should explain themselves to end users, especially focusing on how the soundness and completeness of the explanations impacts the fidelity of end users' mental models. Our findings suggest that completeness is more important than soundness: increasing completeness via certain information types helped participants' mental models and, surprisingly, their perception of the cost/benefit tradeoff of attending to the explanations. We also found that oversimplification, as per many commercial agents, can be a problem: when soundness was very low, participants experienced more mental demand and lost trust in the explanations, thereby reducing the likelihood that users will pay attention to such explanations at all.

Although machine learning is becoming commonly used in today's software, there has been little research into how end users might interact with machine learning systems, beyond communicating simple "right/wrong" judgments. If the users themselves could work hand-in-hand with machine learning systems, the users' understanding and trust of the system could improve and the accuracy of learning systems could be improved as well. We conducted three experiments to understand the potential for rich interactions between users and machine learning systems. The first experiment was a think-aloud study that investigated users' willingness to interact with machine learning reasoning, and what kinds of feedback users might give to machine learning systems. We then investigated the viability of introducing such feedback into machine learning systems, specifically, how to incorporate some of these types of user feedback into machine learning systems, and what their impact was on the accuracy of the system. Taken together, the results of our experiments show that supporting rich interactions between users and machine learning systems is feasible for both user and machine. This shows the potential of rich humancomputer collaboration via on-the-spot interactions as a promising direction for machine learning systems and users to collaboratively share intelligence.

Spreadsheet languages, which include commercial spreadsheets and various research systems, have had a substantial impact on end-user computing. Research shows, however, that spreadsheets often contain faults; thus, we would like to provide at least some of the benefits of formal testing methodologies to the creators of spreadsheets. This article presents a testing methodology that adapts data flow adequacy criteria and coverage monitoring to the task of testing spreadsheets. To accommodate the evaluation model used with spreadsheets, and the interactive process by which they are created, our methodology is incremental. To accommodate the users of spreadsheet languages, we provide an interface to our methodology that does not require an understanding of testing theory. We have implemented our testing methodology in the context of the Forms/3 visual spreadsheet language. We report on the methodology, its time and space costs, and the mapping from the testing strategy to the user interface. In an empirical study, we found that test suites created according to our methodology detected, on average, 81% of the faults in a set of faulty spreadsheets, significantly outperforming randomly generated test suites.