Trust in Automated Software Repair The Effects of Repair Source, Transparency, and Programmer Experience on Perceived Trustworthiness and Trust

摘要

Automation and autonomous systems are becoming increasingly pervasive in society, as are the software systems that control them. There is a need for safe and secure software systems. Automated code repair provides a promising solution. The present research investigates programmers' perceptions of trustworthiness and trust in automated code repair, how those perceptions and intentions differed from code ostensibly repaired by a human, and the effects of repair transparency. The present research comprises two studies, each with a unique sample. The first sample included inexperienced developers (N= 24), and the second sample included experienced developers (N = 24). Participants were presented with five different pieces of code before and after being repaired by an automated code repair program, and were asked to rate the trustworthiness of the repairs and whether they would endorse using the code. Each study was a 2 × 2 between-subjects design with repeated measures. The first factor manipulated the purported source of the repairs (human vs automated code repair program). The second factor manipulated the transparency of the repairs (deleted vs commented out). Results suggest that inexperienced developers find automated code repair more trustworthy than repairs made by a human. Both experienced and inexperienced developers trusted the human repairer less after reviewing the repairs, but did not significantly differ in their intentions to trust the automated code repair program after reviewing the repairs.