Effects of Automation Type on Human Performance in Proofreading Tasks

摘要

Automation is ubiquitous and indispensable in modern working environments. It is adopted andused in not only advanced industrial- and technology-oriented operations, but also ordinary homeor office computational functions. In general, automated systems aim to improve overall workefficiency and productivity of labor-intensive tasks by decreasing the risk of errors, and cognitiveand physical workloads. The systems offer the support for diverse decision-making processes aswell. However, the benefits of automation are not consistently achieved and depend on the typesand features of automation (Onnasch, Wickens, Li, & Manzey, 2014; Parasuraman, Sheridan, &Wickens, 2000). Possible negative side effects have been reported. Sometimes, automation maylead to multi-tasking environments, which allows operators to be distractive with several tasks. Itultimately prolongs task completion time and causes to neglect monitoring and follow-up steps ofthe pre-processing tasks (Endsley, 1996). Furthermore, the operators who excessively depend onautomation are easily deteriorated in skill acquisition, which is necessary for the emergency ormanual operations. Thus, inconsistent performance in automation is a major issue in successfuladoption and trust in automation (Jeong, Park, Park, & Lee, 2017). This paper presents anexperimental study that investigates the main features and causes of the inconsistency in taskperformance in different types of automation. Automated proofreading tasks were used in thisstudy, which is one of the most common types of automation we experience in daily life. Based onthe similar algorithm of the auto-correct function in Microsoft Word, a custom-built program offive proofreading tasks, including one non-automated and four automated proofreading tasks, weredeveloped using Visual Studio 2015 C#. In the non-automated task used as a reference forindividual difference, participants were asked to manually find a typographical error in a sentence.In the automated tasks, auto-correcting functions are provided in two levels (i.e., low and high) ofautomation and two statuses (i.e., routine and failure of automation). The type of automation isdefined as the combinations of a status and a level. Participants identified typographical errors byonly an underlined word at the low-level automation, whereas an underlined word with a possiblesubstituting word was given at the high-level. Additionally, in the routine automation status, acorrect substituting word is provided. On the other hand, a grammatically incorrect word is givenin the failed automation status. Nineteen participants (11 females and 8 males; age mean = 33.8,standard deviation = 19.1) took part in this study. Results of statistical analyses show a clearadvantage in high-routine automation, in terms of both task completion time and accuracy. Whiletask performances of high & routine automation types are quite obvious in both task completiontime and accuracy, those in the failed automation types are mixed and indistinguishable. Differentlevels and statues of failed automation do not much influence task performance. Moreover, taskcompletion time and mental demand are strongly correlated, and the accuracy rate and perceivedtrust show a strong positive correlation. The approaches and outcomes of the current study canprovide some insights into the human-automation interaction systems that support humanperformance and safety, such as in-vehicle warning systems and automated vehicle controls.