Attribution Errors by People and Intelligent Machines

Abstract

Objective To explore the ramifications of attribution errors (AEs), initially in the context of vehicle collisions and then to extend this understanding into the broader and diverse realms of all forms of human–machine interaction. Background This work focuses upon a particular topic that John Senders was examining at the time of his death. He was using the lens of attribution, and its associated errors, to seek to further understand and explore dyadic forms of driver collision. Method We evaluated the utility of the set of Senders’ final observations on conjoint AE in two-vehicle collisions. We extended this evaluation to errors of attribution generally, as applicable to all human–human, human–technology, and prospectively technology–technology interactions. Results As with Senders and his many other contributions, we find evident value in this perspective on how humans react to each other and how they react to emerging forms of technology, such as autonomous systems. We illustrate this value through contemporary examples and prospective analyses. Applications The comprehension and mitigation of AEs can help improve all interactions between people, between intelligent machines and between humans and the machines they work with.

How and why humans trust: A meta-analysis and elaborated model

“…The decomposition of variables thus far is informed by, and guided by, both the overall trust literature and the model of human-interpersonal trust that we have previously offered ( Hancock et al, 2011b ; Schaefer et al, 2016 ; and see also Hancock et al, 2021a , b ). However, the three general predictor categories can themselves be further sub-divided.…”

“…However, as with all meta-analyses, ours is necessarily only a contemporary assessment of such trust and cannot provide a complete anticipation of future developments. In respect of this limitation on forms of punctate assessment, it is important to conduct sequences of meta-analyses in order to establish how the state of understanding evolves across time ( Hancock et al, 2021a , b ). So, in order to facilitate optimal workplace harmony, further empirical examinations are required to address the current shortfalls in the experimental record.…”

“…Yet another meta-analysis reported by Maguin (2010) examined only the associations between interpersonal trust and various team performance characteristics. And most recently, two of our own trust meta-analyses have focused on concerns largely beyond the human-human realm by featuring human-robot trust ( Hancock et al, 2011b , 2021a , b ). Yet another such work has reported on the broader aspects of human-automation trust ( Hancock, 2014 ; Schaefer et al, 2016 ), while the most recent of our group’s efforts have evaluated trust in AI systems ( Kaplan et al, 2021 ).…”

“…This model is founded upon the prior work of Mayer et al (1995) . In the present version, we have superimposed a triad of factors which are explicitly derived from our own program of experimental and integrative work (see, e.g., Hancock et al, 2011a , 2021a , b ; Kaplan et al, 2020 ). In particular, we emphasize a dimension not previously integrated with extant human trust research, namely the context within which the trust interaction occurs.…”

Drivers of partially automated vehicles are blamed for crashes that they cannot reasonably avoid

“…It is a topic of debate whether or not failing to monitor automation and intervene when necessary is a typical ‘human error’ that should be remedied by policies 21 , 44 , 48 , 49 , more training and public education, and increased automation; or whether it is a symptom of inappropriate human-automation interaction design that should be remedied by human-centered design methodologies, for example through shared control 23 . Either way, we believe that such failures would still occur and a reasonable approach for responsibility attribution in such cases should be considered.…”