Automation bias and verification complexity: a systematic review.

No prior work had resolved whether over-reliance on automated systems is restricted to multitasking environments. Prior research has shown that decision support tools can inadvertently introduce novel error types. That uncertainty drove this investigation into the broader literature beyond human factors engineering. It was already known that users sometimes neglect manual information verification when using digital aids. This gap motivated a cross-disciplinary comparison between clinical and industrial settings. Researchers previously assumed that vigilance decreases primarily when individuals manage multiple concurrent demands. However, the prevalence of these errors in specialized diagnostic settings remained poorly understood. This study addresses the discrepancy between established theoretical models and observed performance in complex decision-making tasks.

The aim of this review is to compare human factors and health care literature regarding the association of bias with multitasking. This study addresses the uncertainty surrounding whether over-reliance on technology is restricted to complex, concurrent environments. The researchers seek to clarify the role of task complexity in decision-making errors. No prior work had resolved how these errors manifest across different professional domains. This investigation explores whether diagnostic tasks are more prone to failure than monitoring tasks. The authors intend to evaluate the prevalence of these issues in studies where manual verification is possible. They hope to determine if cognitive load is a more significant predictor than multitasking. This work provides a comprehensive synthesis of evidence to guide future system design and error mitigation strategies.

Review approach involved searching nine major academic databases from 1983 through 2015. The investigators selected evaluations where automated tools assisted human performance and resulted in measurable errors. Participants had to possess the capability to verify system accuracy and execute tasks manually. The team identified and grouped various task types to facilitate cross-disciplinary comparisons. Each activity received a rating based on its specific verification complexity. The researchers noted the presence of multitasking alongside the type of automation employed. They synthesized findings from 40 studies, including six from the health care sector. This rigorous screening process ensured that only relevant performance data were included in the final analysis.

Key findings from the literature reveal that over-reliance on technology occurs in single tasks, contrary to established human factors theories. The researchers identified 40 relevant studies from an initial pool of 890 publications. Bias was frequently observed in diagnostic scenarios rather than simple monitoring activities. High verification complexity emerged as a consistent factor linked to these performance errors. The authors report that the literature is currently fragmented, with large discrepancies in reporting standards. Few studies provided statistical significance testing when compared to a control condition. Results suggest that cognitive load, rather than multitasking, is the primary driver of these decision failures. The data indicate that automated systems can introduce risks even when users are not managing concurrent demands.

The authors propose that over-reliance on automated aids is not exclusively tied to multitasking scenarios. Synthesis and implications suggest that high verification complexity contributes significantly to user error rates. Evidence indicates that diagnostic tasks are particularly susceptible to these cognitive failures. The researchers note that the current body of literature remains highly fragmented across different fields. They highlight that few studies provide robust statistical comparisons against manual control conditions. Strategies for mitigation should prioritize the reduction of cognitive load during high-stakes decision processes. The findings imply that designers must account for task difficulty when implementing automated support systems. Future efforts should aim to standardize how these performance discrepancies are reported in scientific publications.