ABSTRACT
Explainable artificial intelligence aims to bring transparency to artificial intelligence (AI) systems by translating, simplifying, and visualizing its decisions. While society remains skeptical about AI systems, studies show that transparent and explainable AI systems can help improve the Human-AI trust relationship. This manuscript presents two studies that assess three AI decision visualization attribution models that manipulate morphological clarity (MC) and two information presentation-order methods to determine each visualization's impact on the Human-AI trust relationship through increased confidence and cognitive fit (CF). The first study, N = 206 (Avg. age = 37.87 ± 10.51, Male = 123), utilized information presentation methods and visualizations delivered through an online experiment to explore trust in AI by asking participants to complete a visual decision-making task. The second study, N = 19 (24.9 ± 8.3 years old, Male = 10), utilized eye-tracking technology and the same stimuli presentation methods to investigate if cognitive load, inferred through pupillometry measures, mediated the confidence-trust relationship. The results indicate that low MC positively impacts Human-AI trust and that the presentation order of information within an interface in terms of adjacency further influences user trust in AI. We conclude that while adjacency and MC significantly affect cognitive load, cognitive load alone does not mediate the confidence-trust relationship. Our findings interpreted through a combination of CF, situation awareness, and ecological interface design have implications for the design of future AI systems, which may facilitate better collaboration between humans and AI-based decision agents.
ABSTRACT
We report results of a study that utilizes a BCI to drive an interactive interface countermeasure that allows users to self-regulate sustained attention while performing an ecologically valid, long-duration business logistics task. An engagement index derived from EEG signals was used to drive the BCI while fNIRS measured hemodynamic activity for the duration of the task. Participants (n = 30) were split into three groups (1) no countermeasures (NOCM), (2) continuous countermeasures (CCM), and (3) event synchronized, level-dependent countermeasures (ECM). We hypothesized that the ability to self-regulate sustained attention through a neurofeedback mechanism would result in greater task engagement, decreased error rate and improved task performance. Data were analyzed by wavelet coherence analysis, statistical analysis, performance metrics and self-assessed cognitive workload via RAW-TLX. We found that when the BCI was used to deliver continuous interface countermeasures (CCM), task performance was moderately enhanced in terms of total 14,785 (σ = 423) and estimated missed sales 7.46% (σ = 1.76) when compared to the NOCM 14,529 (σ = 510), 9.79% (σ = 2.75), and the ECM 14,180 (σ = 875), 9.62% (σ = 4.91) groups. An "actions per minute" (APM) metric was used to determine interface interaction activity which showed that overall the CCM and ECM groups had a higher APM of 3.460 (SE = 0.140) and 3.317 (SE = 0.139) respectively when compared with the NOCM group 2.65 (SE = 0.097). Statistical analysis showed a significant difference between ECM - NOCM and CCM - NOCM (p < 0.001) groups, but no significant difference between the ECM - CCM groups. Analysis of the RAW-TLX scores showed that the CCM group had lowest total score 7.27 (σ = 3.1) when compared with the ECM 9.7 (σ = 3.3) and NOCM 9.2 (σ = 3.4) groups. No statistical difference was found between the RAW-TLX or the subscales, except for self-perceived performance (p < 0.028) comparing the CCM and ECM groups. The results suggest that providing a means to self-regulate sustained attention has the potential to keep operators engaged over long periods, and moderately increase on-task performance while decreasing on-task error.
