ObjectiveThree digital ranked choice voting ballots were investigated to assess the impact of formatting on usability.BackgroundRanked choice voting (RCV) is a ballot format that allows voters to rank candidate choices in order of preference. The design of the ballot has a notable impact on usability, and this remains a challenge for RCV.MethodThis experiment investigated the usability of three RCV electronic ballot formats (bubble grid, column, and hand-numbered). Completion times, errors, subjective usability ratings, and trust data were collected.ResultsThe three formats took subjects 50-70 seconds on average to vote for a single contest. The ballots also had high error rates, which ranged from 23-38%. The hand-numbered format had the worst performance, as it had the highest completion times, error rates, and the lowest subjective usability ratings.ConclusionThis research highlights usability issues with RCV digital ballots. The formats tested were inefficient and ineffective, and performed poorly in comparison to previously tested plurality digital ballots.ApplicationThis experiment raises concerns about the scalability of RCV to larger elections. Ballots with RCV take longer to vote on and have higher error rates, which should be taken into consideration by election administrators.

ObjectiveThis study explores whether consumer education (CE) with and without driver responsibility content improves drivers' understanding and use of Advanced Driver-Assistance Systems (ADAS).BackgroundResearch suggests that CE with responsibility content may improve driver understanding; however, how other forms of learning interact, such as experiential learning, remains unclear.MethodEighty drivers (21-76 years) participated in a CE study using a Traffic Jam Assist ADAS. The study was conducted as a 2 (CE content: CE with responsibility content and CE without responsibility content) x 2 (driver engagement instructions: hands-free and hands-on) x 3 (timing of mental model assessment (MMA): pre-CE, post-CE, and post-drive).ResultsCE with responsibility content did not significantly enhance MMA understanding or confidence beyond CE without responsibility content in this study (p > .05). Regardless of responsibility content, drivers who received hands-on instructions consistently scored higher compared to hands-free instructions (p < .01). MMA confidence rating decreased at post-CE for all conditions but only returned to baseline levels at post-drive (p < .05) for drivers who received hands-free instructions.ConclusionDrivers who received hands-on instructions scored higher than hands-free at all; however, other forms of learning (e.g., experiential learning) may be necessary when the system substantially changes the driver's role.ApplicationCE may be enough to improve drivers' understanding if an update does not alter how a driver interacts with the system, or how it aligns with their initial mental model.

ObjectiveEvaluate the sensitivity of eye-tracking metrics to variations in mental workload in rail traffic control, and how they may be modulated by operator experience.BackgroundRail traffic control is a safety-critical domain characterized by high cognitive demands. While subjective workload measures are widely used, objective, real-time indicators, particularly eye-tracking metrics, remain underexplored.MethodEighteen professional rail traffic controllers (9 novice, 9 experienced) performed a simulation-based experiment involving four different scenarios with varying traffic density and task complexity, while wearing a head-mounted eye tracker. Measures of gaze behavior and visual strategies (fixations in different areas of interest (AOI)) were computed, and compared to subjective workload measures, to assess the effects of task difficulty and experience.ResultsSome measures like pupil dilation showed substantial changes (by ∼100%), while others like stationary gaze entropy showed more modest changes (∼40%), with increased task demand. Measures such as fixation rate revealed significant interaction effects with experience, indicating distinct visual strategies of novices vs. experts. Compared to experts, novices relied more on global overview displays under higher workload, and reduced planning-related information use, as revealed by AOI analysis. Subjective workload measures corroborated with increases in perceived demand.ConclusionThis study advances mental workload assessment in rail traffic control by demonstrating that eye-tracking metrics are sensitive to workload variations, and also reveal underlying cognitive strategies unique to expertise.ApplicationThese findings support the development of real-time workload monitoring systems and inform training interventions focused on information integration and cognitive strategies in rail traffic control operations.

ObjectiveThis study examined whether multidimensional recurrence quantification analysis (MdRQA) metrics derived from team heart rate data predict team performance and team workload under varied team and task demands.BackgroundAssessing team performance in operational environments is challenging. Traditional approaches require task interruptions or ground-truth data that are often unavailable. Physiological sensing offers a less-intrusive approach, yet most applications aggregate individual-level data or rely on simple pairwise metrics that do not represent how coupling is organized across the whole team. MdRQA instead embeds multivariate time series into a shared phase space, but evidence linking MdRQA metrics to team outcomes remains limited and mixed.MethodTwenty triads completed a cooperative virtual reality choice reaction time task across four conditions manipulating interdependence and time pressure. MdRQA metrics (percent determinism, average diagonal line length, diagonal line entropy) were extracted from heart rate interbeat interval data. Linear mixed-effects models tested whether these metrics predicted self-reported team workload and team performance outcomes including sensitivity and response bias from signal detection theory.ResultsLonger sustained coupling predicted increased sensitivity and fewer errors, while greater entropy predicted poorer performance. Higher determinism and lower entropy predicted a more liberal response bias. MdRQA metrics did not predict perceived workload beyond condition effects.ConclusionMdRQA metrics reflect joint physiological dynamics and predict performance outcomes in triads beyond the effects of task demands alone.ApplicationContinuously computed physiological coupling metrics may serve as less-intrusive, real-time proxies for team performance in operational environments where traditional assessment is infeasible.

ObjectiveTo assess older adults' experience, attitudes, and interests toward virtual reality (VR) applications as part of an early phase in the user-centered focused development of the Cognitive Activity Social Technology (CAST) VR system.BackgroundVR offers opportunities to enhance health and well-being in older adults by promoting cognitive, activity, social, and technology engagement. However, limited research has examined older adults' user experience and perceptions.MethodA multi-site study was conducted with 48 community-dwelling older adults aged 65-90. Participants interacted with 16 VR applications through hands-on sessions and first-person video demonstrations. They rated user experience, overall attitudes toward VR, and interest in using VR for various activities.ResultsMost applications were rated as enjoyable, interesting, and easy to use, with minimal reports of physical discomfort. Participants expressed the greatest interest in VR for art, cultural experiences, education, entertainment, and travel, while showing less interest in VR-based social interaction and fitness. Overall, participants found VR technology to be enjoyable and easy to use, but were neutral about VR's usefulness, likely due to the productivity-oriented nature of the usefulness measure.ConclusionThe findings demonstrated the feasibility, usability, and acceptability of VR applications for cognitive and activity engagement among older adults.ApplicationResults highlight the importance of designing VR experiences that align with older adults' preferences and needs. Findings inform user-centered design of VR systems to enhance cognitive engagement, social connectivity, and well-being among older adults.

ObjectiveWe examine AI-assisted tasks in which inadvertent safeguards can occur. An inadvertent safeguard refers to a counterintuitive scenario in which an AI error at the diagnosis/classification stage nevertheless produces a correct downstream decision recommendation. In this paper, we analyze the structural relationship between the AI's classification space and its recommendation outputs. We formalize the conditions under which inadvertent safeguards arise and compare the performance implications of Stage 3 action-selection support versus Stage 2 information-analysis support.BackgroundRecent research in human-automation and human-AI interaction has allowed for the structural possibility of inadvertent safeguards. However, this phenomenon has not been explicitly theorized or systematically examined.MethodWe first characterized the structural conditions under which inadvertent safeguards arise. We then conducted an experiment in which 90 participants completed a mental rotation task with a simulated AI decision aid that provided either Stage 2 information-analysis support or Stage 3 action-selection support.ResultsWhen inadvertent safeguards occur, Stage 3 action-selection support produced better performance, faster responses, and more positive trust adjustment than Stage 2 information-analysis support. Overall, Stage 3 action-selection support also produced higher perceived reliability, post-condition trust, and usability ratings.ConclusionThe impact of AI errors depends not only on classification accuracy but also on the mapping structure between classifications and actions. Many-to-one mappings, together with misclassifications between recommendation-equivalent states, can preserve correct downstream recommendations despite classification errors.ApplicationDesigners should distinguish classification accuracy from recommendation accuracy when evaluating Stage 3 action-selection support, especially in domains where multiple states can map to the same action.

ObjectiveThis study investigated the effects of correlated color temperature (CCT) on risky decision-making in fatigued individuals and the potential neural mechanisms underlying these effects.BackgroundUnderground facilities (UGFs) are increasingly utilized as military shelters, but the environmental stressors within them can easily lead to mental fatigue among commanders. The blue component within polychromatic white light, as indexed by CCT, has been shown to enhance alertness and cognition through non-visual pathways mediated by intrinsically photosensitive retinal ganglion cells (ipRGCs). However, it remains unclear whether CCT can benefit risky decision-making in fatigued individuals during the day.MethodWe examined the effects of varying CCTs (4000 K, 6500 K, 8500 K, 12,000 K) on fatigued individuals using a within-subjects design (N = 21) and assessed subjective alertness and mood, behavioral performance in attention and risky decision-making, and event-related potentials (ERP), including feedback-related negativity (FRN) and the P300 component.ResultsThe results showed that 12,000 K led to more favorable decisions, reflected by a lower risk ratio, more cash-outs, and higher revenue. ERP analyses revealed larger FRN and P300 amplitudes under 12,000 K, indicating that higher CCT enhanced cognitive control and promoted more conservative decision-making, enabling beneficial feedback-based decisions even under fatigue.ConclusionBlue-enriched white light improves alertness, mood, and decision-making in fatigued individuals, promoting gain-maximizing and risk-minimizing decisions.ApplicationThe current findings demonstrate that CCT may serve as a key parameter for light-based interventions aimed at alleviating fatigue in risky decision-making in UGFs.

ObjectiveWe examine how human-automation shared steering control combined with a haptic shape-changing interface for communicating and negotiating future intent affects driving performance and trust recovery.BackgroundSteering fights in haptic shared control can negatively impact driving performance. Existing solutions do not feature automation explicitly communicating its intent nor supporting intent negotiation with the driver. Trust recovery after faults is also expected to improve with explicit intent communication.MethodThirty participants took part in a continuous driving task consisting of 39 turns at intersections in which the pre-planned turn direction either matched or differed. Haptic communication of intent between driver and automation was metered under three conditions: From automation to driver (One-Way), in both directions (Two-Way), or not available (No-Way). Road obstacles were present at certain intersections to test handling of contingencies and faults. Automation faults were restricted to a second experimental period, leaving a third period to examine trust before and after a period of faults.ResultsDrivers in Two-Way communication displayed significantly improved driving performance when turn intents did not match. After the period of automation faults, drivers in Two-Way communication displayed significantly better overall trust recovery than drivers in One-Way or No-Way communication.ConclusionNegotiation is a viable means to avoid driver overrides and leads to improved trust recovery. Negotiation through a haptic interface should be considered an effective strategy to ensure function and cohesion in human-automation teams.ApplicationHuman-autonomy planning and team building.

ObjectiveTo identify the needs of older adults when interacting with artificial intelligence (AI) technologies, and to propose design recommendations that can inform the development of future AI systems.BackgroundAI technologies are increasingly integrated into healthcare, education, and social applications for older adults. However, many AI systems fail to account for age-related cognitive, perceptual, and emotional changes, resulting in usability barriers, low trust, and limited adoption.MethodA systematic review of 93 studies categorized by application area identified challenges and user needs through thematic analysis, leading to proposed design recommendations. Two AI-powered platforms were then benchmarked against these design recommendations to assess their suitability for older adults.ResultsFindings highlighted older adults' concerns related to usability, personalization, privacy, and explainability of these technologies. Older adults preferred systems that respect human autonomy, adapt to user context and emotions, and offer clear, trustworthy explanations. A set of actionable design recommendations was proposed to improve AI-powered technologies for older adults. Analysis of two selected AI platforms suggested that these systems only partially aligned with the proposed recommendations.ConclusionThis study offers a set of design recommendations that can support the development of AI tools attuned to older adults' needs.ApplicationThis review identifies critical design gaps and offers actionable design recommendations. Findings can inform future participatory, interdisciplinary efforts in age-inclusive AI development.

ObjectiveThis study explored whether and which cognitive abilities are supported by various levels of automation.BackgroundParasuraman's (2000) model of automation provided guidance for designing automation to augment cognitive tasks. Subsequent mixed empirical findings and evolving understanding of attention control in complex tasks suggested an evaluation of how automation supports cognition within a task was needed.MethodParticipants completed cognitive ability tests then a targeting task either manually or with information automation or decision automation.ResultsDecision automation supported working memory only when reliable, while attention control predicted performance regardless of automation level.ConclusionResults suggest automation can support specific processes like working memory, while attention control demands may persist across automation levels. However, because the task involved a relatively constrained form of automation, future research should examine whether these effects hold with more flexible, adaptive systems.ApplicationAutomation should target specific cognitive functions while recognizing that attention control demands may persist. Future research might investigate how other automation designs can better support this foundational ability.

ObjectiveTo identify driving maneuvers in remote driving under end-to-end latency and examine operator adaptation through within-session maneuver transitions.BackgroundRemote driving systems enable operators to control highly automated vehicles when systems encounter operational limits. End-to-end latency disrupts adaptation and vehicle control, although operators may adjust driving behavior. Research on within-session adaptation clusters and maneuvers emerging under latency remains limited.MethodThirty-six participants completed nine remote driving simulator sessions, navigating construction zones under end-to-end latencies (0-640 ms). Hidden Markov Models identified driving maneuvers from vehicle control data (speed, steering, throttle). Dynamic time warping and K-means clustering classified drives by maneuver-transition sequences. Chi-square tests and ANOVAs examined associations among adaptation clusters, participant-specific differences, experimental conditions, and human-factors measures, while participant-clustered factorial analysis examined maneuver composition.ResultsFour maneuvers (adapted, transitional, overcompensation, and compensatory) and three adaptation clusters (No adaptation, Rapid adaptation, and Gradual adaptation) emerged. Adaptation clusters were associated with participant-specific differences rather than latency conditions or repeated trials. Rapid adaptation drives were associated with lower workload and higher trust than No adaptation drives and showed a trend toward increasing trust, whereas trust significantly declined in No adaptation cluster.ConclusionRemote driving operators exhibit adaptation clusters characterized by maneuver-transition sequences. Transitional maneuvers reveal intermediate stages. Cluster distributions were more strongly associated with participant-specific differences than tested experimental conditions, with adaptation patterns differing in trust and workload.ApplicationTraining programs should assess adaptation capabilities and provide interventions rather than uniform protocols. Early cluster identification enables training on maneuver transitions that facilitate progression toward adaptation.

ObjectiveComputationally defined visual salience can predict attentional deployment in complex, e-commerce displays but less is known about the effect of salience on later cognitive processes. For instance, can salience also predict later memory for e-commerce products? If so, visual salience could be leveraged to inform design.BackgroundOnline shopping environments provide a wealth of visual information, with numerous products competing for attention. When consumers visit e-commerce pages, they may not have a specific purchase goal. Understanding attentional guidance in these complex displays is critical. Visual salience may influence what products consumers notice and later remember.MethodsHigh- and low-salience products were identified in naturalistic e-commerce webpages using a computational model. During the experiment, participants viewed the webpages and were told their memory for all products would be tested. Recognition memory for individual products was then tested.ResultsParticipants demonstrated better recognition memory for products that had high visual salience, compared to low, within their studied e-commerce page. These memory results mirror the relationship that exists between overt attention and visual salience and extend them to show a persisting effect of salience on information processing.ConclusionVisual salience enhances product recognition memory in naturalistic e-commerce pages. A computational salience model can, therefore, be useful in predicting subsequent memory.ApplicationStudy results offer insights for e-commerce designers. A computational salience model can be leveraged to improve product visibility and memorability. The model is particularly useful for quantifying salience in complex displays where images vary across feature dimensions.

ObjectiveThis study examines the effectiveness of systematic viewing for closed-circuit television (CCTV) scanning within the domain of (remote) nautical object (bridge/lock) control.BackgroundSystematic viewing during nautical object control is assumed to mitigate the likelihood of missing critical information (e.g. road user) in CCTV streams by structuring operators' visual search. However, prior studies across domains have shown mixed results regarding whether structured scanning approaches enhance performance.MethodForty-two professional operators performed a remote-bridge-control simulation task in which unexpected critical events were embedded. Half of the participants received instructions in a systematic viewing protocol, while the other half operated without protocol guidance.ResultsOperators trained in the protocol showed significantly higher gaze systematicity with longer dwell times and higher coverage of protocol-defined CCTV areas. Despite these changes, both groups detected the same proportion of critical events. Looked-but-failed-to-see errors occurred more often than search errors overall, but the distribution of error types did not differ significantly between the groups. First fixations on the critical events occurred at similar times, although the no-protocol group responded significantly faster to the critical event. Gaze systematicity did not correlate with detection performance.ConclusionThe systematic viewing protocol successfully induced more structured viewing behaviour, with the protocol group engaging in more extensive scene inspection prior to initiating a response. However, higher gaze systematicity did not translate into improved detection performance.ApplicationSystematic viewing protocols can promote more extensive CCTV scanning among professional operators, but do not guarantee the detection of critical events.

ObjectiveWe investigate whether decision-making processes in human-AI teams mirror those in human-only teams in a task where team members hold common and unique information. Specifically, we examine how information inquiry and performance expectations influence these processes.BackgroundHuman-AI decision-making has traditionally been studied as a linear, segmented process where AI provides recommendations and humans make final decisions. However, recent advancements in large language models (LLMs) enable increasingly dynamic interactions similar to the collaborative processes in human-only teams, where members have to share and integrate unique information to reach decisions. Despite the potential for similar dynamics in human-AI teams, research on such configurations-particularly on when unique information is shared-remains limited.MethodWe conducted an experiment using a hidden profile task, a well-established paradigm in team research. We manipulated team type (Nhuman-AI = 98; Nhuman-only = 65, three-member teams) and information inquiry, and created conditions with differing performance expectations.ResultsInformation inquiry enhanced information sharing and, consequently, improved decision-making performance, irrespective of performance expectations and team type. Exploratory analyses suggested that while this effect was consistent across team types, information inquiry patterns may differ.ConclusionOur findings prompt avenues for future research on how humans' communication norms vary when interacting with AI versus human team members. Methodologically, this study demonstrates how established decision-making paradigms can be adapted to incorporate AI, offering insights into both human-only and human-AI team processes.ApplicationThe results have practical implications for designing AI systems that promote effective team decision-making by engaging in information inquiry.

ObjectiveTo examine how task anomalies and time pressure influence cognitive workload in procedural tasks and to identify whether workload effects are transient or sustained.BackgroundProcedural tasks often involve sequential actions for which disruptions, such as operational anomalies or time pressure, may exceed operator capacity. Understanding the temporal dynamics of workload under these conditions is essential for designing adaptive systems to effectively support human performance.MethodThirty participants performed three cooking tasks under nominal and off-nominal conditions (trials with induced operational errors and time pressure). Workload was continuously assessed using subjective ratings and physiological measures including skin conductance level (SCL), percent change in pupil size (PCPS), and blink rate (BR).ResultsSubjective workload was consistently greater in off-nominal conditions. Physiological measures varied in sensitivity to the task condition and type of task: BR reliably indicated elevated task demands, PCPS was sensitive only under multiple operational anomalies, and SCL showed limited responsiveness to all factors. Two distinct workload patterns were observed: isolated operational anomalies led to transient workload spikes, while multiple accumulating anomalies produced sustained high-workload states, reflecting a "workload history" effect.ConclusionCognitive workload responses depend on both the type and accumulation of task disruptions. BR provides a robust real-time marker, while subjective ratings capture perceived effort.ApplicationFindings inform the design of adaptive support systems capable of distinguishing between momentary operational challenges and cumulative strain, enabling tailored operator assistance in procedural and safety-critical domains.