Perceiving justice in the AI city: Global evidence from an eye-tracking study of autonomous mobility

Perceiving justice in the AI city: Global evidence from an eye-tracking study of autonomous mobility | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Perceiving justice in the AI city: Global evidence from an eye-tracking study of autonomous mobility Anu Masso, Mergime Ibrahimi, Ahti-Veikko Pietarinen, Frauke Behrendt, and 1 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8195075/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract AI increasingly shapes urban mobility, yet its social justice implications remain poorly understood, especially across cultural contexts. This study examines how people perceive justice in AI-based mobility systems through visual attention and cognitive framing. Combining large-scale online eye-tracking (n = 1,272) with survey data from 22 cities worldwide, we analyse how individuals attend to and interpret social diversity across autonomous (AV) and human-driven contexts. Three key patterns emerge. First, visual attention to social categories such as nationality or income is structured and transferable across technological settings. Second, attentional variation reflects both individual (age, gender, AI experience) and contextual (regional, cultural) factors. Third, Bayesian modelling shows that justice perceptions depend less on visual engagement than on affective trust and perceived safety. Together, these findings reveal how automation reshapes cognitive-ethical orientations toward justice, illustrating that equitable AI mobility depends as much on perceptual inclusion and social recognition as on algorithmic fairness. Biological sciences/Psychology Social science/Psychology Social science/Science technology and society Artificial intelligence urban mobility autonomous vehicles social justice data justice eye-tracking cross-cultural variation Figures Figure 1 Figure 2 Figure 3 Figure 4 Introduction Artificial intelligence (AI) is rapidly transforming urban governance, planning, and public service delivery. Positioned at the forefront of digital urbanism, AI is often framed as a key enabler of sustainability – particularly in mobility 1 , where optimisation algorithms promise cleaner, more efficient, and adaptive transport systems. Autonomous vehicles (AVs) exemplify this imaginary: they are heralded as intelligent and inclusive alternatives capable of reducing congestion and emissions while expanding access. Yet beneath this sustainability narrative lies a persistent tension between technological promise and social reality. For example, AI-driven mobility systems risk reproducing, rather than remedying, inequalities in access, recognition, and representation 2 3 4 . Sustainable urban AI must therefore be reframed to include not only environmental efficiency but also social and cultural sustainability – where inclusion becomes a foundational design principle, not an afterthought. At the international level, these concerns are increasingly recognised. Reports such as UN-Habitat’s AI & Cities (2022) 5 and the Global Assessment of Responsible AI in Cities (2024) 6 call for context-sensitive governance of AI in urban environments. They emphasise that the ethical and social implications of AI depend not only on technical design but also on how technologies are interpreted and received locally. The OECD’s AI Capability Indicators (2025) 7 similarly underline the need to understand how AI aligns with human abilities and social expectations – a perspective still missing from many urban applications. Against this backdrop, extensive research has examined AI-based urbanism, from the broader promises of data-driven efficiency 7 to critiques of technocratic governance 8 . Critical smart city scholarship has shown that algorithmic systems often reinforce inequality, obscure accountability, and shape perceptions of safety and risk 9 10 . In mobility, these dynamics are especially acute: weak governance can entrench socially and environmentally harmful trajectories, and AVs raise concerns around exclusion, bias, and accessibility. Yet while theoretical critiques are abundant and empirical work has emerged 11 12 13 14 , studies that explore how people actually engage with such systems remain limited – particularly those attentive to the perceptual and cultural diversity of these encounters in urban mobility. Yet despite AI operating within highly diverse social environments, we still know surprisingly little about how this diversity is perceived, negotiated, or transformed within AI-mediated mobility systems. Existing debates on AI justice and inclusion tend to rely on abstract principles rather than lived experience, and they are dominated by Global North perspectives that treat AI as a neutral optimisation tool rather than a socially embedded system shaped by power 15 16 . While research on ‘smart transport’ has expanded 17 18 19 , the perceptual and experiential dimensions of how people encounter these systems remain largely overlooked, particularly across different cultural and socio-economic settings. This risks sidelining how people from diverse contexts perceive, trust, or resist algorithmic mobility systems. These dynamics resonate with findings from recent research on Mobility-as-a-Service (MaaS) 20 , which shows how algorithmic governance can reinforce inequality when sustainability goals are weakly defined or externally imposed. Yet such studies primarily focus on structural dynamics and overlook how these systems are interpreted, contested, or embodied by the people who navigate them. Moreover, cognitive and visual processes 21 22 – how individuals perceive and attend to AI systems in everyday settings – remain largely absent from current debates on urban justice and governance. Recent evidence even suggests that AI technologies can reshape human perception and culture 23 , underscoring the need to examine how these systems are seen and experienced in urban contexts. In this study, we approach justice in AI-based urban mobility not merely as a design or governance issue but as a culturally shaped perceptual experience. We combine large-scale online eye-tracking with survey data to examine how people in 22 cities worldwide see and interpret social diversity in AI-based mobility. While surveys capture explicit attitudes, they reveal little about implicit, affective processes that shape how people relate to AI technologies long before forming conscious opinions 22 . By linking visual attention to justice perceptions, we explore how cultural and contextual factors shape the cognitive construction of inclusion and recognition. We use autonomous vehicles as a salient case within the broader landscape of AI-based urban systems. Specifically, we ask: What patterns characterise visual attention to social diversity in autonomous versus human-driven mobility contexts? Which individual and contextual factors explain attention variation towards social diversity across AV and non-AV (human-driven) settings? How is visual attention to diversity related to participants’ perceptions of justice in AI-based mobility? These questions reveal the aim at cognitive and cultural mechanisms through which AI-based mobility systems are accepted, contested, or mistrusted in everyday life. The findings contribute to designing urban AI systems that are not only technically sound but also socially legitimate and perceptually inclusive. Perceiving and framing justice in AI-based urban mobility Recent debates on algorithmic urbanism emphasise that justice in data-driven governance extends beyond outcomes to the ways people experience and interpret algorithmic systems. Critical approaches to urban and mobility justice 24 25 26 27 28 29 show that technologies are embedded in social and power hierarchies that shape whose needs and movements are prioritised in urban mobility. The concept of data justice expands this perspective by highlighting how data practices reproduce or challenge inequalities of visibility, participation, and control 30 31 32 . In the context of AI-driven mobility, where decisions depend on large-scale data infrastructures, justice concerns not only what algorithms decide but also how they configure who and what becomes knowable and governable. The more recent notion of mobility data justice 33 brings these issues into the transport domain, showing how algorithmic mobility systems distribute movement and agency unequally. Together, these debates underscore an important distinction between computational fairness and broader questions of justice. While fairness has often been operationalised within data-driven frameworks as a computational concern focused on bias detection and procedural control 30 32 , justice extends beyond technical mitigation to encompass the structural, historical, and distributive dimensions of how data and technologies shape social relations 31 . Dencik and colleagues caution that such technical framings risk detaching data practices from their social and political contexts 30 , while Heeks and Renken conceptualise fairness more narrowly as perceived process control in data handling 32 . Taken together, these perspectives provide the foundation for analysing how justice is perceived and felt in AI-based mobility – linking structural questions of data justive with people’s cognitive, affective, and evaluative experiences. This approach understands justice as a dynamic, value-based process grounded in recognition and inclusion 31 . It highlights two complementary dimensions: normative justice, reflecting shared ideals about fairness, and experiential justice, referring to how these ideals are enacted and interpreted in concrete social contexts. Understanding how people move between these two dimensions – between what they value and what they feel – is essential for assessing the societal legitimacy of AI-based mobility. Such experiential judgements do not form in isolation; they are shaped by broader cultural and perceptual processes that influence how people interpret inclusion, recognition, and inequality in cities 34 35 36 . Public perceptions of justice depend not only on technical performance but also on whether individuals feel represented or marginalised within these systems 34 35 36 . Cross-national and social-psychological research shows that responses to inequality and diversity are mediated by both individual predispositions and contextual cues 37 , as described by intergroup contact theory 38 , social identity theory 39 , social categorisation theory 40 , and perceived threat theory 41 . Yet these frameworks rarely address technologically mediated environments. As algorithmic systems increasingly structure access, recognition, and exposure in urban life 42 43 44 45 46 , the perceptual mechanisms underlying justice evaluations may shift in ways not fully captured by existing theories – particularly in the super-diverse contexts described by Vertovec 47 . This highlights the need to examine how AI-based mobility systems shape whether, and for whom, justice is perceived as being realised in cities. This gap calls for integrating insights from perception and social cognition research into the study of AI justice. Perception is not merely a passive process but an active form of sense-making that filters social meaning through visual and affective cues 48 49 . In AI-based mobility systems, such cues often precede conscious reasoning, shaping intuitive judgements of justice and legitimacy. AI-based systems may therefore modulate justice perceptions indirectly – by altering what draws attention, what feels salient, and what remains unseen. Understanding these processes requires not only attitudinal data but methods that capture how attention itself is distributed across socially meaningful stimuli. Visual framing plays a particularly important role in this context. As Grabe and Bucy 21 demonstrate in Image Bite Politics , images profoundly shape how people cognitively and emotionally make sense of complex social phenomena. Building on this insight, we explore how visual attention and framing influence perceptions of justice and inclusion in AI-based mobility. This is especially relevant in the case of autonomous vehicles (AVs), where public engagement is largely mediated through speculative imagery, simulation, or symbolic representation rather than direct use 48 50 . These visual imaginaries do not only depict technology but also encode assumptions about social order – about who is seen as a passenger, who as a pedestrian, and whose safety is prioritised. Visual cues – such as who is chosen to be depicted as passenger or pedestrian, or how risk and safety factors are distributed across the scenarios – subtly influence how justice, legitimacy, and inclusion come to be perceived. Representations of AVs often evoke imaginaries of control, harmony, or environmental efficiency, yet these framings may obscure underlying inequalities in accessibility and recognition. Studying how people visually process such scenes can thus reveal broader normative orientations toward justive and belonging. By combining framework of data justice with insights from visual framing and notion of super-diversity, this study moves beyond static definitions of fairness and diversity to examine how they are perceived, felt, and visually negotiated in encounters with AI-based mobility. Together, these processes form a theoretical bridge between data justice and perceptual justice – linking structural power with individual sense-making. This highlights an important theoretical insight: for justice to be realised in AI-based systems, it is not sufficient that algorithms operate fairly in a technical sense; justice must also be perceived as such by those affected. Understanding these links is crucial for evaluating not only the ethics but also the social intelligibility of AI systems – how people come to ‘see’ algorithmic mobility as fair or exclusionary, and how such perceptions might reinforce or resist existing urban inequalities. Empirical design and methods This study applies a mixed-method design to examine how individuals perceive social diversity and justice in AI-based urban mobility. Combining large-scale online eye-tracking with attitudinal survey data, it investigates how people visually attend to, interpret, and evaluate social groups in autonomous (AV) and non-autonomous (non-AV) mobility scenarios. Rather than focusing on moral decision-making, the study explores the perceptual mechanisms through which social difference and justice are cognitively constructed in AI-mediated environments. Our design builds upon, yet extends, the Moral Machine experiment 51 , which explored public reasoning about ethical trade-offs in autonomous vehicle dilemmas. While Moral Machine was instrumental in foregrounding public engagement with AI ethics, later scholarship has discussed both its methodological strengths and conceptual limits – especially its focus on moral preferences rather than perceptual or contextual dimensions of human-AI interaction 52 53 54 55 . Building on these insights, our study shifts the analytical focus from moral choice to social perception , examining how individuals visually perceive, recognise, and interpret social difference in AI-based mobility environments. We employed a semi-experimental online design combining scenario-based eye-tracking with a structured survey. Figure 1 illustrates the operationalisation of intersectional diversity. Participants viewed scenarios in which either hypothetical autonomous vehicle passengers or pedestrians were prioritised and indicated their choice via mouse click while eye-tracking recorded gaze trajectories. Each scene featured characters differing across multiple diversity attributes – race or ethnicity, gender, age, disability, income, vaccination status, micro-mobility use, and urban or rural residence – modelled after intersectionality and super-diversity theory 56 47 . Scenarios were generated through a Python-based randomisation script to ensure balance and clarity. All participants viewed 7 scenarios in AV condition, and the same 7 scenarios in non-AV condition. Participants were randomly assigned to begin with either AV or non-AV conditions. Participants were recruited through the CINT crowdsourcing platform, producing a globally distributed sample of 1,272 valid participants from 22 cities across Europe, Asia, Africa, and the Americas. City selection reflected the degree of AI and AV experimentation, indexed using the Cities in Motion Index (CIMI) and publicly available data on pilot projects. Stratified quotas aligned participant distributions with national census data for gender, age, and education. Further demographic and contextual details are provided in Supplementary Informaiton Tables S1-S3. Eye-tracking was conducted using the RealEye platform, which employs AI-based webcam gaze prediction (accuracy ≈110 pixels). Visual stimuli were designed to approximate hypothetical yet realistic urban mobility situations while allowing systematic variation in social composition and vehicle type. All visual stimuli were created using the same resolution, brightness, and contrast settings to ensure visual consistency and minimise potential low-level visual confounds (e.g., differences in contrast or sharpness). Fixation data were aggregated into total fixation time per area of interest (AOI) and z-standardised within participants to ensure comparability across conditions. After the eye-tracking task, participants completed a short survey capturing justice perceptions, social sensitivity, and attitudes toward AI. Drawing on the framework of data justice 31 , the survey measured perceived inclusion and recognition in AI-based mobility. Additional indicators included trust in AI, perceived safety, AI experience, gender, minority status, education, and region, reflecting well-established predictors of public responses to automation and social diversity 57 . Ethical clearance was obtained prior to data collection, and all participants gave informed consent. Full details of the survey items, calibration, and validation procedures are presented in the Supplementary Information. Analytically, the study proceeded in three stages. First, mean fixation times across diversity categories depicted in the visual scenarios were compared between AV and non-AV contexts using paired-samples t-tests to establish baseline variation in attention to attributes represented in the stimuli, such as the gender, age, disability status, or income level of the depicted individuals. Repeated-measures ANOVA was then used to analyse variation with participants’ socio-demographic background variables. Second, due to strong correlations between category-level measures, we constructed a Diversity Attention Index (DAI) aggregating fixation times across all categories. The DAI captured each participant’s overall attentiveness to social diversity cues and enabled comparisons between mobility contexts and subsequent modelling of individual and contextual variation. Third, we examined how DAI values related to participants’ justice perceptions and background characteristics using Bayesian multilevel regression models. This approach allowed us to estimate uncertainty in fixed and random effects while accounting for both individual- and city-level variation. We focus on perceived justice as the primary outcome. A validation analysis using an alternative justice measure (normative justice) is included in the Supplementary Information. Together, these steps provide a multilevel understanding of how people attend to and evaluate social difference in AI-based mobility. Integrating perceptual and attitudinal data reveals not only what individuals notice and where , but how they interpret justice and inclusion within algorithmically mediated urban environments. Results Patterns of visual attention to social diversity The first stage of analysis examined how participants visually attended to different social categories in autonomous (AV) and non-autonomous (non-AV) traffic scenarios. Clear and systematic differences emerged depending on both the social category and the transport technology (Figure 2). Nationality, residence, and income received the highest levels of attention across both conditions, indicating that these socially entrenched dimensions remain highly salient in people’s cognitive framing of urban mobility. Each relates to long-standing urban debates about segregation, belonging, and the perceived legitimacy of movement through public space. In contrast, people with reduced mobility received the least attention overall, suggesting that this form of diversity remains visually and cognitively marginal in representations of everyday mobility. Yet, the shorter fixation time may also indicate rapid, automatic normative judgement – that people with reduced mobility should be protected without the need for further deliberation. The high standard deviation (57.1 in the case of AV and 56.8 in the case of non-AV) for this category indicates substantial individual variation – some participants focused strongly on reduced mobility cues, while others almost ignored them. A comparable pattern appeared for micromobility, represented by an electric scooter user. Although scooters have become common in cities, their social meaning is still in flux. This instability was mirrored in gaze patterns, reflecting both curiosity and uncertainty about how new mobility forms fit within existing urban hierarchies. Intermediate levels of attention were observed for age and COVID-19 status. Both categories were textually distinctive, but keeping possibly neutral visual presentation in the stimuli. Despite this visual salience, neither drew strong or consistent attention compared with more normatively charged categories such as nationality or income. This suggests that while participants recognised these differences, they did not interpret them as socially decisive cues in mobility contexts. Category-level variation was statistically confirmed by paired-samples T-test (all p < .001), with full pairwise comparisons presented in Supplementary Table S2. Statistical comparisons (Figure S3 and Table S7 in Supplementary Information) showed that visual attention patterns differed significantly across social categories. In AV contexts, participants directed more attention toward female figures and pedestrians, and slightly more toward younger individuals, Europeans, White and Chinese figures, and those facing economic difficulties (all p < .01). Differences for other groups were minor. These findings indicate that automation amplifies selective perceptual engagement with social difference, particularly for categories associated with visibility, agency, or vulnerability in the traffic scene. Beyond category-level variation, a clear technological effect emerged. Across all dimensions, the shift from non-AV to AV contexts was accompanied by an overall increase in visual attention to social diversity. Mean fixation times increased by approximately 12–16 ms. While the absolute magnitude may seem modest, such shifts are meaningful in eye-tracking research given that the scenarios were tightly matched. The difference therefore reflects a systematic increase in attentional engagement with social cues once automation entered the scene, as also indicated by the effect sizes of 8.32 to 10.66 across categories (t-tests). The increase was most pronounced for people with reduced mobility, but also for micromobility category – categories that are socially newer or less institutionally represented – suggesting that automation may lessen existing gaps in visibility. At the same time, variation across categories increased in the AV condition, indicating wider individual differences in how participants noticed or ignored social diversity under automation. This variability likely reflects the absence of stable perceptual norms in technologically mediated mobility settings. Although total attention increased, the relative hierarchy of categories remained remarkably stable. Participants continued to allocate most attention to nationality, residence, and income, implying that entrenched perceptual hierarchies persist even as the technological mediation of mobility changes. In other words, increases compresses the overall level of attention but does not alter who is seen. People carry forward habitual ways of perceiving urban difference – who stands out, who fades into the background – even as the material infrastructures of mobility evolve. Taken together, these findings show that visual attention to social diversity is patterned yet context-sensitive. Automation increases the total social information processed and amplifies the variability of perceptual engagement. Technological change therefore modulates rather than transforms existing hierarchies of attention. This highlights that justice in AI-based mobility depends not only on algorithmic design or governance but also on perceptual inclusion – what, and whom, people actually see in the automated city. To further explore how these attentional patterns aggregate across individuals and contexts, the next section introduces the Diversity Attention Index (DAI) as a composite measure of perceptual engagement with social diversity. Variation across individual and contextual factors While category-level comparisons revealed systematic variation in how participants attended to different forms of social diversity, the strong intercorrelations between categories indicated that these attentional tendencies were not independent. Individuals who devoted more attention to one category also tended to focus more on others. This coherence suggests an underlying cognitive disposition toward noticing, or overlooking, social difference in mobility contexts. To capture this general tendency, we constructed a composite Diversity Attention Index (DAI) aggregating fixation times across diversity dimensions. Factor analysis supported a one-factor solution, confirming a common latent component underlying attention to diversity (see Supplementary Figure S1). The DAI provided an integrated measure of how people attend to social difference overall, allowing analysis of cross-contextual consistency between AV and non-AV scenarios as well as variation across individual and contextual factors. Building on this index, the second stage of analysis examined how DAI values varied across individual and contextual characteristics (Figure 3, Supplementary Table S8). Attentional orientations to diversity were not evenly distributed but reflected systematic variation across minority status, AI awareness, regional background, age, and family composition. These results demonstrate that attention to diversity is shaped by both sociocultural positioning and experiential familiarity with technology, suggesting that perceptual engagement with social difference is a situated and learned capacity rather than a universal cognitive trait. The strongest and most consistent effect emerged for minority status. Participants who self-identified as ethnic or social minorities displayed lower Diversity Attention Index values than majority respondents in both AV and non-AV contexts. Participants with higher AI awareness showed a similar pattern, displaying lower Diversity Attention Index scores, particularly in non-automated contexts. Regional and sociodemographic factors together added further layers of variation in diversity attention. Respondents from Europe showed higher diversity attention index values compared with those from Asia, while participants from Africa and the Americas showed an intermediate position. These regional contrasts likely reflect distinct cultural framings of social visibility and inequality: in more stratified or rapidly transforming contexts, diversity may be less perceptible, whereas in stable or institutionalised settings it tends to be more visible, more normatively integrated into everyday life, or supported by the greater social and environmental diversity of urban contexts. Similar context-dependent effects appeared across age and family status. Middle-aged participants (45–54) showed the highest attention to social differences, followed by the youngest and oldest groups, while participants aged 30–44 exhibited the lowest overall attention. This pattern suggests generational differences in social engagement and adaptability, possibly reflecting distinct life-stage priorities and exposure to technological and social change. The lower diversity attention among participants aged 30–44 may relate to their greater work and family responsibilities, leaving less cognitive or emotional capacity for engaging with social difference. This may also explain why participants with children showed lower diversity attention overall. Income had smaller yet consistent effects. Those facing economic difficulties displayed marginally higher diversity attention index scores, aligning with research 58 linking empathy and social perception to structural vulnerability. Associations with religiosity were also weaker, indicating that religious participants paid less attention to social diversity. This pattern may suggest that stronger religious identification is linked to a more inward or community-oriented focus, leaving less perceptual openness to broader social difference. Alternatively, it may reflect the influence of shared moral or cultural frameworks that reduce the salience of visible diversity in social perception. Taken together, these findings show that attention to diversity is structured, selective, and context-dependent. Minority identity and technological familiarity both reduce perceptual sensitivity to social difference, yet they do so through distinct mechanisms – one through withdrawal and invisibility, the other through normative selectivity and focus. In this sense, the justie of AI-based mobility depends not only on algorithmic transparency or representational balance but also on how automation reconfigures who is seen, who is overlooked, and whose difference becomes legible within data-driven urban environments. Linking attention to perceived justice indicators The final stage of analysis examined how visual attention to social diversity relates to perceptions of justice in AI-based mobility. A hierarchical Bayesian regression model was used to predict perceived justice from a set of perceptual, experiential, attitudinal, and demographic predictors (Figure 4). Posterior estimates were computed with weakly informative priors and are presented as mean effects with 95% credible intervals. This framework allowed reliable estimation of uncertainty and robust identification of associations across individual and city levels. The justice index was constructed from six correlated survey items capturing people’s lived experiences of justice, inclusion, and recognition in AI-based mobility; detailed variable composition, validation, and comparisons with the general justice value measure are provided in the Supplementary Information. To capture individual and contextual sources of variation, the model included four theoretically grounded predictors reflecting distinct dimensions of social cognition. Trust in automation captured the affective dimension, indicating emotional confidence and perceived safety; minority status served as a social-identity variable reflecting positional sensitivity to justice; and region accounted for cultural and infrastructural variation across cities. Together, these covariates disentangled the relative roles of emotional, social, and contextual factors linking diversity attention to justice perceptions. We next examined how strongly justice orientations were associated with these factors and with visual attention itself. Justice orientations showed only weak links with visual attention. Diversity Attention Index values from AV and non-AV scenarios exhibited non-credible associations with both normative and perceived justice, indicating that variation in attentional engagement with social cues did not substantially predict justice evaluations. However, the results show that justice perceptions in AI-based mobility are rooted primarily in affective and psychological dimensions – measured here by trust in automation and perceived safety – rather than in visual or cognitive engagement with diversity. Participants with greater confidence in automated systems and stronger feelings of safety scored higher on justice perceptions (βtrust = 0.14, 95% CI [0.05, 0.22]; βsafety = 0.11, 95% CI [0.02, 0.19]), and these effects remained stable across model specifications. Other variables, including age, minority status, and AI awareness, showed no credible associations with justice outcomes. Posterior distributions (Figure 4b) further highlight this distinction. The densities for trust and safety are clearly shifted to the right of zero, confirming their positive effects, whereas those for visual attention and demographic predictors cluster symmetrically around zero. Justice perceptions in technologically mediated environments thus depend less on what people see and more on how secure and confident they feel within such systems. This supports recent findings that trust acts as a central mediator of ethical evaluation and social acceptance of AI, suggesting that justice in automated mobility is driven more by affective readiness and perceived control than by cognitive salience of diversity. City-level random intercepts (Figure 4c) reveal marked spatial variation. Respondents in cities such as Kolkata, Johannesburg, Manila, and Jakarta exhibit higher justice orientations, while those in Tokyo, Ankara, and Berlin score lower. These contrasts likely reflect socio-political and infrastructural conditions: in rapidly urbanising regions of the Global South, where technological experimentation coexists with inequality, justice emerges as a salient normative concern. Conversely, in cities with established regulatory systems, justice may be seen as an institutional rather than personal issue, reducing its affective salience. Taken together, these findings indicate that justice in AI-based mobility is less a direct outcome of perceptual engagement with diversity and more an expression of trust, safety, and contextual framing. Visual attention may create the cognitive conditions for recognising difference, yet perceived justice hinge on whether individuals see automated systems as predictable, transparent, and secure. The weak association between attention indices and justice thus reflects a broader decoupling between seeing and valuing : noticing diversity does not automatically lead to perceiving justice. As AI mediates mobility, the ethical landscape of cities becomes increasingly shaped by affective infrastructures – trust, risk perception, and psychological safety – rather than by direct social perception. Justice in the automated city, in other words, depends as much on how people feel within AI systems as on what or whom they notice within them. Discussion This study examined how people perceive and evaluate social diversity and justice in AI-based urban mobility. Combining large-scale online eye-tracking with attitudinal survey data from 22 cities, it approached justice not as an abstract ideal but as a perceptual and affective process embedded in everyday encounters with automation. The findings show that hierarchies of visibility and recognition persist in the automated city, revealing that technologies may carry forward existing social imaginaries rather than transform them. Patterns of visual attention to diversity were structured yet unequal. Participants consistently prioritised traditionally established social categories such as nationality, residence, and income, while paying less attention to people with reduced mobility or micromobility users. These results echo prior studies showing how inequalities of visibility underpin urban and mobility justice 33 29 28 . Automation reduced total attention to social diversity but did not alter its underlying structure, suggesting that perceptions of legitimacy and belonging are reproduced as automation advances. Building on this structural stability of diversity attention, further patterns show that not all groups attend to social cues in the same way. Patterns of reduced attentional engagement among minority participants and AI-aware respondents suggest that perceptual attention to social diversity is shaped by both social position and technological familiarity. Lower Diversity Attention Index scores among minority groups may reflect differentiated experiences of visibility and exclusion, although such mechanisms cannot be directly inferred. Technologically experienced participants showed a similar reduction in diversity attention, likely because stronger assumptions about how automated or rule-based systems operate narrow the cues they treat as relevant. By contrast, less AI-aware participants engaged more broadly across social groups. Together, these patterns indicate that technological literacy does not necessarily enhance sensitivity to social diversity. Individual and contextual variation further revealed how perception intertwines with culture and cognition. Middle-aged participants showed stronger diversity attention, as did respondents from cities in the Europe. These findings are consistent with research linking age, empathy, and technological literacy to individuals’ capacity to understand situations from others’ perspectives 40 38 . Regional contrasts highlight that perceptual inclusion is culturally situated: diversity tends to fade in contexts where inequality and rapid technological change intersect, and becomes more salient where it is institutionalised or politically neutralised 6 3 . Justice orientations showed a similar contextual pattern, with higher values emerging in settings marked by social inequality and rapid change, and lower values in places where justice is viewed more as an institutional than a personal concern. This aligns with research showing that global AI imaginaries differ across historical and political environments 59 13 . These contextual contrasts set the stage for another central insight: the asymmetry between seeing and valuing. Bayesian regression results showed that justice in AI-based mobility depends less on what people visually notice and more on how secure and confident they feel within automated systems. Trust in automation and perceived safety were the strongest predictors of justice attitudes, while visual attention, age, and minority status had no credible effects. This indicates that, although people recognise social differences, justice perceptions are grounded primarily in affective and psychological dimensions rather than in cognitive or perceptual engagement with diversity. The lack of alignment between justice evaluations and visual attention suggests that these operate as distinct cognitive and affective processes, particularly in complex AI-mediated settings. Our interpretation of this decoupling is that the mere presence of justice-oriented mechanisms within AI or smart-city systems is insufficient: justice must also be perceived as being enacted. This positions data justice as a pragmatic, community-level, and continually evolving notion that depends on both institutional design and lived experience. These findings also extend classical social-psychological theories of intergroup perception into a domain where they have rarely been examined: technologically mediated environments. Tajfel’s social identity theory 39 and Turner’s self-categorisation model 39 , and Allport’s social categorisation theory 40 research has shown that individuals allocate attention, empathy, and moral concern unevenly across social categories, reinforcing symbolic boundaries between in-groups and out-groups. Our results suggest that similar perceptual asymmetries persist when interactions are technologically mediated. Automation in mobility does not eliminate the underlying cognitive processes through which people differentiate between social groups ; rather, it reconfigures it through the lens of machine environments. This demonstrates that mechanisms traditionally studied in face-to-face contexts also shape human–AI encounters, where patterns of recognition and misrecognition are increasingly co-produced by both human cognition and technological design. Together, these findings extend debates on data justice 31 30 32 by showing that justice in AI-driven mobility is co-constituted by perception and emotion. Visual attention determines who becomes visible in the automated city, while trust and safety determine whether that visibility is interpreted as fair. Justice thus operates across cognitive and relational layers, linking visual framing 21 with affective infrastructures of trust. This dual structure suggests that algorithmic fairness depends not only on technical optimisation but on how technologies are sensed, trusted, and socially understood. From a governance perspective, ensuring justice requires attention to both perceptual and affective inclusion. Transparency and representational diversity are insufficient if people do not feel recognised within automated environments. Integrating perceptual measures – such as gaze data – into ethical impact assessments may help identify hidden inequalities in how AI systems are seen and trusted. This study moved beyond static definitions of fairness and diversity to examine how they are perceived, felt, and visually negotiated in encounters with AI-based mobility. It did so by combining Taylor’s framework of data justice with insights from visual framing and Vertovec’s notion of super-diversity 47 . We thus highlighted the importance of approaching justice in AI-based urban mobility not merely as a design or governance issue but as a culturally shaped perceptual experience. However, several limitations of this study need to be considered. First, webcam-based eye-tracking enables large-scale, cross-national data collection but offers lower spatial precision than laboratory systems, limiting the capture of fine-grained gaze dynamics 60 . Second, minority status and other sociodemographic indicators relied on self-identification, which varies cross-culturally and does not reflect more complex or intersecting forms of marginalisation. Third, the scenarios were stylised depictions of mobility and cannot reproduce the full complexity of real urban environments. Fourth, justice perceptions were measured immediately after exposure, reflecting situational rather than longer-term evaluations. Finally, although Bayesian multilevel models accounted for uncertainty and cross-city variation, unobserved contextual factors may still influence justice orientations. Declarations Data availability The aggregated data generated and analysed during the current study are available in the Supplementary Data files associated with this article. Information about source data for Figures 2–4 are provided in the Data Availability file. Due to ethical restrictions and confidentiality requirements approved by the Tallinn University of Technology Research Ethics Committee (no. 1063), raw participant-level data cannot be made publicly available. Access to anonymised datasets can be granted by the corresponding author upon reasonable request and under a confidentiality agreement Acknowledgements We thank Edmond Awad for his valuable feedback on an earlier version of this manuscript. We also thank Hannabeth Hansen for her technical assistance in extracting and formatting the database used in this study. We are grateful to all participants for their time and contribution to the study. Funding declarations This work was supported by the European Commission through the H2020 project Finest Twins (grant No. 856602) and the ASTRA development programme of Tallinn University of Technology for 2016–2022 (2014-2020.4.01.16-0032). Consent to participate All participants provided informed consent before taking part in the study. Participation in the anonymous online survey was entirely voluntary, and respondents could discontinue at any time without consequence. Ethics approval The study received ethical approval from the Estonian National Institute for Health Development Research Ethics Committee (Approval No. 1063), which reviewed and approved the research protocol, including the consent procedure. Competing interests The authors declare that they have no competing interests. Author contributions AM led the overall research process, conducted the analyses, and wrote the manuscript draft, revising it based on contributions, revisions, and comments from the co-authors. MI coordinated and conducted the experiment together with AM and contributed to the methods section and the overall conception of the paper. AVP played a key role in the initial experimental design, provided feedback on the eye-tracking procedures and interpretation, and contributed to the interpretation of results and statistical verification. FB contributed to the theoretical framework, particularly the data justice components, and supported the development of the discussion. SB focused on validation analyses, comparison of eye-tracking indicators, selection of appropriate metrics, and provided feedback on justice concepts as well as other sections of the manuscript. All authors reviewed and approved the final version of the manuscript. References Kesselring, S. et al. Sustainable Mobilities in the Neighborhood: Methodological Innovation for Social Change. Sustainability 15 , 3583 (2023). OECD. Automated and Autonomous Driving. OECD https://www.oecd.org/en/publications/automated-and-autonomous-driving_5jlwvzdfk640-en.html (2015). OECD. Safer Roads with Automated Vehicles? OECD https://www.oecd.org/en/publications/safer-roads-with-automated-vehicles_b2881ccb-en.html (2018). Camille-L. AI Machine Learning and Regulation: The Case of Automated Vehicles. ITF https://www.itf-oecd.org/ai-machine-learning-and-regulation-case-automated-vehicles (2025). AI & Cities: Risks, Applications and Governance | UN-Habitat. https://unhabitat.org/ai-cities-risks-applications-and-governance. Global Assessment of Responsible AI in Cities – Research and recommendations to leverage AI for people-centred smart cities | UN-Habitat. https://unhabitat.org/global-assessment-of-responsible-ai-in-cities. Introducing the OECD AI Capability Indicators. OECD https://www.oecd.org/en/publications/introducing-the-oecd-ai-capability-indicators_be745f04-en.html (2025). Chang, H. & Behrendt, F. Riders driving at the limit of AI: geographies of two-wheeled food delivery and traffic safety in Seoul, South Korea. Urban Geography 46 , 1325–1343 (2025). Kitchin, R. The ethics of smart cities and urban science. Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences 374 , 20160115 (2016). Cardullo, P. The Right to the Smart City . (Emerald Publishing, Bingley, UK, 2019). Bucher, T. The algorithmic imaginary: exploring the ordinary affects of Facebook algorithms. Information, Communication & Society 1–15 (2016) doi:10.1080/1369118X.2016.1154086. Jasanoff, S., Kim, S.-H. & Sperling, S. Sociotechnical Imaginaries and Science and Technology Policy: A Cross-National Comparison. 21 (2004). Ruppert, E. Sociotechnical Imaginaries of Different Data Futures: An Experiment in Citizen Data . (2018). Masso, A., Männiste, M. & Calzati, S. The social imaginaries on governance through data: Q-methodological analysis of data professionals’ views in the field of mobility. Futures 152 , 103199 (2023). Alizadeh, T. & Prasad, D. The right to the smart city in the Global South: A research agenda. Urban Studies 61 , 426–444 (2024). Crawford, K. Atlas of AI: Power, Politics, and the Planetary Costs of Artificial Intelligence . (Yale University Press, New Haven, 2021). Sottile, E. et al. An innovative GPS smartphone based strategy for university mobility management: A case study at the University of RomaTre, Italy. Research in Transportation Economics 85 , (2021). Deitz, S., Lobben, A. & Alferez, A. Squeaky wheels: Missing data, disability, and power in the smart city. Big Data & Society 8 , 20539517211047735 (2021). Freudendal-Pedersen, M., Kesselring, S. & Servou, E. What is Smart for the Future City? Mobilities and Automation. Sustainability 11 , 221 (2019). Servou, E., Behrendt, F. & Horst, M. Data, AI and governance in MaaS – Leading to sustainable mobility? Transportation Research Interdisciplinary Perspectives 19 , 100806 (2023). Grabe, M. E. & Bucy, E. P. Image Bite Politics: News and the Visual Framing of Elections . (Oxford University Press, Oxford ; New York, 2009). Norton, P. Autonorama: The Illusory Promise of High-Tech Driving . (Island Press, 2021). Yakura, H. et al. Empirical evidence of Large Language Model’s influence on human spoken communication. Preprint at https://doi.org/10.48550/arXiv.2409.01754 (2025). Urry, J. Mobilities . (Polity Press, Cambridge ; Malden, 2008). Harvey, D. Rebel Cities: From the Right to the City to the Urban Revolution . (Verso Books, 2012). Lefebvre, H. The Production of Space . (Wiley, 1992). Vale, L. J. Just Urban Design: The Struggle for a Public City . (The MIT Press, Cambridge, Massachusetts, 2022). Verlinghieri, E. & Schwanen, T. Transport and mobility justice: Evolving discussions. J Transp Geogr 87 , 102798 (2020). Sheller, M. Mobility Justice: The Politics of Movement in an Age of Extremes . (Verso, London Brooklyn, 2018). Dencik, L., Hintz, A. & Cable, J. Towards data justice? The ambiguity of anti-surveillance resistance in political activism. Big Data & Society 3 , (2016). Taylor, L. What is data justice? The case for connecting digital rights and freedoms globally. Big Data & Society 4 , (2017). Heeks, R. & Renken, J. Data justice for development: What would it mean? Information Development 34 , 90–102 (2018). Behrendt, F. & Sheller, M. Mobility data justice. Mobilities 19 , 151–169 (2024). Lind, E. A. & Tyler, T. R. The Social Psychology of Procedural Justice . (Springer US, Boston, MA, 1988). doi:10.1007/978-1-4899-2115-4. Martin, A. & Taylor, L. Exclusion and inclusion in identification: regulation, displacement and data justice. Information Technology for Development 0 , 1–17 (2020). Young, I. M. & Allen, D. S. Justice and the Politics of Difference . (Princeton University Press, Princeton, New Jersey, 2011). Schlueter, E., Masso, A. & Davidov, E. What factors explain anti-Muslim prejudice? An assessment of the effects of Muslim population size, institutional characteristics and immigration-related media claims. Journal of Ethnic and Migration Studies 0 , 1–16 (2019). Pettigrew, T. F. Intergroup contact theory. Annual Review of Psychology 49 , 65 (1998). Tajfel, H. & Turner, J. The social identity theory of intergroup behaviour. in In S. Worchel & W. G. Austin (eds.). Psychology of Intergroup Relations 7–24 (Nelson-Hall, Chicago, IL., 1986). Gordon W. Allport. The Nature of Prejudice . (Addison-Wesley, Reading (Mass.), 1954). Stephan, W. G. & Stephan, C. W. Intergroup Threat Theory. in The International Encyclopedia of Intercultural Communication 1–12 (John Wiley & Sons, Ltd, 2017). doi:10.1002/9781118783665.ieicc0162. Eubanks, V. Automating Inequality: How High-Tech Tools Profile, Police, and Punish the Poor . (StMartin’s Press, New York, 2018). Lupton, D. Risk: Second Edition . (Routledge, London New York, 2013). Kitchin, R. The real-time city? Big data and smart urbanism. GeoJournal 79 , 1–14 (2014). Baibarac-Duignan, C. & de Lange, M. Controversing the datafied smart city: Conceptualising a ‘making-controversial’ approach to civic engagement. Big Data & Society 8 , 20539517211025557 (2021). Costanza-Chock, S. Design Justice: Community-Led Practices to Build the Worlds We Need . (The MIT Press, Cambridge, 2020). doi:10.7551/mitpress/12255.001.0001. Vertovec, S. Super-diversity and its implications. Ethnic and Racial Studies 30 , 1024–1054 (2007). Martin, R. AV futures or futures with AVs? Bridging sociotechnical imaginaries and a multi-level perspective of autonomous vehicle visualisations in praxis. Humanit Soc Sci Commun 8 , 68 (2021). Kastner, R. M. Moral Judgments and Visual Attention : An Eye-Tracking Investigation. https://www.semanticscholar.org/paper/Moral-Judgments-and-Visual-Attention-%3A-An-Kastner/214e28f1dfddcdcf1e8ecf0113379e4f1b8ac32f (2011). Cugurullo, F. & Acheampong, R. A. Fear of AI: an inquiry into the adoption of autonomous cars in spite of fear, and a theoretical framework for the study of artificial intelligence technology acceptance. AI & Soc 39 , 1569–1584 (2024). Awad, E. et al. The Moral Machine experiment. Nature 563 , 59–64 (2018). Etienne, H. The dark side of the ‘Moral Machine’ and the fallacy of computational ethical decision-making for autonomous vehicles. Law, Innovation and Technology 13 , 85–107 (2021). Formosa, P. & Ryan, M. Making Moral Machines: Why We Need Artificial Moral Agents. AI and Society 1–13 (2020) doi:10.1007/s00146-020-01089-6. Schramowski, P., Turan, C., Jentzsch, S., Rothkopf, C. & Kersting, K. The Moral Choice Machine. Frontiers in Artificial Intelligence 3 , (2020). Schuessler, D. The probability problems of the Moral Machine Experiment. AI Ethics 4 , 501–510 (2024). Crenshaw, K. Mapping the Margins: Intersectionality, Identity Politics, and Violence against Women of Color. Stanford Law Review 43 , 1241–1299 (1991). Kaun, A. & Masso, A. The Data Welfare State . (SAGE Publications, London, 2025). Järv, O., Masso, A., Silm, S. & Ahas, R. The Link Between Ethnic Segregation and Socio-Economic Status: An Activity Space Approach. Tijdschrift voor Economische en Sociale Geografie 112 , 319–335 (2021). Dreamscapes of Modernity: Sociotechnical Imaginaries and the Fabrication of Power . (University of Chicago Press, Chicago, IL, 2015). Ibrahimi, M., Masso, A. & Bellone, M. Sociotechnical imaginaries of autonomous vehicles: Comparing laboratory and online eye-tracking methods. Article accepted for publication in Plos ONE. (2025). Additional Declarations No competing interests reported. Supplementary Files 3AICitiesSupplementaryInformation21.11.25.docx Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-8195075","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":562273633,"identity":"731003f6-efdc-48a9-b073-47c6a81d04da","order_by":0,"name":"Anu Masso","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA3klEQVRIiWNgGAWjYFAD9gYGhgdA2oB4LTyHGRgSEkjSIpFMpBb+2T0GTDdq6uz5Jd8f/JD4w0benIH54AO8Zt85Y8Ccc+xw4szZycwSCQlphjsb2JLx23QjB6iF7UCCwe1kBqCWwwkGB3jMJPDpkAdr+Vdnb3DzMPOPhIT/QC3833/g02IA0pLbxsy44QYzG9CWAyBb2PC6y/BGWsHh3D6gX3qSzSwS0pINNxxmM8brMLkbyRsf53wDhhj7wcc3PtjYyRscb374Aa81QHAAlctMSP0oGAWjYBSMAoIAAPlfSTYjWUz4AAAAAElFTkSuQmCC","orcid":"","institution":"Tallinn University of Technology","correspondingAuthor":true,"prefix":"","firstName":"Anu","middleName":"","lastName":"Masso","suffix":""},{"id":562273634,"identity":"1c02c528-52d4-435d-859f-abed85799a5a","order_by":1,"name":"Mergime Ibrahimi","email":"","orcid":"","institution":"Tallinn University of Technology","correspondingAuthor":false,"prefix":"","firstName":"Mergime","middleName":"","lastName":"Ibrahimi","suffix":""},{"id":562273635,"identity":"65a2c9fc-c755-48aa-8e80-ab1fd0d56fdb","order_by":2,"name":"Ahti-Veikko Pietarinen","email":"","orcid":"","institution":"Hong Kong Babtist University","correspondingAuthor":false,"prefix":"","firstName":"Ahti-Veikko","middleName":"","lastName":"Pietarinen","suffix":""},{"id":562273636,"identity":"347c5b97-cac9-47a0-ba57-a565e895c05f","order_by":3,"name":"Frauke Behrendt","email":"","orcid":"","institution":"Eindhoven University of Technology","correspondingAuthor":false,"prefix":"","firstName":"Frauke","middleName":"","lastName":"Behrendt","suffix":""},{"id":562273637,"identity":"f7d6820f-845c-4abd-bdef-5d99c6402367","order_by":4,"name":"Srijoni Bhattarcherjee","email":"","orcid":"","institution":"Tallinn University of Technology","correspondingAuthor":false,"prefix":"","firstName":"Srijoni","middleName":"","lastName":"Bhattarcherjee","suffix":""}],"badges":[],"createdAt":"2025-11-24 15:39:51","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8195075/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8195075/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":98593374,"identity":"f3ee2ec3-4fdd-4ac3-bc95-9652b705d518","added_by":"auto","created_at":"2025-12-19 11:01:51","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":5185229,"visible":true,"origin":"","legend":"","description":"","filename":"2AICitiesMainText21.11.25.docx","url":"https://assets-eu.researchsquare.com/files/rs-8195075/v1/bef193a156a725b8b419c41e.docx"},{"id":98627901,"identity":"62167692-9e8b-45db-bd06-302af618eba4","added_by":"auto","created_at":"2025-12-19 17:10:46","extension":"json","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":7647,"visible":true,"origin":"","legend":"","description":"","filename":"591512809418413fa1a9ff690f6ea1ea.json","url":"https://assets-eu.researchsquare.com/files/rs-8195075/v1/fb24afa2094f0b359d227daa.json"},{"id":98628437,"identity":"d22134a7-10a6-4d98-8c0a-d8b16466bf28","added_by":"auto","created_at":"2025-12-19 17:11:33","extension":"docx","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":5178386,"visible":true,"origin":"","legend":"","description":"","filename":"3AICitiesSupplementaryInformation21.11.25.docx","url":"https://assets-eu.researchsquare.com/files/rs-8195075/v1/05730900214c823483d9f530.docx"},{"id":98593367,"identity":"93571efe-6cff-4ea5-9dda-f286785d8716","added_by":"auto","created_at":"2025-12-19 11:01:51","extension":"xml","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":122883,"visible":true,"origin":"","legend":"","description":"","filename":"591512809418413fa1a9ff690f6ea1ea1enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-8195075/v1/49baea6983f05a77e3fa7b85.xml"},{"id":98628388,"identity":"e0ca5232-23a1-45ba-9db9-0fc721f0cb7b","added_by":"auto","created_at":"2025-12-19 17:11:23","extension":"jpeg","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":845980,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage1.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8195075/v1/21cc4a5947e149f2631a90d5.jpeg"},{"id":98593371,"identity":"f3cbd7fa-b539-4a88-9ab6-9bd41ba02e3d","added_by":"auto","created_at":"2025-12-19 11:01:51","extension":"jpeg","order_by":5,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":382166,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage2.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8195075/v1/9f3c57ceaeebf1281ef93a2c.jpeg"},{"id":98593370,"identity":"185b19ee-2757-4309-bc4c-b9d11ba9e892","added_by":"auto","created_at":"2025-12-19 11:01:51","extension":"emf","order_by":6,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1839312,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage3.emf","url":"https://assets-eu.researchsquare.com/files/rs-8195075/v1/d88997fc956231c86ce7b6e8.emf"},{"id":98628037,"identity":"04612cc6-ea91-4928-8662-07b40e09310a","added_by":"auto","created_at":"2025-12-19 17:10:54","extension":"jpeg","order_by":7,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":365548,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage4.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8195075/v1/814fa17cfc2a0df70a0cb269.jpeg"},{"id":98628104,"identity":"70612dfe-725b-4faa-b2aa-dfc41e7611df","added_by":"auto","created_at":"2025-12-19 17:10:59","extension":"png","order_by":8,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":218795,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-8195075/v1/e159a02ec7d39a7887f9a773.png"},{"id":98628197,"identity":"0f53c2df-063f-4948-bc33-44e8a01c9a22","added_by":"auto","created_at":"2025-12-19 17:11:08","extension":"png","order_by":9,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":153265,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-8195075/v1/1520b7d3cd0475d593bc3437.png"},{"id":98593369,"identity":"c43cebbe-24ef-406d-be9a-d9f0a3b34255","added_by":"auto","created_at":"2025-12-19 11:01:51","extension":"png","order_by":10,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":18522,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-8195075/v1/357f7650b57f7171a71b3456.png"},{"id":98593378,"identity":"c7a6f364-8189-4ab1-a500-80d115f18eba","added_by":"auto","created_at":"2025-12-19 11:01:51","extension":"png","order_by":11,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":254689,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-8195075/v1/27ffb860fbbed802459dabd0.png"},{"id":98628208,"identity":"8a729b3e-9e85-4ccf-944d-7cc32b9d2ad2","added_by":"auto","created_at":"2025-12-19 17:11:09","extension":"xml","order_by":12,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":116669,"visible":true,"origin":"","legend":"","description":"","filename":"591512809418413fa1a9ff690f6ea1ea1structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-8195075/v1/c8f2a3c777773da65ae36d4a.xml"},{"id":98593380,"identity":"842b63a5-914e-4e99-99a7-06fcd36f41ef","added_by":"auto","created_at":"2025-12-19 11:01:51","extension":"html","order_by":13,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":135077,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-8195075/v1/92c163cea163d8bbccfa3809.html"},{"id":98593360,"identity":"d78ea9cc-d886-41dd-91b0-63754cb0d850","added_by":"auto","created_at":"2025-12-19 11:01:51","extension":"jpeg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":845980,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eExperimental design and operationalisation of intersectional diversity\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eNote:\u003c/strong\u003e The figure illustrates how intersectional diversity was operationalised in the experimental design, enabling analysis of how visual attention to social difference varies across AI-based and human-driven mobility contexts.\u003c/p\u003e","description":"","filename":"floatimage1.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8195075/v1/e98c1170cc504bb7e43c92a2.jpeg"},{"id":98628411,"identity":"c8395ef3-39dc-4b5e-942d-a2b26eb032f8","added_by":"auto","created_at":"2025-12-19 17:11:31","extension":"jpeg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":224804,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eDiversity attention across autonomous (AV) and non-autonomous (non-AV) scenes\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eNote: \u003c/strong\u003eMean fixation times (ms) for AV and non-AV conditions across seven diversity categories, shown with 95% confidence intervals (left panel). Values are based on paired-sample comparisons (N ≈ 1,250 per category); Δ values indicate the mean difference between AV and non-AV scenarios. Rightpanel shows the correlation between AV and non-AV diversity-attention indices (Spearman ρ = 0.83), indicating consistent attention patterns across contexts. Normality tests showed significant deviations (Shapiro–Wilk p \u0026lt; .001), though scatterplots suggested approximate linearity. See Supplementary Figure S1 for full correlation matrix.\u003c/p\u003e","description":"","filename":"floatimage2.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8195075/v1/5109189228c400e37fc304f7.jpeg"},{"id":98593363,"identity":"ffb6dd25-8545-436d-90b2-58e28e6d6230","added_by":"auto","created_at":"2025-12-19 11:01:51","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":69579,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eAV vs non-AV index means (± SD) across socio-demographic and regional subgroups\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eNote\u003c/strong\u003e: Points show subgroup means for AV (green) and non-AV (blue); horizontal lines indicate ±1 standard deviation. Error bars represent 95% confidence intervals. Variables include age (Low/Mid/High tertiles with ranges in labels), gender (Male/Female), income (Comfortable/Handle vs Not managing), education (Higher education: BA/MA/PhD vs Other), marital status (Married/Cohabiting vs Single/Widowed/Divorced), minority status (Yes/No), children (Yes/No), AI experiences (Low vs High awareness), and world region (UN-style groupings by country/city region). Missing and “Don’t know” categories were excluded. Full statistical results are provided in Supplementary Table S1 for transparency.\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-8195075/v1/b7fa3f336b4535ce1bed39a6.png"},{"id":98593365,"identity":"00a63eed-edb1-4869-8ac6-08a479c0150b","added_by":"auto","created_at":"2025-12-19 11:01:51","extension":"jpeg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":365548,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eBayesian model results linking diversity attention to perceived social justice\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003e\u003cstrong\u003eNotes\u003c/strong\u003e\u003c/em\u003e\u003cem\u003e.\u003c/em\u003e Panels show (a) fixed effects and posterior means, (b) posterior distributions of fixed effects, (c) and city-level random intercepts. Estimates are based on a hierarchical Bayesian model with individual- and city-level predictors. Variables represent: Trust in AV (confidence in autonomous systems), Region (city-level grouping), Minority (self-reported minority status), AI experience (familiarity with AI-based systems), AV and non-AV indices (visual attention measures in respective contexts).\u003c/p\u003e","description":"","filename":"floatimage4.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8195075/v1/928784606e6186b2c6a45781.jpeg"},{"id":104400645,"identity":"696e1572-3f4b-4ce0-aa84-de4ae3dedd2c","added_by":"auto","created_at":"2026-03-11 12:10:36","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2565895,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8195075/v1/29a4ec3a-fda3-49f7-97ba-c851f485fc92.pdf"},{"id":98593366,"identity":"8d7cf50d-b303-4b78-8a41-73bef1c133a3","added_by":"auto","created_at":"2025-12-19 11:01:51","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":5178386,"visible":true,"origin":"","legend":"","description":"","filename":"3AICitiesSupplementaryInformation21.11.25.docx","url":"https://assets-eu.researchsquare.com/files/rs-8195075/v1/759256b453ef9bb493f1a5e6.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Perceiving justice in the AI city: Global evidence from an eye-tracking study of autonomous mobility","fulltext":[{"header":"Introduction","content":"\u003cp\u003eArtificial intelligence (AI) is rapidly transforming urban governance, planning, and public service delivery. Positioned at the forefront of digital urbanism, AI is often framed as a key enabler of sustainability \u0026ndash; particularly in mobility \u003csup\u003e1\u003c/sup\u003e, where optimisation algorithms promise cleaner, more efficient, and adaptive transport systems. Autonomous vehicles (AVs) exemplify this imaginary: they are heralded as intelligent and inclusive alternatives capable of reducing congestion and emissions while expanding access. Yet beneath this sustainability narrative lies a persistent tension between technological promise and social reality. For example, AI-driven mobility systems risk reproducing, rather than remedying, inequalities in access, recognition, and representation \u003csup\u003e2\u003c/sup\u003e \u003csup\u003e3\u003c/sup\u003e \u003csup\u003e4\u003c/sup\u003e. Sustainable urban AI must therefore be reframed to include not only environmental efficiency but also social and cultural sustainability \u0026ndash; where inclusion becomes a foundational design principle, not an afterthought.\u003c/p\u003e\n\u003cp\u003eAt the international level, these concerns are increasingly recognised. Reports such as \u003cem\u003eUN-Habitat\u0026rsquo;s AI \u0026amp; Cities\u003c/em\u003e (2022) \u003csup\u003e5\u003c/sup\u003e and the \u003cem\u003eGlobal Assessment of Responsible AI in Cities\u003c/em\u003e (2024) \u003csup\u003e6\u003c/sup\u003e call for context-sensitive governance of AI in urban environments. They emphasise that the ethical and social implications of AI depend not only on technical design but also on how technologies are interpreted and received locally. The OECD\u0026rsquo;s \u003cem\u003eAI Capability Indicators\u003c/em\u003e (2025) \u003csup\u003e7\u003c/sup\u003e similarly underline the need to understand how AI aligns with human abilities and social expectations \u0026ndash; a perspective still missing from many urban applications.\u003c/p\u003e\n\u003cp\u003eAgainst this backdrop, extensive research has examined AI-based urbanism, from the broader promises of data-driven efficiency \u003csup\u003e7\u003c/sup\u003e to critiques of technocratic governance \u003csup\u003e8\u003c/sup\u003e. Critical smart city scholarship has shown that algorithmic systems often reinforce inequality, obscure accountability, and shape perceptions of safety and risk \u003csup\u003e9\u003c/sup\u003e \u003csup\u003e10\u003c/sup\u003e. In mobility, these dynamics are especially acute: weak governance can entrench socially and environmentally harmful trajectories, and AVs raise concerns around exclusion, bias, and accessibility. Yet while theoretical critiques are abundant and empirical work has emerged \u003csup\u003e11\u003c/sup\u003e \u003csup\u003e12\u003c/sup\u003e \u003csup\u003e13\u003c/sup\u003e \u003csup\u003e14\u003c/sup\u003e, studies that explore how people actually engage with such systems remain limited \u0026ndash; particularly those attentive to the perceptual and cultural diversity of these encounters in urban mobility.\u003c/p\u003e\n\u003cp\u003eYet despite AI operating within highly diverse social environments, we still know surprisingly little about how this diversity is perceived, negotiated, or transformed within AI-mediated mobility systems.\u0026nbsp;Existing debates on AI justice and inclusion tend to rely on abstract principles rather than lived experience, and they are dominated by Global North perspectives that treat AI as a neutral optimisation tool rather than a socially embedded system shaped by power \u003csup\u003e15\u003c/sup\u003e \u003csup\u003e16\u003c/sup\u003e. While research on \u0026lsquo;smart transport\u0026rsquo; has expanded \u003csup\u003e17\u003c/sup\u003e \u003csup\u003e18\u003c/sup\u003e \u003csup\u003e19\u003c/sup\u003e, the perceptual and experiential dimensions of how people encounter these systems remain largely overlooked, particularly across different cultural and socio-economic settings. This risks sidelining how people from diverse contexts perceive, trust, or resist algorithmic mobility systems.\u003c/p\u003e\n\u003cp\u003eThese dynamics resonate with findings from recent research on Mobility-as-a-Service (MaaS)\u0026nbsp;\u003csup\u003e20\u003c/sup\u003e, which shows how algorithmic governance can reinforce inequality when sustainability goals are weakly defined or externally imposed.\u0026nbsp;Yet such studies primarily focus on structural dynamics and overlook how these systems are interpreted, contested, or embodied by the people who navigate them. Moreover, cognitive and visual processes \u003csup\u003e21\u003c/sup\u003e \u003csup\u003e22\u003c/sup\u003e \u0026ndash; how individuals perceive and attend to AI systems in everyday settings \u0026ndash; remain largely absent from current debates on urban justice and governance. Recent evidence even suggests that AI technologies can reshape human perception and culture \u003csup\u003e23\u003c/sup\u003e, underscoring the need to examine how these systems are seen and experienced in urban contexts.\u003c/p\u003e\n\u003cp\u003eIn this study, we approach justice in AI-based urban mobility not merely as a design or governance issue but as a culturally shaped perceptual experience. We combine large-scale online eye-tracking with survey data to examine how people in 22 cities worldwide see and interpret social diversity in AI-based mobility. While surveys capture explicit attitudes, they reveal little about implicit, affective processes that shape how people relate to AI technologies long before forming conscious opinions \u003csup\u003e22\u003c/sup\u003e. \u003cem\u003eBy linking visual attention to justice perceptions, we explore how cultural and contextual factors shape the cognitive construction of inclusion and recognition.\u0026nbsp;\u003c/em\u003eWe use autonomous vehicles as a salient case within the broader landscape of AI-based urban systems. Specifically, we ask:\u003c/p\u003e\n\u003col\u003e\n \u003cli\u003eWhat patterns characterise visual attention to social diversity in autonomous versus human-driven mobility contexts?\u003c/li\u003e\n \u003cli\u003eWhich individual and contextual factors explain attention variation towards social diversity across AV and non-AV\u0026nbsp;(human-driven) settings?\u003c/li\u003e\n \u003cli\u003eHow is visual attention to diversity related to participants\u0026rsquo; perceptions of justice in AI-based mobility?\u003c/li\u003e\n\u003c/ol\u003e\n\u003cp\u003eThese questions reveal the aim at cognitive and cultural mechanisms through which AI-based mobility systems are accepted, contested, or mistrusted in everyday life. The findings contribute to designing urban AI systems that are not only technically sound but also socially legitimate and perceptually inclusive.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ePerceiving and framing justice in AI-based urban mobility\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eRecent debates on algorithmic urbanism emphasise that justice in data-driven governance extends beyond outcomes to the ways people experience and interpret algorithmic systems. Critical approaches to urban and mobility justice \u003csup\u003e24\u003c/sup\u003e \u003csup\u003e25\u003c/sup\u003e \u003csup\u003e26\u003c/sup\u003e \u003csup\u003e27\u003c/sup\u003e \u003csup\u003e28\u003c/sup\u003e \u003csup\u003e29\u003c/sup\u003e show that technologies are embedded in social and power hierarchies that shape whose needs and movements are prioritised in urban mobility. The concept of data justice expands this perspective by highlighting how data practices reproduce or challenge inequalities of visibility, participation, and control \u003csup\u003e30\u003c/sup\u003e \u003csup\u003e31\u003c/sup\u003e \u003csup\u003e32\u003c/sup\u003e .\u003c/p\u003e\n\u003cp\u003eIn the context of AI-driven mobility, where decisions depend on large-scale data infrastructures, justice concerns not only what algorithms decide but also how they configure who and what becomes knowable and governable. The more recent notion of mobility data justice \u003csup\u003e33\u003c/sup\u003e brings these issues into the transport domain, showing how algorithmic mobility systems distribute movement and agency unequally. Together, these debates underscore an important distinction between computational fairness and broader questions of justice. While fairness has often been operationalised within data-driven frameworks as a computational concern focused on bias detection and procedural control \u003csup\u003e30\u003c/sup\u003e \u003csup\u003e32\u003c/sup\u003e, justice extends beyond technical mitigation to encompass the structural, historical, and distributive dimensions of how data and technologies shape social relations \u0026nbsp;\u003csup\u003e31\u003c/sup\u003e. Dencik and colleagues caution that such technical framings risk detaching data practices from their social and political contexts \u003csup\u003e30\u003c/sup\u003e, while Heeks and Renken conceptualise fairness more narrowly as perceived process control in data handling \u003csup\u003e32\u003c/sup\u003e.\u003c/p\u003e\n\u003cp\u003eTaken together, these perspectives provide the foundation for analysing how justice is perceived and felt in AI-based mobility \u0026ndash; linking structural questions of data justive with people\u0026rsquo;s cognitive, affective, and evaluative experiences. This approach understands justice as a dynamic, value-based process grounded in recognition and inclusion \u0026nbsp;\u003csup\u003e31\u003c/sup\u003e. It highlights two complementary dimensions: normative justice, reflecting shared ideals about fairness, and experiential justice, referring to how these ideals are enacted and interpreted in concrete social contexts. Understanding how people move between these two dimensions \u0026ndash; between what they value and what they feel \u0026ndash; is essential for assessing the societal legitimacy of AI-based mobility.\u003c/p\u003e\n\u003cp\u003eSuch experiential judgements do not form in isolation; they are shaped by broader cultural and perceptual processes that influence how people interpret inclusion, recognition, and inequality in cities\u003csup\u003e34\u003c/sup\u003e \u003csup\u003e35\u003c/sup\u003e \u003csup\u003e36\u003c/sup\u003e. Public perceptions of justice depend not only on technical performance but also on whether individuals feel represented or marginalised within these systems\u0026nbsp;\u003csup\u003e34\u003c/sup\u003e \u003csup\u003e35\u003c/sup\u003e \u003csup\u003e36\u003c/sup\u003e. Cross-national and social-psychological research shows that responses to inequality and diversity are mediated by both individual predispositions and contextual cues\u0026nbsp;\u003csup\u003e37\u003c/sup\u003e, as described by intergroup contact theory\u0026nbsp;\u003csup\u003e38\u003c/sup\u003e, social identity theory\u0026nbsp;\u003csup\u003e39\u003c/sup\u003e, social categorisation theory\u0026nbsp;\u003csup\u003e40\u003c/sup\u003e,\u0026nbsp;and perceived threat theory\u0026nbsp;\u003csup\u003e41\u003c/sup\u003e. Yet these frameworks rarely address technologically mediated environments. As algorithmic systems increasingly structure access, recognition, and exposure in urban life\u0026nbsp; \u003csup\u003e42\u003c/sup\u003e \u003csup\u003e43\u003c/sup\u003e \u003csup\u003e44\u003c/sup\u003e \u003csup\u003e45\u003c/sup\u003e \u003csup\u003e46\u003c/sup\u003e, the perceptual mechanisms underlying justice evaluations may shift in ways not fully captured by existing theories \u0026ndash; particularly in the super-diverse contexts described by Vertovec\u0026nbsp;\u003csup\u003e47\u003c/sup\u003e. This highlights the need to examine how AI-based mobility systems shape whether, and for whom, justice is perceived as being realised in cities.\u003c/p\u003e\n\u003cp\u003eThis gap calls for integrating insights from perception and social cognition research into the study of AI justice. Perception is not merely a passive process but an active form of sense-making that filters social meaning through visual and affective cues \u003csup\u003e48\u003c/sup\u003e \u003csup\u003e49\u003c/sup\u003e. In AI-based mobility systems, such cues often precede conscious reasoning, shaping intuitive judgements of justice and legitimacy. AI-based systems may therefore modulate justice perceptions indirectly \u0026ndash; by altering what draws attention, what feels salient, and what remains unseen. Understanding these processes requires not only attitudinal data but methods that capture how attention itself is distributed across socially meaningful stimuli.\u003c/p\u003e\n\u003cp\u003eVisual framing plays a particularly important role in this context. As Grabe and Bucy \u003csup\u003e21\u003c/sup\u003e demonstrate in \u003cem\u003eImage Bite Politics\u003c/em\u003e, images profoundly shape how people cognitively and emotionally make sense of complex social phenomena. Building on this insight, we explore how visual attention and framing influence perceptions of justice and inclusion in AI-based mobility. This is especially relevant in the case of autonomous vehicles (AVs), where public engagement is largely mediated through speculative imagery, simulation, or symbolic representation rather than direct use \u003csup\u003e48\u003c/sup\u003e \u003csup\u003e50\u003c/sup\u003e. These visual imaginaries do not only depict technology but also encode assumptions about social order \u0026ndash; about who is seen as a passenger, who as a pedestrian, and whose safety is prioritised.\u0026nbsp;Visual cues \u0026ndash;\u003cem\u003e\u0026nbsp;\u003c/em\u003esuch as who is chosen to be depicted as passenger or pedestrian, or how risk and safety factors are distributed across the scenarios \u0026ndash; subtly influence how justice, legitimacy, and inclusion come to be perceived.\u0026nbsp;Representations of AVs often evoke imaginaries of control, harmony, or environmental efficiency, yet these framings may obscure underlying inequalities in accessibility and recognition. Studying how people visually process such scenes can thus reveal broader normative orientations toward justive and belonging.\u003c/p\u003e\n\u003cp\u003eBy combining framework of data justice with insights from visual framing and notion of super-diversity,\u0026nbsp;this study moves beyond static definitions of fairness and diversity to examine how they are perceived, felt, and visually negotiated in encounters with AI-based mobility.\u0026nbsp;Together, these processes form a theoretical bridge between data justice and perceptual justice \u0026ndash; linking structural power with individual sense-making. This highlights an important theoretical insight: for justice to be realised in AI-based systems, it is not sufficient that algorithms operate fairly in a technical sense; justice must also be \u003cem\u003eperceived\u003c/em\u003e as such by those affected. Understanding these links is crucial for evaluating not only the ethics but also the social intelligibility of AI systems \u0026ndash; how people come to \u0026lsquo;see\u0026rsquo; algorithmic mobility as fair or exclusionary, and how such perceptions might reinforce or resist existing urban inequalities.\u003c/p\u003e"},{"header":"Empirical design and methods","content":"\u003cp\u003eThis study applies a mixed-method design to examine how individuals perceive social diversity and justice in AI-based urban mobility. Combining large-scale online eye-tracking with attitudinal survey data, it investigates how people visually attend to, interpret, and evaluate social groups in autonomous (AV) and non-autonomous (non-AV) mobility scenarios. Rather than focusing on moral decision-making, the study explores the perceptual mechanisms through which social difference and justice are cognitively constructed in AI-mediated environments.\u003c/p\u003e\n\u003cp\u003eOur design builds upon, yet extends, the \u003cem\u003eMoral Machine\u003c/em\u003e experiment \u003csup\u003e51\u003c/sup\u003e, which explored public reasoning about ethical trade-offs in autonomous vehicle dilemmas. While \u003cem\u003eMoral Machine\u003c/em\u003e was instrumental in foregrounding public engagement with AI ethics, later scholarship has discussed both its methodological strengths and conceptual limits \u0026ndash; especially its focus on moral preferences rather than perceptual or contextual dimensions of human-AI interaction \u003csup\u003e52\u003c/sup\u003e \u003csup\u003e53\u003c/sup\u003e \u003csup\u003e54\u003c/sup\u003e \u003csup\u003e55\u003c/sup\u003e. Building on these insights, our study shifts the analytical focus from \u003cem\u003emoral choice\u003c/em\u003e to \u003cem\u003esocial perception\u003c/em\u003e, examining how individuals visually perceive, recognise, and interpret social difference in AI-based mobility environments.\u003c/p\u003e\n\u003cp\u003eWe employed a semi-experimental online design combining scenario-based eye-tracking with a structured survey. Figure 1 illustrates the operationalisation of intersectional diversity. Participants viewed scenarios in which either hypothetical autonomous vehicle passengers or pedestrians were prioritised and indicated their choice via mouse click while eye-tracking recorded gaze trajectories. Each scene featured characters differing across multiple diversity attributes \u0026ndash; race or ethnicity, gender, age, disability, income, vaccination status, micro-mobility use, and urban or rural residence \u0026ndash; modelled after intersectionality and super-diversity theory \u003csup\u003e56\u003c/sup\u003e \u003csup\u003e47\u003c/sup\u003e. Scenarios were generated through a Python-based randomisation script to ensure balance and clarity. All participants viewed 7 scenarios in AV condition, and the same 7 scenarios in non-AV condition. Participants were randomly assigned to begin with either AV or non-AV conditions.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eParticipants were recruited through the CINT crowdsourcing platform, producing a globally distributed sample of 1,272 valid participants from 22 cities across Europe, Asia, Africa, and the Americas. City selection reflected the degree of AI and AV experimentation, indexed using the \u003cem\u003eCities in Motion Index\u003c/em\u003e (CIMI) and publicly available data on pilot projects. Stratified quotas aligned participant distributions with national census data for gender, age, and education. Further demographic and contextual details are provided in Supplementary Informaiton Tables S1-S3.\u003c/p\u003e\n\u003cp\u003eEye-tracking was conducted using the RealEye platform, which employs AI-based webcam gaze prediction (accuracy \u0026asymp;110 pixels). Visual stimuli were designed to approximate hypothetical yet\u0026nbsp;realistic urban mobility situations while allowing systematic variation in social composition and vehicle type. All visual stimuli were created using the same resolution, brightness, and contrast settings to ensure visual consistency and minimise potential low-level visual confounds (e.g., differences in contrast or sharpness). Fixation data were aggregated into total fixation time per area of interest (AOI) and z-standardised within participants to ensure comparability across conditions.\u003c/p\u003e\n\u003cp\u003eAfter the eye-tracking task, participants completed a short survey capturing justice perceptions, social sensitivity, and attitudes toward AI. Drawing on the framework of data justice \u003csup\u003e31\u003c/sup\u003e, the survey measured perceived inclusion and recognition in AI-based mobility. Additional indicators included trust in AI, perceived safety, AI experience, gender, minority status, education, and region, \u003cstrong\u003ereflecting well-established predictors of public responses to automation and social diversity\u0026nbsp;\u003c/strong\u003e\u003csup\u003e57\u003c/sup\u003e. Ethical clearance was obtained prior to data collection, and all participants gave informed consent. Full details of the survey items, calibration, and validation procedures are presented in the Supplementary Information.\u003c/p\u003e\n\u003cp\u003eAnalytically, the study proceeded in three stages. \u003cem\u003eFirst, mean fixation times across diversity categories depicted in the visual scenarios were compared between AV and non-AV contexts using paired-samples t-tests to establish baseline variation in attention to attributes represented in the stimuli, such as the gender, age, disability status, or income level of the depicted individuals. Repeated-measures ANOVA was then used to analyse variation with participants\u0026rsquo; socio-demographic background variables.\u0026nbsp;\u003c/em\u003eSecond, due to strong correlations between category-level measures, we constructed a \u003cem\u003eDiversity Attention Index\u003c/em\u003e (DAI) aggregating fixation times across all categories. The DAI captured each participant\u0026rsquo;s overall attentiveness to social diversity cues and enabled comparisons between mobility contexts and subsequent modelling of individual and contextual variation. Third, we examined how DAI values related to participants\u0026rsquo; justice perceptions and background characteristics using Bayesian multilevel regression models. This approach allowed us to estimate uncertainty in fixed and random effects while accounting for both individual- and city-level variation. We focus on perceived justice as the primary outcome. A validation analysis using an alternative justice measure (normative justice) is included in the Supplementary Information.\u003c/p\u003e\n\u003cp\u003eTogether, these steps provide a multilevel understanding of how people attend to and evaluate social difference in AI-based mobility. Integrating perceptual and attitudinal data reveals not only \u003cem\u003ewhat\u003c/em\u003e individuals notice and \u003cem\u003ewhere\u003c/em\u003e, but \u003cem\u003ehow\u003c/em\u003e they interpret justice and inclusion within algorithmically mediated urban environments.\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003e\u003cem\u003ePatterns of visual attention to social diversity\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eThe first stage of analysis examined how participants visually attended to different social categories in autonomous (AV) and non-autonomous (non-AV) traffic scenarios. Clear and systematic differences emerged depending on both the social category and the transport technology (Figure 2). Nationality, residence, and income received the highest levels of attention across both conditions, indicating that these socially entrenched dimensions remain highly salient in people\u0026rsquo;s cognitive framing of urban mobility. Each relates to long-standing urban debates about segregation, belonging, and the perceived legitimacy of movement through public space.\u003c/p\u003e\n\u003cp\u003eIn contrast, people with reduced mobility received the least attention overall, suggesting that this form of diversity remains visually and cognitively marginal in representations of everyday mobility. Yet, the shorter fixation time may also indicate rapid, automatic normative judgement \u0026ndash; that people with reduced mobility should be protected without the need for further deliberation. The high standard deviation (57.1 in the case of AV and 56.8 in the case of non-AV) for this category indicates substantial individual variation \u0026ndash; some participants focused strongly on reduced mobility cues, while others almost ignored them. A comparable pattern appeared for micromobility, represented by an electric scooter user. Although scooters have become common in cities, their social meaning is still in flux. This instability was mirrored in gaze patterns, reflecting both curiosity and uncertainty about how new mobility forms fit within existing urban hierarchies.\u003c/p\u003e\n\u003cp\u003eIntermediate levels of attention were observed for age and COVID-19 status. Both categories were textually distinctive, but keeping possibly neutral visual presentation in the stimuli. Despite this visual salience, neither drew strong or consistent attention compared with more normatively charged categories such as nationality or income. This suggests that while participants recognised these differences, they did not interpret them as socially decisive cues in mobility contexts. Category-level variation was statistically confirmed by paired-samples T-test (all p \u0026lt; .001), with full pairwise comparisons presented in Supplementary Table S2.\u003c/p\u003e\n\u003cp\u003eStatistical comparisons (Figure S3 and Table S7 in Supplementary Information) showed that visual attention patterns differed significantly across social categories. In AV contexts, participants directed more attention toward female figures and pedestrians, and slightly more toward younger individuals, Europeans, White and Chinese figures, and those facing economic difficulties (all p \u0026lt; .01). Differences for other groups were minor. These findings indicate that automation amplifies selective perceptual engagement with social difference, particularly for categories associated with visibility, agency, or vulnerability in the traffic scene.\u003c/p\u003e\n\u003cp\u003eBeyond category-level variation, a clear technological effect emerged. Across all dimensions, the shift from non-AV to AV contexts was accompanied by an overall increase in visual attention to social diversity. Mean fixation times increased by approximately 12\u0026ndash;16 ms. While the absolute magnitude may seem modest, such shifts are meaningful in eye-tracking research given that the scenarios were tightly matched. The difference therefore reflects a systematic increase in attentional engagement with social cues once automation entered the scene, as also indicated by the effect sizes of 8.32 to 10.66 across categories (t-tests). The increase was most pronounced for people with reduced mobility, but also for micromobility category \u0026ndash;\u0026nbsp;categories that are socially newer or less institutionally represented \u0026ndash; suggesting that automation may lessen existing gaps in visibility. At the same time, variation across categories increased in the AV condition, indicating wider individual differences in how participants noticed or ignored social diversity under automation. This variability likely reflects the absence of stable perceptual norms in technologically mediated mobility settings.\u003c/p\u003e\n\u003cp\u003eAlthough total attention increased, the relative hierarchy of categories remained remarkably stable. Participants continued to allocate most attention to nationality, residence, and income, implying that entrenched perceptual hierarchies persist even as the technological mediation of mobility changes. In other words, increases compresses the overall level of attention but does not alter who is seen. People carry forward habitual ways of perceiving urban difference \u0026ndash; who stands out, who fades into the background \u0026ndash; even as the material infrastructures of mobility evolve.\u003c/p\u003e\n\u003cp\u003eTaken together, these findings show that visual attention to social diversity is patterned yet context-sensitive. Automation increases the total social information processed and amplifies the variability of perceptual engagement. Technological change therefore modulates rather than transforms existing hierarchies of attention. This highlights that justice in AI-based mobility depends not only on algorithmic design or governance but also on perceptual inclusion \u0026ndash; what, and whom, people actually see in the automated city. To further explore how these attentional patterns aggregate across individuals and contexts, the next section introduces the Diversity Attention Index (DAI) as a composite measure of perceptual engagement with social diversity.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eVariation across individual and contextual factors\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eWhile category-level comparisons revealed systematic variation in how participants attended to different forms of social diversity, the strong intercorrelations between categories indicated that these attentional tendencies were not independent. Individuals who devoted more attention to one category also tended to focus more on others. This coherence suggests an underlying cognitive disposition toward noticing, or overlooking, social difference in mobility contexts. To capture this general tendency, we constructed a composite \u003cem\u003eDiversity Attention Index\u003c/em\u003e (DAI) aggregating fixation times across diversity dimensions. Factor analysis supported a one-factor solution, confirming a common latent component underlying attention to diversity (see Supplementary Figure S1). The DAI provided an integrated measure of how people attend to social difference overall, allowing analysis of cross-contextual consistency between AV and non-AV scenarios as well as variation across individual and contextual factors.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eBuilding on this index, the second stage of analysis examined how DAI values varied across individual and contextual characteristics\u003c/strong\u003e (Figure 3, Supplementary Table S8). Attentional orientations to diversity were not evenly distributed but reflected systematic variation across minority status, AI awareness, regional background, age, and family composition. These results demonstrate that attention to diversity is shaped by both sociocultural positioning and experiential familiarity with technology, suggesting that perceptual engagement with social difference is a situated and learned capacity rather than a universal cognitive trait.\u003c/p\u003e\n\u003cp\u003eThe strongest and most consistent effect emerged for minority status. Participants who self-identified as ethnic or social minorities displayed lower Diversity Attention Index values than majority respondents in both AV and non-AV contexts. Participants with higher AI awareness showed a similar pattern, displaying lower Diversity Attention Index scores, particularly in non-automated contexts. Regional and sociodemographic factors together added further layers of variation in diversity attention. Respondents from Europe showed higher diversity attention index values compared with those from Asia, while participants from Africa and the Americas showed an intermediate position. These regional contrasts likely reflect distinct cultural framings of social visibility and inequality: in more stratified or rapidly transforming contexts, diversity may be less perceptible, whereas in stable or institutionalised settings it tends to be more visible, more normatively integrated into everyday life, or supported by the greater social and environmental diversity of urban contexts.\u003c/p\u003e\n\u003cp\u003eSimilar context-dependent effects appeared across age and family status. Middle-aged participants (45\u0026ndash;54) showed the highest attention to social differences, followed by the youngest and oldest groups, while participants aged 30\u0026ndash;44 exhibited the lowest overall attention. This pattern suggests generational differences in social engagement and adaptability, possibly reflecting distinct life-stage priorities and exposure to technological and social change. The lower diversity attention among participants aged 30\u0026ndash;44 may relate to their greater work and family responsibilities, leaving less cognitive or emotional capacity for engaging with social difference. This may also explain why participants with children showed lower diversity attention overall.\u003c/p\u003e\n\u003cp\u003eIncome had smaller yet consistent effects. Those facing economic difficulties displayed marginally higher diversity attention index scores, aligning with research \u003csup\u003e58\u003c/sup\u003e linking empathy and social perception to structural vulnerability. Associations with religiosity were also weaker, indicating that religious participants paid less attention to social diversity. This pattern may suggest that stronger religious identification is linked to a more inward or community-oriented focus, leaving less perceptual openness to broader social difference. Alternatively, it may reflect the influence of shared moral or cultural frameworks that reduce the salience of visible diversity in social perception.\u003c/p\u003e\n\u003cp\u003eTaken together, these findings show that attention to diversity is structured, selective, and context-dependent. Minority identity and technological familiarity both reduce perceptual sensitivity to social difference, yet they do so through distinct mechanisms \u0026ndash; one through withdrawal and invisibility, the other through normative selectivity and focus. In this sense, the justie of AI-based mobility depends not only on algorithmic transparency or representational balance but also on how automation reconfigures who is seen, who is overlooked, and whose difference becomes legible within data-driven urban environments.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eLinking attention to perceived justice indicators\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eThe final stage of analysis examined how visual attention to social diversity relates to perceptions of justice in AI-based mobility. A hierarchical Bayesian regression model was used to predict perceived justice from a set of perceptual, experiential, attitudinal, and demographic predictors (Figure 4). Posterior estimates were computed with weakly informative priors and are presented as mean effects with 95% credible intervals. This framework allowed reliable estimation of uncertainty and robust identification of associations across individual and city levels. The justice index was constructed from six correlated survey items capturing people\u0026rsquo;s lived experiences of justice, inclusion, and recognition in AI-based mobility; detailed variable composition, validation, and comparisons with the general justice value measure are provided in the Supplementary Information.\u003c/p\u003e\n\u003cp\u003eTo capture individual and contextual sources of variation, the model included four theoretically grounded predictors reflecting distinct dimensions of social cognition. Trust in automation captured the affective dimension, indicating emotional confidence and perceived safety; minority status served as a social-identity variable reflecting positional sensitivity to justice; and region accounted for cultural and infrastructural variation across cities. \u003cstrong\u003eTogether, these covariates disentangled the relative roles of emotional, social, and contextual factors linking diversity attention to justice perceptions.\u0026nbsp;\u003c/strong\u003e\u003cem\u003eWe next examined how strongly justice orientations were associated with these factors and with visual attention itself.\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eJustice orientations showed only weak links with visual attention. Diversity Attention Index values from AV and non-AV scenarios exhibited non-credible associations with both normative and perceived justice, indicating that variation in attentional engagement with social cues did not substantially predict justice evaluations. However, the results show that justice perceptions in AI-based mobility are rooted primarily in affective and psychological dimensions \u0026ndash; measured here by trust in automation and perceived safety \u0026ndash; rather than in visual or cognitive engagement with diversity. Participants with greater confidence in automated systems and stronger feelings of safety scored higher on justice perceptions (\u0026beta;trust = 0.14, 95% CI [0.05, 0.22]; \u0026beta;safety = 0.11, 95% CI [0.02, 0.19]), and these effects remained stable across model specifications. Other variables, including age, minority status, and AI awareness, showed no credible associations with justice outcomes.\u003c/p\u003e\n\u003cp\u003ePosterior distributions (Figure 4b) further highlight this distinction. The densities for trust and safety are clearly shifted to the right of zero, confirming their positive effects, whereas those for visual attention and demographic predictors cluster symmetrically around zero. Justice perceptions in technologically mediated environments thus depend less on what people \u003cem\u003esee\u003c/em\u003e and more on how secure and confident they \u003cem\u003efeel\u003c/em\u003e within such systems. This supports recent findings that trust acts as a central mediator of ethical evaluation and social acceptance of AI, suggesting that justice in automated mobility is driven more by affective readiness and perceived control than by cognitive salience of diversity.\u003c/p\u003e\n\u003cp\u003eCity-level random intercepts (Figure 4c) reveal marked spatial variation. Respondents in cities such as Kolkata, Johannesburg, Manila, and Jakarta exhibit higher justice orientations, while those in Tokyo, Ankara, and Berlin score lower. These contrasts likely reflect socio-political and infrastructural conditions: in rapidly urbanising regions of the Global South, where technological experimentation coexists with inequality, justice emerges as a salient normative concern. Conversely, in cities with established regulatory systems, justice may be seen as an institutional rather than personal issue, reducing its affective salience.\u003c/p\u003e\n\u003cp\u003eTaken together, these findings indicate that justice in AI-based mobility is less a direct outcome of perceptual engagement with diversity and more an expression of trust, safety, and contextual framing. Visual attention may create the cognitive conditions for recognising difference, yet perceived justice hinge on whether individuals see automated systems as predictable, transparent, and secure. The weak association between attention indices and justice thus reflects a broader decoupling between \u003cem\u003eseeing\u003c/em\u003e and \u003cem\u003evaluing\u003c/em\u003e: noticing diversity does not automatically lead to perceiving justice. As AI mediates mobility, the ethical landscape of cities becomes increasingly shaped by affective infrastructures \u0026ndash; trust, risk perception, and psychological safety \u0026ndash; rather than by direct social perception. Justice in the automated city, in other words, depends as much on how people \u003cem\u003efeel\u003c/em\u003e within AI systems as on what or whom they \u003cem\u003enotice\u003c/em\u003e within them.\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eThis study examined how people perceive and evaluate social diversity and justice in AI-based urban mobility. Combining large-scale online eye-tracking with attitudinal survey data from 22 cities, it approached justice not as an abstract ideal but as a perceptual and affective process embedded in everyday encounters with automation. The findings show that hierarchies of visibility and recognition persist in the automated city, revealing that technologies may carry forward existing social imaginaries rather than transform them.\u003c/p\u003e\n\u003cp\u003ePatterns of visual attention to diversity were structured yet unequal. Participants consistently prioritised traditionally established social categories such as nationality, residence, and income, while paying less attention to people with reduced mobility or micromobility users. These results echo prior studies showing how inequalities of visibility underpin urban and mobility justice \u003csup\u003e33\u003c/sup\u003e \u003csup\u003e29\u003c/sup\u003e \u003csup\u003e28\u003c/sup\u003e . Automation reduced total attention to social diversity but did not alter its underlying structure, suggesting that perceptions of legitimacy and belonging are reproduced as automation advances.\u003c/p\u003e\n\u003cp\u003eBuilding on this structural stability of diversity attention, further patterns show that not all groups attend to social cues in the same way. Patterns of reduced attentional engagement among minority participants and AI-aware respondents suggest that perceptual attention to social diversity is shaped by both social position and technological familiarity. Lower Diversity Attention Index scores among minority groups may reflect differentiated experiences of visibility and exclusion, although such mechanisms cannot be directly inferred. Technologically experienced participants showed a similar reduction in diversity attention, likely because stronger assumptions about how automated or rule-based systems operate narrow the cues they treat as relevant. By contrast, less AI-aware participants engaged more broadly across social groups. Together, these patterns indicate that technological literacy does not necessarily enhance sensitivity to social diversity.\u003c/p\u003e\n\u003cp\u003eIndividual and contextual variation further revealed how perception intertwines with culture and cognition. Middle-aged participants showed stronger diversity attention, as did respondents from cities in the Europe. \u003cstrong\u003eThese findings are consistent with research linking age, empathy, and technological literacy to individuals\u0026rsquo; capacity to understand situations from others\u0026rsquo; perspectives\u003c/strong\u003e \u003csup\u003e40\u003c/sup\u003e \u003csup\u003e38\u003c/sup\u003e . Regional contrasts highlight that perceptual inclusion is culturally situated: diversity tends to fade in contexts where inequality and rapid technological change intersect, and becomes more salient where it is institutionalised or politically neutralised \u003csup\u003e6\u003c/sup\u003e \u003csup\u003e3\u003c/sup\u003e. Justice orientations showed a similar contextual pattern, with higher values emerging in settings marked by social inequality and rapid change, and lower values in places where justice is viewed more as an institutional than a personal concern. This aligns with research showing that global AI imaginaries differ across historical and political environments \u003csup\u003e59\u003c/sup\u003e \u003csup\u003e13\u003c/sup\u003e.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eThese contextual contrasts set the stage for another central insight: the asymmetry between seeing and valuing.\u003c/strong\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003eBayesian regression results showed that justice in AI-based mobility depends less on what people visually notice and more on how secure and confident they feel within automated systems. Trust in automation and perceived safety were the strongest predictors of justice attitudes, while visual attention, age, and minority status had no credible effects. This indicates that, although people recognise social differences, justice perceptions are grounded primarily in affective and psychological dimensions rather than in cognitive or perceptual engagement with diversity. The lack of alignment between justice evaluations and visual attention suggests that these operate as distinct cognitive and affective processes, particularly in complex AI-mediated settings. Our interpretation of this decoupling is that the mere presence of justice-oriented mechanisms within AI or smart-city systems is insufficient: justice must also be perceived as being enacted. This positions data justice as a pragmatic, community-level, and continually evolving notion that depends on both institutional design and lived experience.\u003c/p\u003e\n\u003cp\u003eThese findings also extend classical social-psychological theories of intergroup perception into a domain where they have rarely been examined: technologically mediated environments. Tajfel\u0026rsquo;s social identity theory\u003csup\u003e39\u003c/sup\u003e and Turner\u0026rsquo;s self-categorisation model \u003csup\u003e39\u003c/sup\u003e, and Allport\u0026rsquo;s social categorisation theory \u003csup\u003e40\u003c/sup\u003e research has shown that individuals allocate attention, empathy, and moral concern unevenly across social categories, reinforcing symbolic boundaries between in-groups and out-groups. Our results suggest that similar perceptual asymmetries persist when interactions are technologically mediated. \u003cstrong\u003eAutomation in mobility does not eliminate the underlying cognitive processes through which people differentiate between social groups\u003c/strong\u003e; rather, it reconfigures it through the lens of machine environments. This demonstrates that mechanisms traditionally studied in face-to-face contexts also shape human\u0026ndash;AI encounters, where patterns of recognition and misrecognition are increasingly co-produced by both human cognition and technological design.\u003c/p\u003e\n\u003cp\u003eTogether, these findings extend debates on data justice\u0026nbsp;\u003csup\u003e31\u003c/sup\u003e \u003csup\u003e30\u003c/sup\u003e \u003csup\u003e32\u003c/sup\u003e by showing that justice in AI-driven mobility is co-constituted by perception and emotion. Visual attention determines who becomes visible in the automated city, while trust and safety determine whether that visibility is interpreted as fair. Justice thus operates across cognitive and relational layers, linking visual framing\u0026nbsp;\u003csup\u003e21\u003c/sup\u003e \u0026nbsp;with affective infrastructures of trust. This dual structure suggests that algorithmic fairness depends not only on technical optimisation but on how technologies are sensed, trusted, and socially understood.\u003c/p\u003e\n\u003cp\u003eFrom a governance perspective, ensuring justice requires attention to both perceptual and affective inclusion. Transparency and representational diversity are insufficient if people do not feel recognised within automated environments. Integrating perceptual measures \u0026ndash; such as gaze data \u0026ndash; into ethical impact assessments may help identify hidden inequalities in how AI systems are seen and trusted. This study moved beyond static definitions of fairness and diversity to examine how they are perceived, felt, and visually negotiated in encounters with AI-based mobility. It did so by combining Taylor\u0026rsquo;s framework of data justice with insights from visual framing and Vertovec\u0026rsquo;s notion of super-diversity \u003csup\u003e47\u003c/sup\u003e. We thus highlighted the importance of approaching justice in AI-based urban mobility not merely as a design or governance issue but as a culturally shaped perceptual experience.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eHowever, several limitations of this study need to be considered.\u0026nbsp;\u003c/strong\u003eFirst, webcam-based eye-tracking enables large-scale, cross-national data collection but offers lower spatial precision than laboratory systems, limiting the capture of fine-grained gaze dynamics \u003csup\u003e60\u003c/sup\u003e. Second, minority status and other sociodemographic indicators relied on self-identification, which varies cross-culturally and does not reflect more complex or intersecting forms of marginalisation. Third, the scenarios were stylised depictions of mobility and cannot reproduce the full complexity of real urban environments. Fourth, justice perceptions were measured immediately after exposure, reflecting situational rather than longer-term evaluations. Finally, although Bayesian multilevel models accounted for uncertainty and cross-city variation, unobserved contextual factors may still influence justice orientations.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eData availability\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe aggregated data generated and analysed during the current study are available in the Supplementary Data files associated with this article. Information about source data for Figures 2\u0026ndash;4 are provided in the Data Availability file. Due to ethical restrictions and confidentiality requirements approved by the Tallinn University of Technology Research Ethics Committee (no. 1063), raw participant-level data cannot be made publicly available. Access to anonymised datasets can be granted by the corresponding author upon reasonable request and under a confidentiality agreement\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgements\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe thank Edmond Awad for his valuable feedback on an earlier version of this manuscript. We also thank Hannabeth Hansen for her technical assistance in extracting and formatting the database used in this study. We are grateful to all participants for their time and contribution to the study.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding declarations\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis work was supported by the European Commission through the H2020 project Finest Twins (grant No. 856602) and the ASTRA development programme of Tallinn University of Technology for 2016\u0026ndash;2022 (2014-2020.4.01.16-0032).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent to participate\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll participants provided informed consent before taking part in the study. Participation in the anonymous online survey was entirely voluntary, and respondents could discontinue at any time without consequence.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthics approval\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe study received ethical approval from the Estonian National Institute for Health Development Research Ethics Committee (Approval No. 1063), which reviewed and approved the research protocol, including the consent procedure.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare that they have no competing interests.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAM led the overall research process, conducted the analyses, and wrote the manuscript draft, revising it based on contributions, revisions, and comments from the co-authors. MI coordinated and conducted the experiment together with AM and contributed to the methods section and the overall conception of the paper. AVP played a key role in the initial experimental design, provided feedback on the eye-tracking procedures and interpretation, and contributed to the interpretation of results and statistical verification. FB contributed to the theoretical framework, particularly the data justice components, and supported the development of the discussion. SB focused on validation analyses, comparison of eye-tracking indicators, selection of appropriate metrics, and provided feedback on justice concepts as well as other sections of the manuscript. All authors reviewed and approved the final version of the manuscript.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eKesselring, S. \u003cem\u003eet al.\u003c/em\u003e Sustainable Mobilities in the Neighborhood: Methodological Innovation for Social Change. \u003cem\u003eSustainability\u003c/em\u003e \u003cstrong\u003e15\u003c/strong\u003e, 3583 (2023).\u003c/li\u003e\n\u003cli\u003eOECD. Automated and Autonomous Driving. \u003cem\u003eOECD\u003c/em\u003e https://www.oecd.org/en/publications/automated-and-autonomous-driving_5jlwvzdfk640-en.html (2015).\u003c/li\u003e\n\u003cli\u003eOECD. Safer Roads with Automated Vehicles? \u003cem\u003eOECD\u003c/em\u003e https://www.oecd.org/en/publications/safer-roads-with-automated-vehicles_b2881ccb-en.html (2018).\u003c/li\u003e\n\u003cli\u003eCamille-L. AI Machine Learning and Regulation: The Case of Automated Vehicles. \u003cem\u003eITF\u003c/em\u003e https://www.itf-oecd.org/ai-machine-learning-and-regulation-case-automated-vehicles (2025).\u003c/li\u003e\n\u003cli\u003eAI \u0026amp; Cities: Risks, Applications and Governance | UN-Habitat. https://unhabitat.org/ai-cities-risks-applications-and-governance.\u003c/li\u003e\n\u003cli\u003eGlobal Assessment of Responsible AI in Cities \u0026ndash; Research and recommendations to leverage AI for people-centred smart cities | UN-Habitat. https://unhabitat.org/global-assessment-of-responsible-ai-in-cities.\u003c/li\u003e\n\u003cli\u003eIntroducing the OECD AI Capability Indicators. \u003cem\u003eOECD\u003c/em\u003e https://www.oecd.org/en/publications/introducing-the-oecd-ai-capability-indicators_be745f04-en.html (2025).\u003c/li\u003e\n\u003cli\u003eChang, H. \u0026amp; Behrendt, F. Riders driving at the limit of AI: geographies of two-wheeled food delivery and traffic safety in Seoul, South Korea. \u003cem\u003eUrban Geography\u003c/em\u003e \u003cstrong\u003e46\u003c/strong\u003e, 1325\u0026ndash;1343 (2025).\u003c/li\u003e\n\u003cli\u003eKitchin, R. The ethics of smart cities and urban science. \u003cem\u003ePhilosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences\u003c/em\u003e \u003cstrong\u003e374\u003c/strong\u003e, 20160115 (2016).\u003c/li\u003e\n\u003cli\u003eCardullo, P. \u003cem\u003eThe Right to the Smart City\u003c/em\u003e. (Emerald Publishing, Bingley, UK, 2019).\u003c/li\u003e\n\u003cli\u003eBucher, T. The algorithmic imaginary: exploring the ordinary affects of Facebook algorithms. \u003cem\u003eInformation, Communication \u0026amp; Society\u003c/em\u003e 1\u0026ndash;15 (2016) doi:10.1080/1369118X.2016.1154086.\u003c/li\u003e\n\u003cli\u003eJasanoff, S., Kim, S.-H. \u0026amp; Sperling, S. Sociotechnical Imaginaries and Science and Technology Policy: A Cross-National Comparison. 21 (2004).\u003c/li\u003e\n\u003cli\u003eRuppert, E. \u003cem\u003eSociotechnical Imaginaries of Different Data Futures: An Experiment in Citizen Data\u003c/em\u003e. (2018).\u003c/li\u003e\n\u003cli\u003eMasso, A., M\u0026auml;nniste, M. \u0026amp; Calzati, S. The social imaginaries on governance through data: Q-methodological analysis of data professionals\u0026rsquo; views in the field of mobility. \u003cem\u003eFutures\u003c/em\u003e \u003cstrong\u003e152\u003c/strong\u003e, 103199 (2023).\u003c/li\u003e\n\u003cli\u003eAlizadeh, T. \u0026amp; Prasad, D. The right to the smart city in the Global South: A research agenda. \u003cem\u003eUrban Studies\u003c/em\u003e \u003cstrong\u003e61\u003c/strong\u003e, 426\u0026ndash;444 (2024).\u003c/li\u003e\n\u003cli\u003eCrawford, K. \u003cem\u003eAtlas of AI: Power, Politics, and the Planetary Costs of Artificial Intelligence\u003c/em\u003e. (Yale University Press, New Haven, 2021).\u003c/li\u003e\n\u003cli\u003eSottile, E. \u003cem\u003eet al.\u003c/em\u003e An innovative GPS smartphone based strategy for university mobility management: A case study at the University of RomaTre, Italy. \u003cem\u003eResearch in Transportation Economics\u003c/em\u003e \u003cstrong\u003e85\u003c/strong\u003e, (2021).\u003c/li\u003e\n\u003cli\u003eDeitz, S., Lobben, A. \u0026amp; Alferez, A. Squeaky wheels: Missing data, disability, and power in the smart city. \u003cem\u003eBig Data \u0026amp; Society\u003c/em\u003e \u003cstrong\u003e8\u003c/strong\u003e, 20539517211047735 (2021).\u003c/li\u003e\n\u003cli\u003eFreudendal-Pedersen, M., Kesselring, S. \u0026amp; Servou, E. What is Smart for the Future City? Mobilities and Automation. \u003cem\u003eSustainability\u003c/em\u003e \u003cstrong\u003e11\u003c/strong\u003e, 221 (2019).\u003c/li\u003e\n\u003cli\u003eServou, E., Behrendt, F. \u0026amp; Horst, M. Data, AI and governance in MaaS \u0026ndash; Leading to sustainable mobility? \u003cem\u003eTransportation Research Interdisciplinary Perspectives\u003c/em\u003e \u003cstrong\u003e19\u003c/strong\u003e, 100806 (2023).\u003c/li\u003e\n\u003cli\u003eGrabe, M. E. \u0026amp; Bucy, E. P. \u003cem\u003eImage Bite Politics: News and the Visual Framing of Elections\u003c/em\u003e. (Oxford University Press, Oxford ; New York, 2009).\u003c/li\u003e\n\u003cli\u003eNorton, P. \u003cem\u003eAutonorama: The Illusory Promise of High-Tech Driving\u003c/em\u003e. (Island Press, 2021).\u003c/li\u003e\n\u003cli\u003eYakura, H. \u003cem\u003eet al.\u003c/em\u003e Empirical evidence of Large Language Model\u0026rsquo;s influence on human spoken communication. Preprint at https://doi.org/10.48550/arXiv.2409.01754 (2025).\u003c/li\u003e\n\u003cli\u003eUrry, J. \u003cem\u003eMobilities\u003c/em\u003e. (Polity Press, Cambridge ; Malden, 2008).\u003c/li\u003e\n\u003cli\u003eHarvey, D. \u003cem\u003eRebel Cities: From the Right to the City to the Urban Revolution\u003c/em\u003e. (Verso Books, 2012).\u003c/li\u003e\n\u003cli\u003eLefebvre, H. \u003cem\u003eThe Production of Space\u003c/em\u003e. (Wiley, 1992).\u003c/li\u003e\n\u003cli\u003eVale, L. J. \u003cem\u003eJust Urban Design: The Struggle for a Public City\u003c/em\u003e. (The MIT Press, Cambridge, Massachusetts, 2022).\u003c/li\u003e\n\u003cli\u003eVerlinghieri, E. \u0026amp; Schwanen, T. Transport and mobility justice: Evolving discussions. \u003cem\u003eJ Transp Geogr\u003c/em\u003e \u003cstrong\u003e87\u003c/strong\u003e, 102798 (2020).\u003c/li\u003e\n\u003cli\u003eSheller, M. \u003cem\u003eMobility Justice: The Politics of Movement in an Age of Extremes\u003c/em\u003e. (Verso, London Brooklyn, 2018).\u003c/li\u003e\n\u003cli\u003eDencik, L., Hintz, A. \u0026amp; Cable, J. Towards data justice? The ambiguity of anti-surveillance resistance in political activism. \u003cem\u003eBig Data \u0026amp; Society\u003c/em\u003e \u003cstrong\u003e3\u003c/strong\u003e, (2016).\u003c/li\u003e\n\u003cli\u003eTaylor, L. What is data justice? The case for connecting digital rights and freedoms globally. \u003cem\u003eBig Data \u0026amp; Society\u003c/em\u003e \u003cstrong\u003e4\u003c/strong\u003e, (2017).\u003c/li\u003e\n\u003cli\u003eHeeks, R. \u0026amp; Renken, J. Data justice for development: What would it mean? \u003cem\u003eInformation Development\u003c/em\u003e \u003cstrong\u003e34\u003c/strong\u003e, 90\u0026ndash;102 (2018).\u003c/li\u003e\n\u003cli\u003eBehrendt, F. \u0026amp; Sheller, M. Mobility data justice. \u003cem\u003eMobilities\u003c/em\u003e \u003cstrong\u003e19\u003c/strong\u003e, 151\u0026ndash;169 (2024).\u003c/li\u003e\n\u003cli\u003eLind, E. A. \u0026amp; Tyler, T. R. \u003cem\u003eThe Social Psychology of Procedural Justice\u003c/em\u003e. (Springer US, Boston, MA, 1988). doi:10.1007/978-1-4899-2115-4.\u003c/li\u003e\n\u003cli\u003eMartin, A. \u0026amp; Taylor, L. Exclusion and inclusion in identification: regulation, displacement and data justice. \u003cem\u003eInformation Technology for Development\u003c/em\u003e \u003cstrong\u003e0\u003c/strong\u003e, 1\u0026ndash;17 (2020).\u003c/li\u003e\n\u003cli\u003eYoung, I. M. \u0026amp; Allen, D. S. \u003cem\u003eJustice and the Politics of Difference\u003c/em\u003e. (Princeton University Press, Princeton, New Jersey, 2011).\u003c/li\u003e\n\u003cli\u003eSchlueter, E., Masso, A. \u0026amp; Davidov, E. What factors explain anti-Muslim prejudice? An assessment of the effects of Muslim population size, institutional characteristics and immigration-related media claims. \u003cem\u003eJournal of Ethnic and Migration Studies\u003c/em\u003e \u003cstrong\u003e0\u003c/strong\u003e, 1\u0026ndash;16 (2019).\u003c/li\u003e\n\u003cli\u003ePettigrew, T. F. Intergroup contact theory. \u003cem\u003eAnnual Review of Psychology\u003c/em\u003e \u003cstrong\u003e49\u003c/strong\u003e, 65 (1998).\u003c/li\u003e\n\u003cli\u003eTajfel, H. \u0026amp; Turner, J. The social identity theory of intergroup behaviour. in \u003cem\u003eIn S. Worchel \u0026amp; W. G. Austin (eds.). Psychology of Intergroup Relations\u003c/em\u003e 7\u0026ndash;24 (Nelson-Hall, Chicago, IL., 1986).\u003c/li\u003e\n\u003cli\u003eGordon W. Allport. \u003cem\u003eThe Nature of Prejudice\u003c/em\u003e. (Addison-Wesley, Reading (Mass.), 1954).\u003c/li\u003e\n\u003cli\u003eStephan, W. G. \u0026amp; Stephan, C. W. Intergroup Threat Theory. in \u003cem\u003eThe International Encyclopedia of Intercultural Communication\u003c/em\u003e 1\u0026ndash;12 (John Wiley \u0026amp; Sons, Ltd, 2017). doi:10.1002/9781118783665.ieicc0162.\u003c/li\u003e\n\u003cli\u003eEubanks, V. \u003cem\u003eAutomating Inequality: How High-Tech Tools Profile, Police, and Punish the Poor\u003c/em\u003e. (StMartin\u0026rsquo;s Press, New York, 2018).\u003c/li\u003e\n\u003cli\u003eLupton, D. \u003cem\u003eRisk: Second Edition\u003c/em\u003e. (Routledge, London New York, 2013).\u003c/li\u003e\n\u003cli\u003eKitchin, R. The real-time city? Big data and smart urbanism. \u003cem\u003eGeoJournal\u003c/em\u003e \u003cstrong\u003e79\u003c/strong\u003e, 1\u0026ndash;14 (2014).\u003c/li\u003e\n\u003cli\u003eBaibarac-Duignan, C. \u0026amp; de Lange, M. Controversing the datafied smart city: Conceptualising a \u0026lsquo;making-controversial\u0026rsquo; approach to civic engagement. \u003cem\u003eBig Data \u0026amp; Society\u003c/em\u003e \u003cstrong\u003e8\u003c/strong\u003e, 20539517211025557 (2021).\u003c/li\u003e\n\u003cli\u003eCostanza-Chock, S. \u003cem\u003eDesign Justice: Community-Led Practices to Build the Worlds We Need\u003c/em\u003e. (The MIT Press, Cambridge, 2020). doi:10.7551/mitpress/12255.001.0001.\u003c/li\u003e\n\u003cli\u003eVertovec, S. Super-diversity and its implications. \u003cem\u003eEthnic and Racial Studies\u003c/em\u003e \u003cstrong\u003e30\u003c/strong\u003e, 1024\u0026ndash;1054 (2007).\u003c/li\u003e\n\u003cli\u003eMartin, R. AV futures or futures with AVs? Bridging sociotechnical imaginaries and a multi-level perspective of autonomous vehicle visualisations in praxis. \u003cem\u003eHumanit Soc Sci Commun\u003c/em\u003e \u003cstrong\u003e8\u003c/strong\u003e, 68 (2021).\u003c/li\u003e\n\u003cli\u003eKastner, R. M. Moral Judgments and Visual Attention : An Eye-Tracking Investigation. https://www.semanticscholar.org/paper/Moral-Judgments-and-Visual-Attention-%3A-An-Kastner/214e28f1dfddcdcf1e8ecf0113379e4f1b8ac32f (2011).\u003c/li\u003e\n\u003cli\u003eCugurullo, F. \u0026amp; Acheampong, R. A. Fear of AI: an inquiry into the adoption of autonomous cars in spite of fear, and a theoretical framework for the study of artificial intelligence technology acceptance. \u003cem\u003eAI \u0026amp; Soc\u003c/em\u003e \u003cstrong\u003e39\u003c/strong\u003e, 1569\u0026ndash;1584 (2024).\u003c/li\u003e\n\u003cli\u003eAwad, E. \u003cem\u003eet al.\u003c/em\u003e The Moral Machine experiment. \u003cem\u003eNature\u003c/em\u003e \u003cstrong\u003e563\u003c/strong\u003e, 59\u0026ndash;64 (2018).\u003c/li\u003e\n\u003cli\u003eEtienne, H. The dark side of the \u0026lsquo;Moral Machine\u0026rsquo; and the fallacy of computational ethical decision-making for autonomous vehicles. \u003cem\u003eLaw, Innovation and Technology\u003c/em\u003e \u003cstrong\u003e13\u003c/strong\u003e, 85\u0026ndash;107 (2021).\u003c/li\u003e\n\u003cli\u003eFormosa, P. \u0026amp; Ryan, M. Making Moral Machines: Why We Need Artificial Moral Agents. \u003cem\u003eAI and Society\u003c/em\u003e 1\u0026ndash;13 (2020) doi:10.1007/s00146-020-01089-6.\u003c/li\u003e\n\u003cli\u003eSchramowski, P., Turan, C., Jentzsch, S., Rothkopf, C. \u0026amp; Kersting, K. The Moral Choice Machine. \u003cem\u003eFrontiers in Artificial Intelligence\u003c/em\u003e \u003cstrong\u003e3\u003c/strong\u003e, (2020).\u003c/li\u003e\n\u003cli\u003eSchuessler, D. The probability problems of the Moral Machine Experiment. \u003cem\u003eAI Ethics\u003c/em\u003e \u003cstrong\u003e4\u003c/strong\u003e, 501\u0026ndash;510 (2024).\u003c/li\u003e\n\u003cli\u003eCrenshaw, K. Mapping the Margins: Intersectionality, Identity Politics, and Violence against Women of Color. \u003cem\u003eStanford Law Review\u003c/em\u003e \u003cstrong\u003e43\u003c/strong\u003e, 1241\u0026ndash;1299 (1991).\u003c/li\u003e\n\u003cli\u003eKaun, A. \u0026amp; Masso, A. \u003cem\u003eThe Data Welfare State\u003c/em\u003e. (SAGE Publications, London, 2025).\u003c/li\u003e\n\u003cli\u003eJ\u0026auml;rv, O., Masso, A., Silm, S. \u0026amp; Ahas, R. The Link Between Ethnic Segregation and Socio-Economic Status: An Activity Space Approach. \u003cem\u003eTijdschrift voor Economische en Sociale Geografie\u003c/em\u003e \u003cstrong\u003e112\u003c/strong\u003e, 319\u0026ndash;335 (2021).\u003c/li\u003e\n\u003cli\u003e\u003cem\u003eDreamscapes of Modernity: Sociotechnical Imaginaries and the Fabrication of Power\u003c/em\u003e. (University of Chicago Press, Chicago, IL, 2015).\u003c/li\u003e\n\u003cli\u003eIbrahimi, M., Masso, A. \u0026amp; Bellone, M. Sociotechnical imaginaries of autonomous vehicles: Comparing laboratory and online eye-tracking methods. Article accepted for publication in Plos ONE. (2025).\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Artificial intelligence, urban mobility, autonomous vehicles, social justice, data justice, eye-tracking, cross-cultural variation","lastPublishedDoi":"10.21203/rs.3.rs-8195075/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8195075/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eAI increasingly shapes urban mobility, yet its social justice implications remain poorly understood, especially across cultural contexts. This study examines how people perceive justice in AI-based mobility systems through visual attention and cognitive framing. Combining large-scale online eye-tracking (n\u0026thinsp;=\u0026thinsp;1,272) with survey data from 22 cities worldwide, we analyse how individuals attend to and interpret social diversity across autonomous (AV) and human-driven contexts. Three key patterns emerge. First, visual attention to social categories such as nationality or income is structured and transferable across technological settings. Second, attentional variation reflects both individual (age, gender, AI experience) and contextual (regional, cultural) factors. Third, Bayesian modelling shows that justice perceptions depend less on visual engagement than on affective trust and perceived safety. Together, these findings reveal how automation reshapes cognitive-ethical orientations toward justice, illustrating that equitable AI mobility depends as much on perceptual inclusion and social recognition as on algorithmic fairness.\u003c/p\u003e","manuscriptTitle":"Perceiving justice in the AI city: Global evidence from an eye-tracking study of autonomous mobility","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-12-19 11:01:46","doi":"10.21203/rs.3.rs-8195075/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"4833daec-b584-49d1-9309-b52aafd3c83e","owner":[],"postedDate":"December 19th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":59872817,"name":"Biological sciences/Psychology"},{"id":59872818,"name":"Social science/Psychology"},{"id":59872819,"name":"Social science/Science technology and society"}],"tags":[],"updatedAt":"2026-03-03T00:09:05+00:00","versionOfRecord":[],"versionCreatedAt":"2025-12-19 11:01:46","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8195075","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8195075","identity":"rs-8195075","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}