Abstract
Intelligent cockpits are transforming from passive control interfaces into proactive collaborative partners in connected vehicles. While industry practice emphasizes high instrumentality, entertainment value, and anthropomorphic features, whether these design attributes reduce users' psychological burden and promote collaboration willingness in cognitively demanding driving tasks remains unclear. Drawing on Technology Affordance Theory, this study investigates how three perceived attributes (instrumentality, entertainment, anthropomorphism) are associated with human–machine collaboration willingness through cognitive load as mediating mechanism. Structural equation modeling analysis of 503 Chinese licensed drivers reveals divergent pathway effects. Perceived instrumentality and entertainment enhance collaboration willingness by significantly reducing cognitive load (partial mediation). However, perceived anthropomorphism exhibits a paradoxical pattern: it directly promotes collaboration willingness through affective channels while showing null effects on cognitive load reduction. This challenges the assumption that anthropomorphism equates to usability, revealing an affordance conflict wherein social cognitive processing demands offset interaction convenience in multitasking contexts. The findings establish boundary conditions for anthropomorphic design effectiveness and suggest context-adaptive strategies prioritizing cognitive resource optimization in safety–critical environments.
Keywords: Intelligent cockpit, Cognitive load, Human–machine collaboration willingness, Perceived anthropomorphism, Technology affordance theory
Introduction
Intelligent cockpits represent a revolutionary mode of human–machine interaction that has profoundly transformed driving experiences. While these systems are designed to enhance user experience through advanced functionalities, their rapid proliferation has exposed a critical design tension. Despite high adoption rates and user interest, significant usability challenges persist, including interaction inefficiency and feature complexity that may erode human–machine collaboration willingness—users' psychological inclination to establish collaborative relationships with intelligent systems and jointly accomplish tasks [1]. This tension is particularly salient in driving, a cognitively demanding task where drivers must simultaneously manage substantial information from the cockpit interface while maintaining primary attention on safety–critical driving operations [2]. The question thus emerges: Are intelligent cockpits genuine collaborative partners that reduce psychological burden, or do they impose additional cognitive costs that undermine collaboration?
Prior literature has provided starkly divergent answers. From the enhancement perspective, empirical research by Waytz et al. [3] found that endowing vehicle systems with anthropomorphic features (such as names or human-like voices) significantly strengthens drivers' trust in autonomous driving technology and alleviates psychological tension when confronting complex technologies. Song et al. [4] further confirmed that service robots with high instrumentality can significantly improve users' work efficiency and satisfaction. However, research from the burden perspective has revealed potential problems with intelligent systems. In cognitive load testing of in-vehicle human–machine interaction, Strayer et al. [5] found that even hands-free and eyes-free intelligent voice interaction requires drivers to allocate substantial cognitive resources, resulting in a cognitive tunneling effect that leads to the neglect of traffic information. Wang et al. [6] further indicated that excessively anthropomorphic robots trigger users' social comparison anxiety and job insecurity in high-pressure contexts. Will users genuinely reduce their willingness due to certain attributes of intelligent cockpits? This study aims to reconcile these contradictory findings in the literature.
Previous research has explored various driving factors influencing human–machine collaboration willingness, which can be categorized into two domains: factors related to technological systems and factors related to individual users. For instance, numerous studies have revealed that functional characteristics of systems, such as ease of use and usefulness, are critical to user adoption [7, 8]. Additionally, individual psychological traits such as technology anxiety and trust propensity have been identified as important predictors of collaboration willingness [9, 10]. However, human–machine collaboration willingness is a complex psychological outcome that cannot be accurately explained or predicted based solely on single-dimensional technological characteristics or individual factors. Technology Affordance Theory provides a powerful theoretical lens for understanding this complexity [11, 12]. This theory emphasizes that the action possibilities offered by technological objects to users are multidimensional, and different affordance dimensions may exert differentiated influences on users [13, 14]. In the intelligent cockpit context, technology affordances manifest as diversified perceived attributes: perceived instrumentality reflects the system's practical support for completing driving tasks [15, 16], perceived anthropomorphism embodies the possibility of socialized interaction provided by the system [17, 18], and perceived entertainment represents the system's capacity to create pleasurable experiences [8]. However, a more comprehensive perspective that simultaneously incorporates the multidimensional perceived attributes of intelligent cockpits to systematically reveal their differentiated influence mechanisms on users' collaboration willingness is currently lacking [19, 20]. Particularly in high cognitive load contexts such as driving, different perceived attributes may operate through distinct psychological pathways, which have not been adequately explored in existing literature [21]. Therefore, the second objective of this study is to investigate how human–machine collaboration willingness is triggered by specific perceived attributes of intelligent cockpits.
Although previous research has focused on the direct effects of driving factors on human–machine collaboration willingness, surprisingly, its underlying intrinsic mechanisms have received limited attention [22]. This may lead to two limitations: first, it may result in mixed or inconsistent outcomes,second, it remains unclear what psychological processes occur before users develop human–machine collaboration willingness. Regarding the former, mixed results may stem from the relative scarcity of research on mechanisms explaining why identified predictors relate to collaboration willingness. As Sweller [23] noted, investigating mediating variables that transmit the association between interactive media use and behavioral outcomes is crucial. Regarding the latter, knowledge about the psychological mechanisms underlying continuous usage behavior of intelligent systems is limited. The potentially complex experiences associated with intelligent cockpit use may consume people's cognitive resources and create severe psychological stress [24, 25]. Moreover, revealing these mechanisms is equally important for practitioners, as users' high-frequency usage is critical for automobile manufacturers and system developers. This study argues that complex experiences in intelligent cockpit interaction led to changes in cognitive load (i.e., the total amount of mental energy required to process information in working memory), which in turn affects human–machine collaboration willingness [26, 27]. Therefore, the third objective of this study is to explore how perceived attributes are linked to human–machine collaboration willingness through cognitive load.
To address the research gaps, this study aims to investigate the antecedents of human–machine collaboration willingness by revealing factors associated with the perceived attributes of intelligent cockpits. Secondly, it seeks to uncover the underlying mechanisms through which these perceived attributes influence collaboration willingness. Specifically, grounded in Technology Affordance Theory [13, 14] and Social Exchange Theory [28, 29], this study hypothesizes that three forms of perceived attributes (perceived instrumentality, perceived anthropomorphism, and perceived entertainment) have their unique influences on users' cognitive load experience (i.e., the influence of each attribute persists when controlling for the others), which subsequently affects human–machine collaboration willingness. This study makes several contributions to existing research. First, by identifying three specific perceived attribute antecedents leading to collaboration willingness, this study echoes previous literature regarding how technological characteristics may trigger users' psychological responses [4, 19], enriching the antecedent space of human–machine collaboration willingness by simultaneously integrating perceived instrumentality, perceived anthropomorphism, and perceived entertainment attributes. Second, by establishing the critical mediating role of users' cognitive load in the development of human–machine collaboration willingness, this study reveals the underlying psychological processes behind willingness formation, enriching the research literature on intelligent cockpits [30]. Finally, this study transcends existing human–computer interaction explanatory frameworks (such as the Technology Acceptance Model and the Unified Theory of Acceptance and Use of Technology) by adopting a cognitive psychology perspective to investigate how users leverage limited psychological resources to cope with intelligent systems, opening new avenues for researching intelligent cockpit usage experiences [31, 32].
Theoretical background
Human–machine collaboration willingness
Human–machine collaboration willingness refers to an individual's psychological inclination and behavioral intention to establish collaborative relationships with technological systems and jointly accomplish tasks [1]. In the context of intelligent connected vehicles, this manifests as users' willingness to engage in information sharing, task delegation, and continuous interaction with intelligent cockpits. As artificial intelligence technology reshapes human–machine relationships from a unidirectional tool-user model to a bidirectional partner-collaborator model [33], traditional technology usage intention is no longer sufficient to encompass this dynamic relationship. In contrast, collaboration willingness emphasizes three core characteristics: reciprocity, proactivity, and continuity [34], meaning that users not only passively receive services but also expect to establish reciprocal, trusting, and long-term goal-oriented relationships [35].
In intelligent cockpit scenarios, human–machine collaboration willingness possesses unique connotations. Le et al. [31] noted that the effectiveness of in-vehicle collaboration depends on whether drivers are willing to engage in control delegation, that is, trusting the system's autonomous judgment while maintaining supervision [3]. This willingness extends beyond the functional usage level to the relational level [36]. For instance, users are willing to provide their personal preferences to the system, allowing it to handle certain driving tasks and adopt recommendations based on connected data. Research indicates that when users perceive the system as a collaborative partner, they exhibit higher levels of engagement and creativity [4]. However, the high-risk and high cognitive load characteristics of driving tasks pose challenges to collaboration willingness [37]. Users must weigh system value against potential safety risks and cognitive costs when making decisions. If users perceive that collaboration threatens control authority or increases cognitive burden, their collaboration willingness will be suppressed even if the system is functionally powerful [6, 38]. Furthermore, the connectivity capabilities of intelligent cockpits extend the scope of collaboration to vehicle–road coordination and cloud-based ecosystems [39]. This requires users to develop new mental models, not only trusting in-vehicle systems but also accepting machines as proactive collaborators within intelligent transportation ecosystems [32].
Technology affordance theory
Technology Affordance Theory offers a foundational framework for examining how intelligent cockpit attributes influence user behavior. This theory originates from Gibson’s [11] ecological psychology, was introduced into the design field by Norman [12], and was systematized in information systems research by Leonardi [13]. Technology affordance is defined as the action possibilities that technological objects offer to users, supporting them in achieving specific goals [13]. Unlike traditional feature-based perspectives, this theory emphasizes relationality and contextuality, meaning that the same technological features can present differentiated affordances under different user capabilities or task contexts [40]. This perspective explains how complex systems collectively shape user perceptions and behaviors through multidimensional action possibilities [14]. To systematically examine how intelligent cockpit attributes influence driver behavior, this study synthesizes existing frameworks to categorize technology affordances into three core dimensions: functional, social, and hedonic [8, 15, 16]. These dimensions form a dynamic system of action possibilities shaping drivers' multi-faceted interaction roles: perceived instrumentality addresses the task performer role seeking efficiency,perceived anthropomorphism activates the social interactor role through conversational scripts; and perceived entertainment satisfies the experience seeker role through emotional regulation [21].
Beyond categorization, these affordances function as a cognitive regulation system operating along two pathways. Instrumental and hedonic affordances act as cognitive amplifiers, facilitating cognitive surplus by substituting manual tasks with automation or providing emotional buffers against stress. Conversely, anthropomorphic affordances can become cognitive constraints by introducing social processing demands—such as intention inference and conversational politeness management—that compete for limited mental resources [41, 42]. This integration clarifies that different affordances are not independent traits but interconnected mechanisms within a cognitive resource economy that either release or consume drivers’ psychological resources (see Table 1).
Table 1.
Dimensions of intelligent cockpit perceived attributes based on technology affordance theory
| Affordance type | Perceived attribute dimension | Theoretical connotation | Manifestation in intelligent cockpits | Corresponding user role | Connectivity enhancement |
|---|---|---|---|---|---|
| Functional affordance | Perceived instrumentality | The system provides practical tools and capability support needed to complete tasks, directly responding to users' core task requirements [4] | Through functions such as navigation, route planning, and driving assistance, the system provides practical value for users to safely and efficiently complete travel tasks | Task performer | Real-time vehicle–road coordination data significantly enhances decision support capabilities and traffic condition prediction accuracy |
| Social affordance | Perceived anthropomorphism | Based on the CASA paradigm [42], the system provides possibilities for socialized interaction through human-like features, transforming human–machine relationships | Through human-like features such as natural language interaction, emotion recognition, and personalized responses, the intelligent cockpit transforms from a tool into a virtual partner, providing emotional support and companionship | Social interactor | Cloud-based large models support more natural and fluent human–machine dialogue and contextualized personalized interaction |
| Hedonic affordance | Perceived entertainment | The system provides emotional value beyond functional requirements, satisfying users' hedonic needs through pleasurable experiences [8] | Through multimedia content, gamified design, and immersive interaction, the system provides positive emotional experiences for users, transforming driving from task execution into a pleasurable journey | Experience seeker | Massive cloud-based entertainment resources and intelligent recommendation algorithms provide personalized content services |
Social exchange theory
Social Exchange Theory [28, 43] provides a critical lens for understanding how drivers evaluate intelligent cockpit interactions through cost–benefit calculation based on reciprocity norms. In human–machine collaboration contexts, users maintain relationships when perceived benefits exceed perceived costs [44]. This study applies this framework to intelligent cockpits by conceptualizing perceived attributes as exchange rewards and cognitive load as exchange cost. The central theoretical proposition is that drivers perform an implicit cost–benefit analysis when deciding whether to engage with intelligent cockpit features. Perceived instrumentality, anthropomorphism, and entertainment represent potential rewards—efficiency gains, emotional connection, and pleasurable experiences respectively. However, these rewards must be weighed against cognitive load, the mental effort required to process cockpit interactions while managing primary driving tasks [25, 26]. Given the severe constraints on human working memory and attentional resources in safety–critical driving contexts [23], excessive cognitive costs can trigger a reciprocity imbalance that suppresses collaboration willingness [29].
This exchange logic generates testable predictions about the mediating role of cognitive load. When intelligent cockpits reduce psychological resource consumption through optimized design, users perceive high net-value services and reciprocate with increased collaboration willingness [20]. Conversely, when design features introduce costs that outweigh benefits—such as anthropomorphic elements that provide emotional appeal but impose social processing demands—users withdraw engagement to protect cognitive resources [45]. The stability of human–machine collaboration thus depends on maintaining a favorable cost–benefit ratio wherein the system demonstrates its capacity to deliver functional value while controlling cognitive costs [28].
Research hypotheses and model
Research hypotheses
Perceived attributes of intelligent cockpits and human–machine collaboration willingness
Based on Technology Affordance Theory, technological objects prompt subjects to interact with them to achieve specific goals by providing action possibilities [46]. In the context of intelligent connected vehicles, intelligent cockpits, as technological objects, directly influence users' collaboration willingness through their perceived attributes by supporting users in completing travel tasks. Perceived instrumentality reflects users' cognition of the practical value of intelligent cockpits in enhancing or supplementing driving capabilities [47]. In connected environments, intelligent cockpits significantly improve travel efficiency and safety through functions such as real-time traffic condition acquisition, automatic route planning, and driving assistance. Research indicates that high instrumentality of service robots can reduce technological barriers and increase collaboration possibilities [48]. When users perceive that the cockpit can effectively resolve congestion and optimize routes, they regard it as a reliable cooperative partner, thereby stimulating human–machine collaboration willingness. Accordingly, this study proposes:
H1a: Perceived instrumentality significantly and positively influences users' human–machine collaboration willingness.
Perceived anthropomorphism refers to the degree to which users perceive that intelligent cockpits possess human-like characteristics (such as natural voice and emotional understanding) [31]. Humans tend to develop positive evaluations of anthropomorphic objects and establish emotional connections with them, thereby promoting collaboration [1, 49]. In intelligent cockpits, anthropomorphic assistants can not only reduce the sense of technological distance through natural dialogue but also make users feel understood by perceiving user emotions (such as fatigue) and providing care [36]. The personalised services brought by connectivity further strengthen this emotional bond, prompting users to be more willing to collaborate with the system. Accordingly, this study proposes:
H1b: Perceived anthropomorphism significantly and positively influences users' human–machine collaboration willingness.
Perceived entertainment reflects the pleasurable feelings users experience during usage [34]. Intelligent cockpits not only provide rich multimedia content but also create immersive interactive experiences. Research indicates that perceived entertainment can stimulate positive emotions and reduce users' psychological barriers and effort expectations regarding technology use [4]. When the interaction process itself is pleasurable, users' psychological resistance to human–machine collaboration decreases, while their willingness to explore and fault tolerance increase accordingly [4]. Particularly with the support of intelligent recommendation algorithms, entertainment content aligned with preferences can create a positive driving and riding atmosphere, further enhancing users' human–machine collaboration willingness. Accordingly, this study proposes:
H1c: Perceived entertainment significantly and positively influences users' human–machine collaboration willingness.
Perceived attributes of intelligent cockpits and cognitive load
Cognitive load refers to the psychological resources required by individuals to complete tasks [2]. As driving is a high cognitive demand task, the perceived attributes of intelligent cockpits directly influence users' resource allocation and psychological burden. According to Social Exchange Theory, technological systems should provide value to users by reducing burdens [43]. Perceived instrumentality means the system can efficiently share tasks. Connected cockpits substitute for users' tedious information collection and decision-making processes by automatically processing massive amounts of data (such as environmental perception and route planning) [48]. High instrumentality not only reduces the mental effort required for decision-making but also lowers the cognitive costs associated with user supervision and error correction, due to the system's high reliability. When users are confident that the system can accurately execute tasks, their cognitive load will significantly decrease. Accordingly, this study hypothesizes:
H2a: Perceived instrumentality significantly and negatively influences users' cognitive load.
Perceived anthropomorphism significantly reduces users' cognitive load in dual task driving contexts by providing intuitive interaction paradigms that facilitate seamless navigation. First, anthropomorphic design (such as natural voice assistants) activates humans' deeply ingrained communication instincts, greatly lowering interaction barriers. Compared to traditional mechanical menu operations, natural conversational methods do not require users to memorize complex commands or perform cumbersome cognitive conversions, thereby reducing the mental effort required for information processing [50]. Second, in driving scenarios where visual resources are highly scarce, anthropomorphic interaction plays a critical role in sensory diversion. This non-visual channel interaction method avoids visual resource competition with traffic monitoring, enabling drivers to maintain attention on the environment and significantly alleviating the psychological burden arising from difficulties in attention allocation [51, 52]. Overall, the natural form of anthropomorphic interaction enables users to complete tasks with minimal cognitive cost, thereby effectively releasing psychological resources. Accordingly, this study proposes:
H2b: Perceived anthropomorphism significantly and negatively influences users' cognitive load.
High entertainment experiences can directly weaken users' subjective perception of mental effort. According to the Cognitive Absorption theory proposed by Agarwal and Karahanna [53], when users experience pleasure and enjoyment during usage, they enter a state of deep engagement. This state shields attention from task difficulty, making operations that originally require substantial psychological resources feel subjectively effortless, thereby directly reducing perceived cognitive load. Furthermore, the perceived entertainment attribute directly reduces the psychological costs of human–machine collaboration by providing emotional resources. Research by Song et al. [4] found that the entertainment value of technological systems can significantly reduce users' effort expectancy. In driving contexts, tedious or monotonous interactions consume drivers' willpower resources, whereas engaging and interesting cockpit interactions can continuously replenish emotional energy, preventing cognitive resource depletion and thereby maintaining lower cognitive load levels. In short, the higher the entertainment value of intelligent cockpits, the lower the mental effort and cognitive costs required for users to complete interaction tasks. Accordingly, this study proposes:
H2c: Perceived entertainment significantly and negatively influences users' cognitive load.
Cognitive load and human–machine collaboration willingness
In the high cognitive demand context of connected intelligent cockpit driving, cognitive load levels directly constrain users' collaboration willingness. First, according to Cognitive Resource Theory, individuals' psychological resources are limited. When users are in a high load state due to coping with complex traffic conditions, their working memory and cognitive resources approach depletion, making it difficult to bear the additional cognitive investment required for human–machine collaboration (such as understanding voice prompts and confirming system recommendations) [54, 55]. At this time, to ensure driving safety, users tend to adopt resource-conservation strategies and actively limit non-essential interactions with the cockpit. Second, from the perspective of attention competition, high cognitive load changes users' evaluation of system feedback. In extremely busy driving states, information provided by intelligent cockpits (even if assistive) is easily perceived by users as external interference rather than support, thereby triggering the discomfort of cognitive crowding [56]. To eliminate this interference, users instinctively refuse the system's collaboration requests and may even disable certain intelligent functions to maintain attention focus. Finally, high load impedes users' judgment of the value of collaboration. When the brain is in an overloaded state, users lack sufficient cognitive capacity to evaluate the long-term benefits of system recommendations and tend to focus more on immediate operational burdens and risks [32, 33]. This myopic effect undermines users' trust in and dependence on cockpit intelligent functions, further reducing their human–machine collaboration willingness. Accordingly, this study proposes:
H3: Users' cognitive load significantly and negatively influences their human–machine collaboration willingness.
The mediating role of cognitive load
Based on the reciprocity norm of Social Exchange Theory, perceived attributes, as resources provided by the system, stimulate users to reciprocate with collaboration willingness by improving users' psychological states (reducing cognitive load) [30]. Perceived instrumentality enhances collaboration willingness by reducing cognitive load. Instrumental functions liberate users from tedious information processing and driving operations, endowing users with cognitive surplus [20]. According to the Conservation of Resources Theory, this conservation of psychological resources constitutes high-value gains for users, thereby promoting their positive responses to the system [57, 58]. In the intelligent cockpit context, when the system assists users in sharing monitoring, judgment, and control tasks, users' cognitive burden decreases, thereby enhancing their acceptance of and trust in human–machine collaboration. That is, perceived instrumentality enhances collaboration willingness by alleviating users' cognitive burden. Accordingly, this study proposes:
H4a: Cognitive load mediates the relationship between perceived instrumentality and human–machine collaboration willingness.
Perceived anthropomorphism influences collaboration willingness by altering users' cognitive load levels. The primary goal of perceived anthropomorphic design (such as natural voice interaction) is to align with human instincts and intuitions, thereby alleviating users' mental burden. Research by Waytz et al. [3] demonstrates that in driving contexts, endowing vehicles with anthropomorphic features can reduce drivers' psychological stress, enabling them to transfer control to the system more easily. When users feel that interaction with the system is intuitive and effortless, their collaboration willingness increases accordingly. However, Goudey and Bonnin [59] point out that the impact of anthropomorphism on acceptance depends on whether it increases users' cognitive processing difficulty. If the degree of anthropomorphism does not match the system's capabilities, it may instead require users to invest more effort in understanding the system, leading to increased workload and a suppressed usage intention. Therefore, regardless of whether anthropomorphism specifically brings load reduction or load increase, changes in cognitive load constitute the key psychological mechanism determining whether users are willing to collaborate. Accordingly, this study proposes:
H4b: Cognitive load mediates the relationship between perceived anthropomorphism and human–machine collaboration willingness.
Perceived entertainment enhances collaboration willingness by reducing users' subjective perception of mental effort (i.e., reducing cognitive load). Unlike instrumentality, which directly substitutes for users in task execution, entertainment primarily serves a psychological buffering function. According to research by Agarwal and Karahanna [53], pleasurable interaction experiences induce a state of cognitive absorption, which can significantly reduce users' subjective perception of mental effort. In other words, driven by positive emotions, users feel less tired or burdened when processing interaction tasks of the same difficulty. This low cognitive load state, brought about by entertainment, essentially reduces the psychological costs of users' participation in human–machine collaboration. When users perceive that interacting with the system will not excessively consume their limited cognitive resources, their resistance to collaboration disappears, and they exhibit a higher willingness to maintain this relaxed and pleasant collaborative relationship. Accordingly, this study proposes:
H4c: Cognitive load mediates the relationship between perceived entertainment and human–machine collaboration willingness.
Research model
Based on the preceding reasoning, this study constructs a conceptual model, as shown in Fig. 1, aiming to elucidate the mechanisms by which perceived attributes of intelligent cockpits (perceived instrumentality, perceived anthropomorphism, perceived entertainment) influence human–machine collaboration willingness. The model proposes a dual-pathway hypothesis: perceived attributes not only directly and positively predict human–machine collaboration willingness but also exert mediating effects by reducing cognitive load.
Fig. 1.
The proposed research model
Research methodology
Pre-experience procedure
Participant recruitment
The research team employed a combined online and offline recruitment strategy, targeting potential users from diverse backgrounds through social media platforms and interpersonal networks. All 503 participants met the following inclusion criteria: aged 18–60 years and holding a valid driver's license. The sample exhibited balanced gender distribution (49.70% male, 50.30% female), with age predominantly 26–35 years (40.16%), followed by 18–25 years (31.81%) and 36–45 years (21.07%). Education levels were dominated by bachelor's/associate degrees (52.49%) and master's degrees or above (31.61%). Occupational distribution showed diversity: corporate employees (19.88%), students (19.09%), government/public institution personnel (17.50%), internet/IT/technology professionals (14.91%), and other occupations (28.62%). Monthly income concentrated in the 5,001–10,000 yuan range (30.42%), followed by 5,000 yuan and below (28.23%) and 10,001–20,000 yuan (27.44%). Driving experience was predominantly 1–3 years (43.74%), followed by 3–5 years (25.85%) and less than 1 year (21.47%).
To ensure participants had concrete understanding of the intelligent cockpit before evaluation, all participants completed the experiment within a specialized laboratory setting equipped with a high-fidelity driving simulator. The controlled pre-experience procedure consisted of three phases: participants first watched a 5-min instructional video detailing the cockpit's core intelligent functions (voice interaction, navigation, entertainment), then engaged in a 30-min free interaction session to experience the system's responsiveness and interface dynamics, and finally immediately completed the subjective measurement scales. This laboratory-based approach ensured that all responses were grounded in direct, contextualized interaction with the system, minimizing external environmental noise.
Testing materials and procedures
This study selected vehicle models equipped with Huawei HarmonyOS cockpit (HarmonySpace), such as AITO M7 and M9, as the testing platform. This system integrates voice interaction, intelligent navigation, and a multimedia ecosystem, possessing high technological representativeness and market coverage. Its unified operating system architecture helps control confounding variables introduced by brand differences. To ensure testing professionalism, the research team underwent two weeks of HarmonyOS system training during preparation, covering operational logic, distributed architecture, multimodal interaction, and vehicle-machine coordination principles. The testing environment adopted a standardized simulated driving cockpit equipped with high-fidelity driving simulators and data collection equipment, with professional consultation support from technical personnel to ensure procedural standardization.
The research process strictly adhered to academic ethical standards and data protection requirements. All participants signed informed consent forms, clearly informing them that their data would be used solely for academic research purposes, and all technical information was handled in accordance with confidentiality requirements. The testing process followed standardized procedures in a controlled environment, divided into four main steps as shown in Fig. 2. First, researchers introduced the testing purpose and procedures to participants in a quiet, appropriately lit indoor environment with the simulated cockpit prepared. Second, researchers demonstrated basic operations through instructional videos, including voice wake-up, navigation settings, and entertainment function invocation, ensuring participants mastered these operations before proceeding. Third, participants entered a 30-min free experience session, freely exploring various functions including voice assistant interaction, intelligent navigation planning, driving assistance features, and multimedia entertainment. Fourth, participants immediately completed a questionnaire measuring their evaluations of perceived instrumentality, perceived anthropomorphism, perceived entertainment, cognitive load, and human–machine collaboration willingness. This immediate measurement approach ensured the timeliness and authenticity of feedback data, maintaining the research design's independence and ensuring the objectivity and academic value of research conclusions.
Fig. 2.
Standardized process of pre-experience procedure based on Huawei HarmonyOS cockpit
Questionnaire survey
This study employed a 5-point Likert scale (1 = strongly disagree, 5 = strongly agree) to measure all research constructs. The scale design followed Churchill's procedure [60], conducting bidirectional English-Chinese translation and adjustment of mature scales. All scales were adapted from validated authoritative literature and appropriately adjusted according to the intelligent cockpit context. The questionnaire comprised three main sections. The first section provided introductory instructions, stating the survey purpose and respondent anonymity, and briefly introducing relevant concepts (such as intelligent cockpit, human–machine collaboration, etc.) to assist respondents' understanding. The second section was the core measurement section, primarily measuring perceived attributes of intelligent cockpits (perceived instrumentality, perceived anthropomorphism, perceived entertainment), cognitive load, and human–machine collaboration willingness. The third section collected respondents' demographic information, including gender, age, education, occupation, monthly income, and driving experience.
The Perceived Instrumentality (PI) scale was adapted from Leung et al. [48] and Song et al. [4], primarily measuring information richness, ease of use, and task support capability of intelligent cockpits [7, 61], comprising 5 items. Representative items include "Using the intelligent cockpit can improve my efficiency in completing travel tasks" (PI1), "Using the intelligent cockpit can help me complete driving tasks more accurately (e.g., navigation, route planning)" (PI2), "Using the intelligent cockpit can stably and reliably complete driving assistance tasks (e.g., accurately respond every time)" (PI3), "Using the intelligent cockpit can help me better cope with complex driving scenarios" (PI4), and "Using the intelligent cockpit can improve the overall quality of my driving task completion" (PI5).
The Perceived Anthropomorphism (PA) scale was adapted from Balakrishnan et al. [62] and Wang et al. [6], primarily measuring interaction authenticity and human-like characteristics [63], comprising 5 items. Representative items include "The voice interaction method of the intelligent cockpit is very similar to conversing with a real person" (PA1), "The intelligent cockpit can understand my intentions, just like a human" (PA2), "The responses of the intelligent cockpit make me feel like I am interacting with an emotional entity" (PA3), "The 'personality' exhibited by the intelligent cockpit in interaction makes me feel it is like an assistant" (PA4), and "The intelligent cockpit can perceive my emotional changes (e.g., fatigue, anxiety)" (PA5).
The Perceived Entertainment (PE) scale was derived from Moon and Kim [64] and Van der Heijden [8], primarily measuring imagination and emotional experience quality [65], comprising 4 items. Representative items include "Using the intelligent cockpit makes me feel pleasant when completing travel tasks" (PE1), "Using the intelligent cockpit makes me feel interested when completing travel tasks" (PE2), "Using the intelligent cockpit keeps me in a pleasant mood during travel" (PE3), and "The process of using the intelligent cockpit itself is pleasant" (PE4).
The Cognitive Load (CL) scale was adapted from Paas [27] and Naismith et al. [66], primarily measuring users' perception of mental effort and cognitive resource consumption during intelligent cockpit use, comprising 5 items. Representative items include "Using the intelligent cockpit requires me to invest high mental effort" (CL1), "Using the intelligent cockpit is cognitively effortful" (CL2), "Using the intelligent cockpit occupies a large amount of my thinking resources" (CL3), "Using the intelligent cockpit makes me feel cognitively burdened" (CL4), and "When operating the intelligent cockpit, I need to invest a large amount of attentional resources" (CL5).
The Human–Machine Collaboration Willingness (CW) scale was adapted from You and Robert [1] and Franç et al. [35], primarily measuring users' collaborative tendency toward intelligent cockpits and continuous usage intention [67, 68], comprising 6 items. Representative items include "I am willing to work with the intelligent cockpit as a team to jointly complete travel tasks" (CW1), "I am willing to share my personal information and preferences with the intelligent cockpit so that it can better assist me" (CW2), "I am willing to delegate the intelligent cockpit to autonomously handle some tasks (e.g., automatically plan routes based on connected traffic conditions)" (CW3), "I am willing to trust and adopt suggestions or reminders proactively provided by the intelligent cockpit based on connected data (e.g., congestion alerts ahead)" (CW4), "I am willing to communicate and interact with the intelligent cockpit (e.g., through voice) to jointly complete travel tasks" (CW5), and "Compared to operating completely independently, I am more willing to complete travel tasks with the assistance of the intelligent cockpit" (CW6).
Data analysis strategy
This study employed SPSS v28.0.1.1 and SmartPLS v4.0 for data analysis. First, SPSS v28.0.1.1 was used for data entry, cleaning, and descriptive statistical analysis, which reflected the overall distribution through sample characteristics, and for conducting normality testing and outlier processing. Second, Partial Least Squares Structural Equation Modeling (PLS-SEM) was adopted to test research hypotheses. The main reasons for selecting PLS-SEM are that this method has lower requirements for sample distribution, is suitable for path analysis of complex models, and can simultaneously handle measurement models and structural models. The analysis was divided into two phases: first, assessing the reliability (Cronbach's α and CR both greater than 0.7) and validity (convergent validity AVE greater than 0.5, discriminant validity tested through Fornell-Larcker criterion and HTMT less than 0.85) of the measurement model; second, testing research hypotheses among latent variables through path coefficient analysis and verifying the significance of mediating effects using the Bootstrap method (5000 resamples, 95% confidence interval). Additionally, the model's predictive relevance indicator Q2, effect size f2, and collinearity diagnostic VIF were reported to comprehensively evaluate model quality.
Results
Measurement model assessment
Reliability and validity analysis
This study employed Cronbach's alpha (α) and Composite Reliability (CR) to assess internal consistency, and used Average Variance Extracted (AVE) to assess convergent validity. As shown in Table 2, the α and CR values of all constructs exceeded the threshold of 0.7, indicating good reliability of the scales. Furthermore, all AVE values were greater than 0.5, confirming that latent variables explained more than half of the variance in their indicators, possessing adequate convergent validity.
Table 2.
Reliability and validity analysis
| Measurements | Path | Factor loading | T | p | α | CR | AVE |
|---|---|---|---|---|---|---|---|
|
Perceived instrumentality (PI) |
PI1 ← PI | 0.876 | 86.127 | 0.000 | 0.888 | 0.889 | 0.692 |
| PI2 ← PI | 0.821 | 50.289 | 0.000 | ||||
| PI3 ← PI | 0.813 | 47.956 | 0.000 | ||||
| PI4 ← PI | 0.825 | 52.569 | 0.000 | ||||
| PI5 ← PI | 0.822 | 48.134 | 0.000 | ||||
|
Perceived anthropomorphism (PA) |
PA1 ← PA | 0.808 | 46.024 | 0.000 | 0.864 | 0.864 | 0.647 |
| PA2 ← PA | 0.815 | 49.260 | 0.000 | ||||
| PA3 ← PA | 0.803 | 49.063 | 0.000 | ||||
| PA4 ← PA | 0.807 | 44.582 | 0.000 | ||||
| PA5 ← PA | 0.790 | 40.662 | 0.000 | ||||
|
Perceived entertainment (PE) |
PE1 ← PE | 0.845 | 56.936 | 0.000 | 0.849 | 0.852 | 0.688 |
| PE2 ← PE | 0.840 | 63.852 | 0.000 | ||||
| PE3 ← PE | 0.836 | 58.116 | 0.000 | ||||
| PE4 ← PE | 0.797 | 41.601 | 0.000 | ||||
|
Cognitive load (CL) |
CL1 ← CL | 0.821 | 56.689 | 0.000 | 0.876 | 0.876 | 0.668 |
| CL2 ← CL | 0.828 | 60.150 | 0.000 | ||||
| CL3 ← CL | 0.825 | 60.181 | 0.000 | ||||
| CL4 ← CL | 0.817 | 60.774 | 0.000 | ||||
| CL5 ← CL | 0.797 | 46.725 | 0.000 | ||||
|
Human–machine collaborative willingness (CW) |
CW1 ← CW | 0.786 | 47.369 | 0.000 | 0.883 | 0.883 | 0.632 |
| CW2 ← CW | 0.784 | 49.427 | 0.000 | ||||
| CW3 ← CW | 0.802 | 55.243 | 0.000 | ||||
| CW4 ← CW | 0.800 | 48.707 | 0.000 | ||||
| CW5 ← CW | 0.810 | 56.685 | 0.000 | ||||
| CW6 ← CW | 0.786 | 49.108 | 0.000 |
Discriminant validity
Discriminant validity was verified through two methods. First, according to the Fornell-Larcker criterion, the square root of AVE for each construct (diagonal values) was greater than its correlation coefficients with other constructs. Second, the Heterotrait-Monotrait Ratio (HTMT) between all variable pairs was below the strict threshold of 0.85. Both test results indicate good statistical distinctiveness among latent variables, as shown in Tables 3 and 4 below.
Table 3.
Discriminant validity
| PI | PA | PE | CL | CW | |
|---|---|---|---|---|---|
| PI | 0.832 | ||||
| PA | 0.334 | 0.804 | |||
| PE | 0.246 | 0.269 | 0.830 | ||
| CL | −0.520 | −0.277 | −0.547 | 0.818 | |
| CW | 0.524 | 0.405 | 0.468 | −0.645 | 0.795 |
Table 4.
HTMT discriminant validity
| PI | PA | PE | CL | CW | |
|---|---|---|---|---|---|
| PI | |||||
| PA | 0.381 | ||||
| PE | 0.282 | 0.314 | |||
| CL | 0.588 | 0.317 | 0.633 | ||
| CW | 0.591 | 0.463 | 0.539 | 0.732 |
Structural equation model assessment
Predictive relevance indicator Q2
The model demonstrated strong predictive capability, with Q2 values for endogenous variables CL (0.303) and CW (0.320) both exceeding the strong prediction standard of 0.26. Meanwhile, collinearity diagnostic results showed that the Variance Inflation Factor (VIF) for all measurement items and latent variables was below 5, indicating no collinearity issues in the model, as shown in Table 5 below.
Table 5.
Predictive relevance indicator Q2
| SSO | SSE | Q2 (= 1-SSE/SSO) | |
|---|---|---|---|
| CL | 2515 | 1753.843 | 0.303 |
| CW | 3018 | 2052.459 | 0.320 |
Collinearity diagnostic VIF
A diagnostic analysis of the model's collinearity was conducted, with the results presented in Tables 6 and 7. It can be concluded that the VIF between measurement variables is below 5, and the VIF between latent variable factors is also below 5, indicating that the model does not have collinearity issues.
Table 6.
Collinearity of measurement variable indicators
| VIF | |
|---|---|
| CL1 | 2.061 |
| CL2 | 2.078 |
| CL3 | 2.087 |
| CL4 | 1.961 |
| CL5 | 1.860 |
| CW1 | 1.888 |
| CW2 | 1.879 |
| CW3 | 1.980 |
| CW4 | 1.987 |
| CW5 | 2.055 |
| CW6 | 1.861 |
| PA1 | 1.969 |
| PA2 | 1.925 |
| PA3 | 1.928 |
| PA4 | 1.879 |
| PA5 | 1.760 |
| PE1 | 2.073 |
| PE2 | 1.942 |
| PE3 | 1.902 |
| PE4 | 1.750 |
| PI1 | 2.684 |
| PI2 | 2.077 |
| PI3 | 1.999 |
| PI4 | 2.107 |
| PI5 | 2.046 |
Table 7.
Collinearity of latent variable indicators
| VIF | |
|---|---|
| PI → CL | 1.160 |
| PI → CW | 1.460 |
| PA → CL | 1.175 |
| PA → CW | 1.176 |
| PE → CL | 1.111 |
| PE → CW | 1.471 |
| CL → CW | 1.845 |
Common method bias assessment
Given that all core variables were measured via a single questionnaire at one time point, we assessed the potential for common method bias (CMB) using Harman’s single-factor test [69]. Following established procedures, we conducted an unrotated exploratory factor analysis including all 25 measurement items in SPSS. The results, presented in Table 8, show that the first factor accounts for 36.893% of the total variance, which is substantially below the commonly accepted threshold of 50% [70]. Additionally, the first five factors collectively explain 66.777% of the variance, indicating that variance is distributed across multiple factors rather than concentrated in a single factor. These findings suggest that common method variance is unlikely to severely distort our structural model findings.
Table 8.
Total variance explained for common method bias test
| Component | Initial eigenvalues | Extraction sums of squared loadings | ||||
|---|---|---|---|---|---|---|
| Total | % of Variance | Cumulative % | Total | % of Variance | Cumulative % | |
| 1 | 9.223 | 36.893 | 36.893 | 9.223 | 36.893 | 36.893 |
| 2 | 2.596 | 10.385 | 47.278 | 2.596 | 10.385 | 47.278 |
| 3 | 2.334 | 9.336 | 56.614 | 2.334 | 9.336 | 56.614 |
| 4 | 1.433 | 5.732 | 62.346 | 1.433 | 5.732 | 62.346 |
| 5 | 1.108 | 4.432 | 66.777 | 1.108 | 4.432 | 66.777 |
Extraction method: principal component analysis
Effect size (f2)
Effect size (f2) is used to measure the magnitude of the impact on the coefficient of determination (R2) of endogenous variables after removing exogenous variables, thereby assessing the substantive contribution of specific variables to the model. This study determined the level of influence according to Cohen's [71] recommended standards: 0.02, 0.15, and 0.35 represent small, medium, and large effect sizes, respectively. The specific effect sizes for each path are summarized in Table 9 below.
Table 9.
Effect size assessment f2
| f2 | |
|---|---|
| PI → CL | 0.259 |
| PI → CW | 0.067 |
| PA → CL | 0.001 |
| PA → CW | 0.058 |
| PE → CL | 0.324 |
| PE → CW | 0.030 |
| CL → CW | 0.180 |
Coefficient of determination (R2)
The coefficient of determination (R2) indicates the proportion of variance in endogenous variables explained by their predictors in the structural model. As shown in Table 10, the model explains 45.8% of variance in cognitive load (R2 = 0.458, adjusted R2 = 0.455) and 51.5% of variance in collaboration willingness (R2 = 0.515, adjusted R2 = 0.511). According to Hair et al. [72], R2 values of 0.25, 0.50, and 0.75 represent weak, moderate, and substantial explanatory power respectively. Our results indicate that the model demonstrates moderate explanatory power for cognitive load (45.8%, exceeding the 0.25 threshold but below the substantial 0.75 level) and substantial explanatory power for collaboration willingness (51.5%, surpassing the moderate 0.50 threshold). This confirms that the three perceived attributes (instrumentality, anthropomorphism, entertainment) and cognitive load collectively account for a meaningful proportion of variance in users' human–machine collaboration willingness.
Table 10.
Coefficient of Determination (R2)
| Endogenous variable | R2 | Adjusted R2 | Interpretation |
|---|---|---|---|
| CL | 0.458 | 0.455 | Moderate explanatory power |
| CW | 0.515 | 0.511 | Substantial explanatory power |
Following Hair et al. [72], R2 values of 0.25, 0.50, and 0.75 indicate weak, moderate, and substantial levels respectively
Path coefficient significance
This study verified research hypotheses through standardized path coefficients and their significance levels. Path coefficients reflect the direction and strength of influence between variables (values ranging from −1 to 1). Following the recommendations of Hair Jr et al. [73], this study employed the Bootstrapping method (5000 resamples) to calculate T-values of path coefficients for significance assessment. Significance criteria were set as: T > 1.96 (p < 0.05), T > 2.57 (p < 0.01), and T > 3.29 (p < 0.001). The specific path coefficients and test results of the structural model are shown in Fig. 3 and the Table 11 below, respectively.
Fig. 3.
PLS results of the structural model. Note: The dotted line indicates that the relationship is not valid. *** p < 0.001
Table 11.
PLS summary of PLS-SEM path analysis
| Path | O | M | STDEV | T | p |
|---|---|---|---|---|---|
| CL → CW | −0.401 | −0.402 | 0.040 | 10.060 | 0.000 |
| PA → CL | −0.023 | −0.024 | 0.035 | 0.675 | 0.500 |
| PA → CW | 0.182 | 0.183 | 0.034 | 5.393 | 0.000 |
| PE → CL | −0.442 | −0.442 | 0.033 | 13.393 | 0.000 |
| PE → CW | 0.145 | 0.145 | 0.035 | 4.137 | 0.000 |
| PI → CL | −0.403 | −0.404 | 0.033 | 12.211 | 0.000 |
| PI → CW | 0.218 | 0.219 | 0.039 | 5.537 | 0.000 |
From the table above, CL → CW exhibits a significant negative effect (β = −0.401, p < 0.001). In the perceived anthropomorphism dimension, PA → CL shows no significant effect (β = −0.023, p > 0.05), whereas PA → CW exhibits a significant positive effect (β = 0.182, p < 0.001). In the perceived entertainment dimension, PE → CL exhibits a significant negative effect (β = −0.442, p < 0.001), while PE → CW exhibits a significant positive effect (β = 0.145, p < 0.001). In the perceived instrumentality dimension, PI → CL exhibits a significant negative effect (β = −0.403, p < 0.001), and PI → CW exhibits a significant positive effect (β = 0.218, p < 0.001). Therefore, H1a, H1b, H1c, H2a, H2c, and H3 are supported, while H2b is not supported.
Mediation analysis
To demonstrate the mediating effect, this study employed the Bootstrap mediation effect test method to examine whether the mediating effect is significant, using a Bias-Corrected confidence interval (95%) and 5000 resamples to test the mediation effect results, as shown in Table 12. Following established guidelines [74, 75], we classified mediation effects based on the significance and direction of direct and indirect paths. Partial mediation occurs when both the direct effect (predictor → outcome) and indirect effect (predictor → mediator → outcome) are statistically significant, with 95% confidence intervals excluding zero, indicating that the predictor influences the outcome through both direct and mediated pathways. Full mediation occurs when the indirect effect is significant while the direct effect becomes non-significant after including the mediator, suggesting the relationship operates entirely through the mediating variable. No mediation is concluded when the indirect effect is non-significant (95% CI includes zero), regardless of direct effect significance.
Table 12.
Bootstrap mediation effect test
| Path | Effect | β(O) | M | STDEV | T | p | 2.50% | 97.50% | Proportion |
|---|---|---|---|---|---|---|---|---|---|
| PI → CL → CW | Direct effect | 0.218 | 0.219 | 0.039 | 5.537 | 0.000 | 0.139 | 0.294 | ~ 57% |
| Indirect effect | 0.162 | 0.162 | 0.021 | 7.616 | 0.000 | 0.122 | 0.205 | ~ 43% | |
| Total effect | 0.380 | 0.381 | 0.035 | 10.973 | 0.000 | 0.308 | 0.443 | ||
| PA → CL → CW | Direct effect | 0.182 | 0.183 | 0.034 | 5.393 | 0.000 | 0.114 | 0.245 | 100.0% |
| Indirect effect | 0.009 | 0.010 | 0.014 | 0.671 | 0.502 | −0.018 | 0.036 | ||
| Total effect | 0.191 | 0.192 | 0.037 | 5.184 | 0.000 | 0.116 | 0.261 | ||
| PE → CL → CW | Direct effect | 0.145 | 0.145 | 0.035 | 4.137 | 0.000 | 0.078 | 0.214 | ~ 45% |
| Indirect effect | 0.177 | 0.177 | 0.022 | 7.964 | 0.000 | 0.137 | 0.223 | ~ 55% | |
| Total effect | 0.323 | 0.322 | 0.033 | 9.706 | 0.000 | 0.257 | 0.389 |
According to the above table, the mediation analysis reveals the following patterns. For the PI → CL → CW pathway, both the direct effect (β = 0.218, p < 0.001) and indirect effect (β = 0.162, p < 0.001) are significant with 95% confidence intervals excluding zero, indicating partial mediation. The indirect effect accounts for approximately 43% of the total effect, demonstrating that perceived instrumentality is associated with collaboration willingness through both direct pathways and the cognitive load mediation mechanism. For the PA → CL → CW pathway, while the direct effect is significant (β = 0.182, p < 0.001), the indirect effect through cognitive load is non-significant (β = 0.009, p = 0.502, 95% CI: −0.018 to 0.036), indicating no mediation. This pattern suggests that perceived anthropomorphism is associated with collaboration willingness exclusively through affective channels without cognitive load reduction, representing the affordance conflict phenomenon discussed in our theoretical framework. For the PE → CL → CW pathway, both the direct effect (β = 0.145, p < 0.001) and indirect effect (β = 0.177, p < 0.001) are significant, confirming partial mediation. The indirect effect accounts for approximately 55% of the total effect, indicating that perceived entertainment is associated with collaboration willingness primarily through cognitive load reduction pathways, though a substantial direct pathway also exists. Therefore, hypotheses H4a and H4c are supported, while H4b is not supported.
Discussion
Research findings
This study yielded several important findings. First, perceived instrumentality and perceived entertainment of intelligent cockpits are associated with significantly reduced users' cognitive load and predict human–machine collaboration willingness through this mediating pathway. Against the backdrop of rapid development in connected vehicle technology, drivers need to simultaneously process multi-source information from the road environment, vehicle systems, and intelligent cockpits. When intelligent cockpits effectively share users' information processing and decision-making tasks through functions such as real-time traffic condition acquisition, intelligent navigation planning, and driving assistance, users report significantly reduced psychological resources and cognitive effort required to complete travel tasks. This finding is highly consistent with previous research [4, 48], indicating that highly instrumental technological systems can release users' cognitive resources by automating the processing of complex tasks. When users perceive that intelligent cockpits can stably and reliably execute tasks such as navigation and route planning, they do not need to continuously monitor and verify system outputs,as a result, the limited cognitive resources originally used for dual-task processing can be reallocated. Mediation effect analysis reveals that perceived instrumentality accounts for 43% of the total effect, consistent with a cognitive load reduction mechanism of cognitive load reduction: collaboration promotion.
Second, the relationship between perceived entertainment and cognitive load is even slightly stronger than that of perceived instrumentality. When users experience pleasure and enjoyment while using intelligent cockpits, they often report a state of cognitive absorption, which is associated with reduced subjective perception of mental effort [53]. Even when facing interaction tasks of the same difficulty, driven by positive emotions, users feel more relaxed and comfortable, with lower cognitive burden. This finding is consistent with the unique value of hedonic systems in driving contexts: perceived entertainment not only provides emotional satisfaction but also dissolves users' sensitivity to cognitive costs through the pleasure-masking effect. This finding is highly consistent with the Conservation of Resources Theory. When intelligent cockpits create cognitive surplus for users, based on the reciprocity principle of Social Exchange Theory, users demonstrate higher collaboration willingness and trust levels [24, 29].
However, the most theoretically significant finding is that perceived anthropomorphism shows no significant relationship with cognitive load reduction, with mediation analysis confirming complete pathway failure. This null result reveals a critical tension: while anthropomorphism directly enhances collaboration willingness through affective channels, it fails to deliver cognitive efficiency benefits, challenging the HCI assumption that anthropomorphism equates to ease of use [12, 76]. Three theoretical mechanisms may explain this paradox, though these remain speculative and warrant empirical investigation. First, anthropomorphic cues may create expectation-capability mismatch by elevating intelligence expectations beyond current technological capabilities; when system limitations force drivers into frequent error corrections, the intended convenience is offset [3, 59]. Second, human-like features may trigger social processing protocols—a social cognitive tax—requiring users to interpret system intentions and manage conversational norms, which compete with driving tasks for limited cognitive resources [41, 42]. Third, context-dependent trust asymmetry may allow anthropomorphism to build affective rapport while failing to reduce workload, necessitating continuous supervisory attention rather than delegation [9, 10].
Comparing the three perceived attributes clarifies this paradoxical pattern. Perceived instrumentality directly reduces load through functional substitution, while perceived entertainment reduces effort perception through emotional buffering [77]. In contrast, perceived anthropomorphism exhibits pathway-specific divergence: it succeeds affectively while failing cognitively. The mediation effect proportion of anthropomorphism is only 4.7% (far lower than perceived instrumentality's approximately 43% and perceived entertainment's approximately 55%), empirically confirming its failure in the cognitive pathway. This pattern—strong affective benefits coexisting with absent cognitive benefits—creates psychological tension wherein users feel willing to collaborate yet bear full cognitive burden, revealing what we term affordance conflict in the theoretical implications section below.
Finally, cognitive load serves as a critical mediator in the model, demonstrating significant negative association with collaboration willingness. When users are in a high-load state due to coping with complex road conditions, their working memory and thinking resources approach depletion, making it difficult to bear the additional cognitive investment required for human–machine collaboration [23]. From the perspective of attention competition, high cognitive load changes users' evaluation of system feedback. In extremely busy driving states, even assistive information provided by intelligent cockpits is easily perceived by users as external interference rather than support, thereby triggering the discomfort of cognitive crowding. High cognitive load is also associated with impaired judgment of collaboration value—when the brain is in an overloaded state, users lack sufficient cognitive surplus to evaluate the long-term benefits of system recommendations and focus more on immediate operational burden and risks. This myopic pattern relates to weakened trust in and dependence on cockpit intelligent functions. This study identifies cognitive load as a critical psychological mechanism through which perceived attributes are associated with collaboration willingness, revealing that affective and cognitive processes constitute relatively independent dual-pathway systems.
Theoretical implications
This study makes four theoretical contributions to human–machine interaction and intelligent vehicle research, each addressing specific gaps in existing literature. First, this study extends Technology Affordance Theory by introducing affordance conflict—the phenomenon where positive affordances fail in safety–critical contexts through competing psychological pathways. Previous research predominantly applies linear technology acceptance models [78]. Our findings reveal that technological attributes influence behavior through multi-pathway mechanisms generating internal conflicts. Perceived anthropomorphism demonstrates strong affective effects while exhibiting null cognitive effects. This challenges the assumption that perceived benefits translate uniformly across psychological mechanisms [13, 14]. By integrating cognitive load as mediator, we construct a dual-pathway framework. This framework explains when affordances considered universally positive—such as anthropomorphism [42]—may fail in resource-constrained contexts. This identification of affordance conflict establishes boundary conditions: design principles effective in low-load, hedonic contexts may prove counterproductive in high-load, safety–critical environments [25].
To formalize this contribution, we define affordance conflict as the condition wherein a single technology attribute activates competing psychological pathways producing contradictory outcomes simultaneously. Unlike design trade-offs (sacrificing one benefit for another) or design tensions (competing goals), affordance conflict involves pathway-specific divergence wherein the same attribute succeeds affectively while failing cognitively. Three theoretical mechanisms explain this conflict. Expectation-capability mismatch occurs when anthropomorphic design creates promises of human-like competence through affective cues, yet technological limitations force frequent error corrections—transforming a positive affordance into a cognitive constraint. Social cognitive tax emerges when human-like features activate positive affective responses while simultaneously triggering social processing protocols (intention inference, conversational norms) that compete with safety–critical tasks in resource-constrained driving [25, 54]. Context-dependent trust asymmetry manifests when anthropomorphism builds affective trust through emotional connection but fails to establish capability-based cognitive trust, necessitating continuous supervisory attention rather than delegation [9]. These mechanisms collectively explain why anthropomorphism exhibits pathway-specific divergence. Table 13 illustrates how anthropomorphism uniquely exhibits this conflict pattern compared to the consistent positive effects of instrumentality and entertainment across pathways.
Table 13.
Pathway effects of perceived attributes
| Attribute | Affective Pathway | Cognitive Pathway | Pattern |
|---|---|---|---|
| Instrumentality | Moderate positive | Strong load reduction (β = −0.403***) | Consistent positive |
| Entertainment | Strong positive | Moderate load reduction (β = −0.442***) | Consistent positive |
| Anthropomorphism | Strong positive (β = 0.182***) | No reduction (β = −0.023, ns) | Affordance Conflict |
*** p < 0.001, ns not significant
Second, this study advances Social Exchange Theory by revealing asymmetric exchange mechanisms in resource-constrained contexts. In these contexts, emotional benefits and cognitive costs operate independently. Unlike prior research focusing on single-dimensional value exchange [79], we adopt a resource conservation perspective [24]. This demonstrates that driving interaction constitutes a resource game between drivers and intelligent systems. Perceived instrumentality and entertainment create cognitive surplus and reduce exchange costs. In contrast, anthropomorphism increases emotional benefits without reducing cognitive costs, thereby weakening net exchange value. This asymmetric pattern extends Social Exchange Theory beyond benefit maximization logic [28, 29]. It demonstrates that cognitively demanding contexts require navigating trade-offs among value dimensions. Cognitive load's mediating role confirms that psychological resource consumption functions as critical exchange cost moderating reciprocal responses. This provides empirical grounding that collaboration stability depends on maintaining favorable cost–benefit ratios across multiple dimensions simultaneously [44].
Third, by revealing anthropomorphic attributes' cognitive pathway failure, this study engages broader discourse on algorithmic mediation and trust calibration [80, 81]. In high uncertainty driving contexts, users develop epistemic dependence on intelligent systems for critical environmental information [10]. Our affordance conflict framework reveals that anthropomorphic design may interfere with trust calibration by triggering high-capability expectations through affective cues while failing to reduce information processing costs cognitively. This aligns with Shin's [80] demonstration that trust formation involves dynamic calibration rather than static reliability assessment. When drivers experience expectation-capability mismatch through voice recognition failures, they engage in negative trust recalibration despite maintaining affective rapport—creating fragile collaboration wherein emotional acceptance coexists with cognitive skepticism. Furthermore, our social cognitive tax mechanism directly addresses Shin's [81] emphasis on balancing emotional appeal and cognitive clarity in AI systems. While anthropomorphism succeeds in creating emotional appeal (affective pathway success), it fails to provide cognitive clarity—instead introducing social processing demands that increase rather than reduce cognitive load. This imbalance explains why anthropomorphism enhances collaboration willingness through emotional channels without translating to cognitive trust or workload reduction. When design attributes functioning as cognitive constraints exacerbate mental burden within epistemic dependence relationships, deep collaboration mechanisms are hindered despite emotional acceptance. This extends intelligent cockpit research to broader HCI theory regarding balance between emotional appeal and cognitive clarity in algorithm-mediated systems [81]. It provides empirical foundation for understanding how anthropomorphic design simultaneously strengthens affective bonds while weakening cognitive trust. This phenomenon has implications across AI technologies including virtual assistants, chatbots, and autonomous robots.
Fourth, this study confirms affective and cognitive mechanisms' independence in generating collaboration willingness. This corrects traditional technology acceptance models' limitations. While existing research identifies multiple influencing factors [4, 48], systematic mechanism exploration remains limited. This hinders explanation of contradictory responses to anthropomorphism [30]. Cognitive load partially mediates perceived instrumentality and entertainment but not anthropomorphism. This confirms that affective and cognitive mechanisms operate independently: attributes can succeed affectively while failing cognitively, and vice versa. According to Conservation of Resources Theory [24, 82], processing social information in anthropomorphic interactions may cause ego depletion given limited cognitive resources. This affective-cognitive decoupling reveals inevitable design trade-offs in resource-constrained contexts. It challenges technology acceptance models' assumption [7, 78] of unidirectional, consistent causal chains from features to intentions. Our findings identify three refinements. First, contextual cognitive load moderates’ anthropomorphism effectiveness. Second, affective and cognitive pathways operate independently. Third, pursuing extremes in one dimension may create negative effects in another. This addresses significant gaps in intelligent vehicle HCI literature by empirically demonstrating differentiated psychological channel effects [83].
Practical implications
The practical relevance of these findings is underscored by current industry trends. In 2025, users' average daily interaction frequency with intelligent cockpits reached 12.7 times, representing a 215% increase from 2023 [84]. Survey data indicate that 61.3% of users consider intelligent cockpit configurations a significant enhancement to vehicle purchase interest, with 17.4% viewing them as essential requirements, and the Chinese market penetration rate is projected to reach 75% [85]. However, despite voice interaction achieving a 98.6% success rate, functional instability accounts for 45% of user complaints [86], and touchscreen interaction steps increased from 1.86 in 2023 to 2.42 in 2024 [87]. Against this backdrop of rapid adoption but persistent usability challenges, our findings offer several important insights for practitioners.
First, the findings offer intelligent cockpit designers and automobile manufacturers a systematic and comprehensive picture that help them better understand how perceived attributes from different dimensions relate to cognitive load and are associated with users' human–machine collaboration willingness. For instance, intelligent cockpit developers may need to proactively manage users' perception of system anthropomorphism by reducing unnecessary social interactions while providing clear and easily understandable guidance when emotional interaction is needed. Meanwhile, automobile manufacturers can alleviate cognitive load by carefully designing and managing the degree of anthropomorphism in voice assistants. For instance, they can automatically reduce the assistant's social expressiveness during complex driving tasks and restore emotional interaction when driving tasks are simple or during parking.
Second, intelligent cockpit service providers can redesign and renovate their system interfaces by incorporating context-aware functionality that dynamically adjusts interaction modes based on real-time driving load, thereby alleviating users' concerns about cognitive overload. Specifically, the system should be able to recognize high-load driving scenarios (such as urban congestion and complex intersections) and automatically switch to a high-instrumentality, low-anthropomorphism streamlined interaction mode, providing only critical navigation guidance and safety alerts. Conversely, in low-load scenarios (such as highway cruising or parking breaks), the system can appropriately increase anthropomorphism and entertainment, providing richer emotional interaction and more engaging content. This context-adaptive design may effectively reduce users' cognitive burden while leveraging the emotional value of anthropomorphism at appropriate moments, achieving an optimal balance between technical efficiency and emotional experience.
Third, intelligent cockpit designers should recognize that anthropomorphism is not always the optimal choice in all contexts. The anthropomorphism paradox revealed by this study indicates that excessive pursuit of human-like interaction may backfire in high cognitive load driving scenarios. Therefore, designers need to reexamine the current industry-wide practice of pursuing extreme anthropomorphism. On one hand, manufacturers can provide users with customization options for anthropomorphism levels, allowing users to adjust the personification degree of voice assistants according to personal preferences and driving habits; on the other hand, the system can intelligently recommend the most suitable anthropomorphism settings for users through machine learning algorithms based on users' historical interaction data and driving behavior patterns. This personalized and controllable anthropomorphism design can respect user preferences while avoiding the potential cognitive costs associated with anthropomorphism.
Finally, the instrumentality and entertainment functions of intelligent cockpits should receive more attention and optimization. This study found that perceived instrumentality and perceived entertainment not only directly promote collaboration willingness but also exert indirect facilitation through significantly reducing cognitive load. Therefore, manufacturers should prioritize resources to enhance core instrumental functions such as navigation accuracy, intelligent route planning, and driving assistance reliability, ensuring that the system can genuinely share driving tasks for users and create cognitive surplus. Meanwhile, the design of entertainment functions should not be neglected. By providing high-quality multimedia content, personalized entertainment recommendations, and immersive interactive experiences, intelligent cockpits can leverage the buffering effect of pleasant emotions to reduce users' sensitivity to cognitive costs. Particularly in long-distance driving or waiting scenarios, entertainment functions can effectively alleviate fatigue and boredom, maintaining users' positive attitudes toward the system and collaboration willingness. It is worth noting that instrumental and entertainment functions should intelligently switch according to driving contexts, avoiding the distraction of drivers' attention through entertainment content during high-load driving, ensuring that driving safety remains the primary consideration.
Limitations and future research
This study has several limitations suggesting future research directions. First, the specific triggers of cognitive load in anthropomorphic interactions—vocal emotion, social pressure, or uncanny valley effects—remain unclear. Future research should utilize neurophysiological tools (eye-tracking, EEG, fNIRS) to move beyond subjective questionnaires and pinpoint mechanistic load sources, providing direct empirical evidence for theoretical conjectures regarding social cognitive tax and trust asymmetries underlying affordance conflict. Second, our laboratory procedure, while ensuring high internal validity, may not fully replicate real-world driving stressors and multitasking complexities. Drivers likely prioritize safety–critical information differently under unpredictable traffic conditions than in simulators. Naturalistic driving studies or long-term field experiments are necessary to validate whether the identified affordance conflict and its impact on collaboration willingness persist across varying environmental uncertainty and extended usage.
Third, several boundary conditions limit generalizability. The sample was restricted to Chinese licensed drivers, and the testing platform to HarmonyOS (AITO M7, M9) with specific design philosophy. User perceptions of anthropomorphism may vary across cultural contexts or competing platforms (Tesla, BMW, Mercedes). Furthermore, our results primarily apply to Level 2 assisted driving. As automation shifts to L3 +, the driver's transition from active controller to passive observer may fundamentally alter cognitive load dynamics and anthropomorphic feature utility. Future research should systematically examine these boundaries through cross-cultural, multi-platform, and cross-autonomy-level comparisons. Finally, the cross-sectional correlational design precludes strict causal inference and cannot capture dynamic response evolution such as cognitive adaptation or trust collapse over time. Longitudinal tracking or experimental manipulation is required to validate these theoretical mechanisms and observe habituation patterns. Moreover, expanding the model to include perceived safety and privacy would provide a more comprehensive framework for predicting human–machine collaboration in intelligent vehicle ecosystems.
Conclusion
Although intelligent cockpits have significantly enhanced driving information acquisition efficiency and interaction experience through advanced connected technologies, the relationships between their diversified perceived attributes and users' psychological resource patterns exhibit complex differentiation, ultimately jointly predicting users' human–machine collaboration willingness. Similar to recent research that has begun focusing on the trade-offs between costs and benefits of human–machine interaction in high-load contexts (e.g., [20, 32, 54]), this study identifies perceived instrumentality (i.e., functional utility) and perceived entertainment (i.e., emotional buffering) as key predictors associated with reduced cognitive load and subsequently enhanced human–machine collaboration willingness. Among these, the relationships between these perceived attributes and collaboration willingness are primarily mediated by cognitive load as a core psychological mechanism,however, it is noteworthy that while perceived anthropomorphism demonstrates strong direct positive association with collaboration willingness through emotional connection, it shows no significant relationship with load reduction in the cognitive pathway. This pattern suggests a profound tension between emotional appeal and cognitive efficiency in safety–critical tasks.
Authors’ contributions
S. L. conceived and designed the study, developed the theoretical framework, conducted the data collection and analysis, interpreted the results, and wrote the manuscript. The author read and approved the final manuscript.
Funding
The authors declare that no funds, grants, or other support were received during the preparation of this manuscript.
Data availability
The datasets used and analyzed during the current study are available from the corresponding author on reasonable request.
Declarations
Ethics approval and consent to participate
This study was approved by the Guangzhou College of Technology and Business Bioethics Committee (Approval No. GZCTB-EC-2025–041) and adhered to the principles of the Declaration of Helsinki. All participants provided informed consent before participating in the survey. The research was conducted in accordance with ethical guidelines for human research. Participation was voluntary and anonymous, and participants were informed about the purpose of the study.
Consent for publication
Not applicable.
Competing interests
The authors declare no competing interests.
Footnotes
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
References
- 1.You S, Robert LP Jr. Human-robot similarity and willingness to work with a robotic Co-worker. Proceedings of the 2018 ACM/IEEE International Conference on Human-Robot Interaction. 2018. pp. 251-260. 10.1145/3171221.3171281.
- 2.You Z, Ma N, Wang Y, Jiang Y. Cognitive mechanism and evaluation method of human-machine interaction in intelligent vehicle cockpit. Scientific Research Publishing, Inc. USA; 2023. https://books.google.co.jp/books?hl=zh-CN&lr=&id=fe7QEAAAQBAJ&oi=fnd&pg=PP2&dq=Cognitive+mechanism+and+evaluation+method+of+human-machine+interaction+in+intelligent+vehicle+co&ots=VdSHlaDAo6&sig=1dU-B9RWtxRVesVr5VdyHTqhYhU.
- 3.Waytz A, Heafner J, Epley N. The mind in the machine: anthropomorphism increases trust in an autonomous vehicle. J Exp Soc Psychol. 2014;52:113–7. 10.1016/j.jesp.2014.01.005. [Google Scholar]
- 4.Song Y, Zhang M, Hu J, Cao X. Dancing with service robots: the impacts of employee-robot collaboration on hotel employees’ job crafting. Int J Hosp Manag. 2022;103:103220. 10.1016/j.ijhm.2022.103220. [Google Scholar]
- 5.Strayer DL, Cooper JM, Turrill J, Coleman J, Medeiros-Ward N, Biondi F. Measuring cognitive distraction in the automobile. 2013. https://trid.trb.org/View/1252566. [DOI] [PubMed]
- 6.Wang PX, Kim S, Kim M. Robot anthropomorphism and job insecurity: the role of social comparison. J Bus Res. 2023;164:114003. 10.1016/j.jbusres.2023.114003. [Google Scholar]
- 7.Davis FD. Perceived usefulness, perceived ease of use, and user acceptance of information technology. MIS Q. 1989;13(3):319. 10.2307/249008. [Google Scholar]
- 8.Van Der Heijden H. User acceptance of hedonic information systems1. MIS Q. 2004;28(4):695–704. 10.2307/25148660. [Google Scholar]
- 9.Lee JD, See KA. Trust in automation: designing for appropriate reliance. Hum Factors. 2004;46(1):50–80. 10.1518/hfes.46.1.50.30392. [DOI] [PubMed] [Google Scholar]
- 10.Parasuraman R, Riley V. Humans and automation: use, misuse, disuse, abuse. Hum Factors. 1997;39(2):230–53. 10.1518/001872097778543886. [Google Scholar]
- 11.Gibson JJ. The ecological approach to visual perception. Houghton Mifflin; 1979. https://berlinarchaeology.wordpress.com/wp-content/uploads/2022/03/gibson-james-j.-19862015-the-ecological-approach-to-visual-perception-2015-re-issue.pdf.
- 12.Norman DA. The psychology of everyday things. New York: Basic Books; 1988. p. 272.
- 13.Leonardi PM. When flexible routines meet flexible technologies: Affordance, constraint, and the imbrication of human and material agencies1. MIS Q. 2011;35(1):147–67. 10.2307/23043493. [Google Scholar]
- 14.Volkoff O, Strong DM. Critical realism and affordances: theorizing IT-associated organizational change processes1. MIS Q. 2013;37(3):819–34. 10.25300/MISQ/2013/37.3.07. [Google Scholar]
- 15.Hartson R. Cognitive, physical, sensory, and functional affordances in interaction design. Behav Inf Technol. 2003;22(5):315–38. 10.1080/01449290310001592587. [Google Scholar]
- 16.Zhang P. Technical opinionMotivational affordances: Reasons for ICT design and use. Commun ACM. 2008;51(11):145–7. 10.1145/1400214.1400244. [Google Scholar]
- 17.Bradner E. Social affordances of computer-mediated communication technology: Understanding adoption. CHI ’01 Extended Abstracts on Human Factors in Computing Systems. 2001. pp. 67-68. 10.1145/634067.634111.
- 18.Treem JW, Leonardi PM. Social media use in organizations: exploring the affordances of visibility, editability, persistence, and association. Ann Int Commun Assoc. 2013;36(1):143–89. 10.1080/23808985.2013.11679130. [Google Scholar]
- 19.Leung L. Impacts of net-generation attributes, seductive properties of the internet, and gratifications-obtained on internet use. Telemat Informatics. 2003;20(2):107–29. 10.1016/S0736-5853(02)00019-9. [Google Scholar]
- 20.Magliocca P, Canestrino R, Carayannis EG, Gagliardi AR. Understanding human–technology interaction: evolving boundaries. Eur J Innov Manag. 10.1108/EJIM-04-2024-0341.
- 21.Pozzi G, Pigni F, Vitari C. Affordance theory in the IS discipline: a review and synthesis of the literature. AMCIS 2014 Proceedings; 2014. https://shs.hal.science/halshs-01923663v1.
- 22.Fu S, Li H, Liu Y, Pirkkalainen H, Salo M. Social media overload, exhaustion, and use discontinuance: examining the effects of information overload, system feature overload, and social overload. Inf Process Manage. 2020;57(6):102307. 10.1016/j.ipm.2020.102307. [Google Scholar]
- 23.Sweller J. Cognitive load during problem solving: effects on learning. Cogn Sci. 1988;12(2):257–85. 10.1207/s15516709cog1202_4. [Google Scholar]
- 24.Hobfoll SE. Conservation of resources: a new attempt at conceptualizing stress. Am Psychol. 1989;44(3):513. [DOI] [PubMed] [Google Scholar]
- 25.Wickens CD. Multiple resources and mental workload. Hum Factors. 2008;50(3):449–55. 10.1518/001872008X288394. [DOI] [PubMed] [Google Scholar]
- 26.Kahneman D. Attention and Effort. Prentice-Hall; 1973. https://philpapers.org/rec/kahaae.
- 27.Paas FGWC. Training strategies for attaining transfer of problem-solving skill in statistics: a cognitive-load approach. J Educ Psychol. 1992;84(4):429–34. 10.1037/0022-0663.84.4.429. [Google Scholar]
- 28.Blau P. Exchange and power in social life. John Wiley & Sons; 1964. https://www.taylorfrancis.com/books/mono/10.4324/9780203792643/exchange-power-social-life-peter-blau.
- 29.Cropanzano R, Mitchell MS. Social exchange theory: an interdisciplinary review. J Manage. 2005;31(6):874–900. 10.1177/0149206305279602. [Google Scholar]
- 30.Wang B, Liu Y, Parker SK. How does the use of information communication technology affect individuals? A work design perspective. Acad Manage Ann. 2020;14(2):695–725. 10.5465/annals.2018.0127. [Google Scholar]
- 31.Le KBQ, Sajtos L, Fernandez KV. Employee-(ro)bot collaboration in service: an interdependence perspective. J Serv Manag. 2023;34(2):176–207. 10.1108/JOSM-06-2021-0232. [Google Scholar]
- 32.Sourav MSG, Javaid A, Cheng L. A review of AI in human-machine cooperation: Machine perspective. ACM Transactions on Autonomous and Adaptive Systems. 2025:3774318. 10.1145/3774318.
- 33.Kunz WH, Sajtos L, Flavián C. Beyond replacement: human-machine collaboration in the age of AI. J Serv Manag. 2025;36(4):477–94. 10.1108/JOSM-04-2025-0194. [Google Scholar]
- 34.Caniëls MCJ, Chiocchio F, Van Loon NPAA. Collaboration in project teams: the role of mastery and performance climates. Int J Proj Manage. 2019;37(1):1–13. 10.1016/j.ijproman.2018.09.006. [Google Scholar]
- 35.Franç NA, Chiocchio O, Grenier SOTA, Neill NA, Savaria K, Willms JD. The effects of collaboration on performance: a multilevel validation in project teams. Int J Project Organisation Manag. 2012;4(1):1. 10.1504/IJPOM.2012.045362.
- 36.Lu L, Cai R, Gursoy D. Developing and validating a service robot integration willingness scale. Int J Hosp Manag. 2019;80:36–51. 10.1016/j.ijhm.2019.01.005. [Google Scholar]
- 37.Cai M, Zhang WY, Zhang K. Manuhub: a semantic web system for ontology-based service management in distributed manufacturing environments. IEEE Trans Syst Man Cybern Part A Syst Hum. 2011;41(3):574–82. 10.1109/TSMCA.2010.2076395. [Google Scholar]
- 38.Sun J, Huang Y, Huang X, Zhang J, Zhang H. Effect of proactive interaction on trust in autonomous vehicles. Sustainability. 2024;16(8):3404. 10.3390/su16083404. [Google Scholar]
- 39.Ali F, Dogan S, Chen X, Cobanoglu C, Limayem M. Friend or a foe: understanding generation Z employees’ intentions to work with service robots in the hotel industry. Int J Hum Comput Interact. 2023;39(1):111–22. 10.1080/10447318.2022.2041880. [Google Scholar]
- 40.Majchrzak A, Markus ML. Technology affordances and constraints in management information systems (MIS). In Encyclopedia of management theory. SAGE Publications; 2013. pp. 832-836. https://papers.ssrn.com/abstract=2192196.
- 41.Epley N, Waytz A, Akalis S, Cacioppo JT. When we need a human: motivational determinants of anthropomorphism. Soc Cogn. 2008;26(2):143–55. 10.1521/soco.2008.26.2.143. [Google Scholar]
- 42.Nass C, Moon Y. Machines and mindlessness: Social responses to computers. J Soc Issues. 2000;56(1):81–103. 10.1111/0022-4537.00153. [Google Scholar]
- 43.Homans GC. Social behavior as exchange. Am J Sociol. 1958;63(6):597–606. 10.1086/222355. [Google Scholar]
- 44.Thibaut JW, Kelley HH. The Social Psychology of Groups. John Wiley & Sons; 1959. https://books.google.co.jp/books?hl=zh-CN&lr=&id=KDH5Hc9F2AkC&oi=fnd&pg=PA1&dq=Thibaut,+J.+W.,+%26+Kelley,+H.+H.+(1959).+The+social+psychology+of+groups.+John+Wiley+%26+Sons.&ots=T0zkHm0cBd&sig=E5cd9KspQdcgG6QRGGqQjaEuyAo.
- 45.Zhou P, Xie Y, Liang C. How to increase consumers’ continued use intention of artificial intelligence voice assistants? The role of anthropomorphic features. Electron Mark. 2023;33(1):60. 10.1007/s12525-023-00681-0. [Google Scholar]
- 46.Letondal C, Vinot JL, Pauchet S, Boussiron C, Rey S, Becquet V, Lavenir C. Being in the sky: Framing tangible and embodied interaction for future airliner cockpits. Proceedings of the Twelfth International Conference on Tangible, Embedded, and Embodied Interaction. 2018. pp.656-666. 10.1145/3173225.3173229.
- 47.Yin M, Jiang S, Niu X. Can AI really help? The double-edged sword effect of AI assistant on employees’ innovation behavior. Comput Human Behav. 2024;150:107987. 10.1016/j.chb.2023.107987. [Google Scholar]
- 48.Leung XY, Zhang H, Lyu J, Bai B. Why do hotel frontline employees use service robots in the workplace? A technology affordance theory perspective. Int J Hosp Manag. 2023;108:103380. 10.1016/j.ijhm.2022.103380. [Google Scholar]
- 49.Mariani M, Borghi M. Customers’ evaluation of mechanical artificial intelligence in hospitality services: a study using online reviews analytics. Int J Contemp Hosp Manage. 2021;33(11):3956–76. 10.1108/IJCHM-06-2020-0622. [Google Scholar]
- 50.Dittrich M, Ko S, Rieger T. Impact of voice assistants’ conversational style on cognitive driver distraction. Proceedings of the 25th ACM International Conference on Intelligent Virtual Agents. 2025. pp. 1-4. 10.1145/3717511.3749294.
- 51.Mosaferchi S, Ingenito M, Vitiello G, Mortezapour A, Naddeo A. Towards more trustworthy automated driving: Comparing anthropomorphic and biophilic interfaces: comparing biophilic and anthropomorphic in automated driving. Proceedings of the 16th Biannual Conference of the Italian SIGCHI Chapter. 2025:1-6. 10.1145/3750069.3750440.
- 52.Wong PNY, Brumby DP, Babu HVR, Kobayashi K. Voices in self-driving cars should be assertive to more quickly grab a distracted driver’s attention. Proceedings of the 11th International Conference on Automotive User Interfaces and Interactive Vehicular Applications. 2019. pp. 165-176. 10.1145/3342197.3344535.
- 53.Agarwal R, Karahanna E. Time flies when you’re having fun: Cognitive absorption and beliefs about information technology usage1. MIS Q. 2000;24(4):665–94. 10.2307/3250951. [Google Scholar]
- 54.Cai H, Lin Y. Coordinating cognitive assistance with cognitive engagement control approaches in human–machine collaboration. IEEE Trans Syst Man Cybern Part A Syst Hum. 2012;42(2):286–94. 10.1109/TSMCA.2011.2169953. [Google Scholar]
- 55.Lam K, Yaici Y. Impact of cognitive complexity in In-vehicle voice interfaces on driver workload and performance: a simulator-based analysis of safety, performance, and economic impact. 2025. https://urn.kb.se/resolve?urn=urn:nbn:se:kth:diva-369441.
- 56.Li X, Wang Y, Liu Y. Enhancing driver performance under cognitive load: The role of tactile and combined auditory-tactile warnings in simulated underground mining transport vehicle environments. Ergonomics. 2025:1-18. 10.1080/00140139.2025.2574348. [DOI] [PubMed]
- 57.Freund L. As if conscious: the instrumentality theorem in artificial intelligence. AI Ethics. 2025;5(6):5925–40. 10.1007/s43681-025-00832-3. [Google Scholar]
- 58.Gabrecht KM. Human factors of semi-autonomous robots for urban search and rescue [Thesis (University of Nottingham only)]. University of Nottingham; 2016. https://eprints.nottingham.ac.uk/35458/.
- 59.Goudey A, Bonnin G. Must smart objects look human? Study of the impact of anthropomorphism on the acceptance of companion robots. Recherche Et Applications En Marketing (English Edition). 2016;31(2):2–20. 10.1177/2051570716643961. [Google Scholar]
- 60.Churchill GA. A paradigm for developing better measures of marketing constructs. J Mark Res. 1979;16(1):64–73. 10.1177/002224377901600110. [Google Scholar]
- 61.Hyun MY, O’Keefe RM. Virtual destination image: Testing a telepresence model. J Bus Res. 2012;65(1):29–35. 10.1016/j.jbusres.2011.07.011. [Google Scholar]
- 62.Balakrishnan J, Abed SS, Jones P. The role of meta-UTAUT factors, perceived anthropomorphism, perceived intelligence, and social self-efficacy in chatbot-based services? Technol Forecast Soc Change. 2022;180:121692. 10.1016/j.techfore.2022.121692. [Google Scholar]
- 63.Lee KC, Chung N. Empirical analysis of consumer reaction to the virtual reality shopping mall. Comput Human Behav. 2008;24(1):88–104. 10.1016/j.chb.2007.01.018. [Google Scholar]
- 64.Moon J-W, Kim Y-G. Extending the TAM for a world-wide-web context. Inform Manag. 2001;38(4):217–30. 10.1016/S0378-7206(00)00061-6. [Google Scholar]
- 65.Deng Z, Hong Z, Ren C, Zhang W, Xiang F. What predicts patients’ adoption intention toward mHealth services in China: empirical study. JMIR Mhealth Uhealth. 2018;6(8):e172. 10.2196/mhealth.9316. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 66.Naismith LM, Cheung JJH, Ringsted C, Cavalcanti RB. Limitations of subjective cognitive load measures in simulation-based procedural training. Med Educ. 2015;49(8):805–14. 10.1111/medu.12732. [DOI] [PubMed] [Google Scholar]
- 67.Olya HGT, Gavilyan Y. Configurational models to predict residents’ support for tourism development. J Travel Res. 2017;56(7):893–912. 10.1177/0047287516667850. [Google Scholar]
- 68.Olya HGT, Han H. Antecedents of space traveler behavioral intention. J Travel Res. 2020;59(3):528–44. 10.1177/0047287519841714. [Google Scholar]
- 69.Podsakoff PM, MacKenzie SB, Lee J-Y, Podsakoff NP. Common method biases in behavioral research: a critical review of the literature and recommended remedies. J Appl Psychol. 2003;88(5):879. [DOI] [PubMed] [Google Scholar]
- 70.Podsakoff PM, Organ DW. Self-reports in organizational research: problems and prospects. J Manage. 1986;12(4):531–44. 10.1177/014920638601200408. [Google Scholar]
- 71.Cohen J. Set correlation and contingency tables. Appl Psychol Meas. 1988;12(4):425–34. 10.1177/014662168801200410. [Google Scholar]
- 72.Hair JF, Risher JJ, Sarstedt M, Ringle CM. When to use and how to report the results of PLS-SEM. Eur Bus Rev. 2019;31(1):2–24. 10.1108/EBR-11-2018-0203. [Google Scholar]
- 73.Hair FJ Jr, Sarstedt M, Hopkins LG, Kuppelwieser V. Partial least squares structural equation modeling (PLS-SEM): an emerging tool in business research. Eur Bus Rev. 2014;26(2):106-121. 10.1108/EBR-10-2013-0128.
- 74.Baron RM, Kenny DA. The moderator–mediator variable distinction in social psychological research: conceptual, strategic, and statistical considerations. J Pers Soc Psychol. 1986;51(6):1173–82. 10.1037/0022-3514.51.6.1173. [DOI] [PubMed] [Google Scholar]
- 75.Zhao X, Lynch JG, Chen Q. Reconsidering baron and kenny: myths and truths about mediation analysis. J Consum Res. 2010;37(2):197–206. 10.1086/651257. [Google Scholar]
- 76.Reeves B, Nass C. The media equation: how people treat computers, television, and new media like real people. Cambridge, UK. 1996;10(10):19–36. [Google Scholar]
- 77.Fredrickson BL. The role of positive emotions in positive psychology: the broaden-and-build theory of positive emotions. Am Psychol. 2001;56(3):218–26. 10.1037/0003-066X.56.3.218. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 78.Venkatesh, Morris, Davis, Davis. User acceptance of information technology: Toward a unified view. Mis Quarterly. 2003;27(3):425. 10.2307/30036540.
- 79.Halbesleben JRB, Neveu J-P, Paustian-Underdahl SC, Westman M. Getting to the “COR”: Understanding the role of resources in conservation of resources theory. J Manage. 2014;40(5):1334–64. 10.1177/0149206314527130. [Google Scholar]
- 80.Shin D. Artificial misinformation: Exploring human-algorithm interaction online. Cham: Palgrave Macmillan; 2024. p. 286.
- 81.Shin D. Debiasing AI: Rethinking the intersection of innovation and sustainability (1st ed.). Routledge; 2025. 10.1201/9781003530244.
- 82.Baumeister RF, Vohs KD. Self-regulation, ego depletion, and motivation. Soc Personal Psychol Compass. 2007;1(1):115–28. 10.1111/j.1751-9004.2007.00001.x. [Google Scholar]
- 83.Li W, Li G, Tan R, Wang C, Sun Z, Li Y, et al. Review and perspectives on human emotion for connected automated vehicles. Automot Innov. 2024;7(1):4–44. 10.1007/s42154-023-00270-z. [Google Scholar]
- 84.Baidu Intelligent Cloud. 2025 intelligent cockpit voice revolution: In-depth evaluation and comparison of multimodal large models. 2025. https://cloud.baidu.com/article/3829077.
- 85.People.cn. Fujian Provincial Department of Industry and Information Technology. 2023. https://gxt.fujian.gov.cn/zwgk/xw/hydt/xydt/202306/t20230627_6193658.htm.
- 86.Power JD. Pulse of the 2024 China automotive market: How to break through red ocean competition and the low-value dilemma. 2024. https://china.jdpower.com/zh-hans/press-releases/2024-China-Market-Insight.
- 87.Power JD, Tongji University HVR Lab. J.D. Power 2024 China intelligent cockpit study: AI technology application improves user experience and breaks the homogeneity dilemma of intelligent cockpits. 2024. CRI Online. https://auto.cri.cn/20241212/271aa089-6944-2d2a-b9ea-19e896e9e39b.html.
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Data Availability Statement
The datasets used and analyzed during the current study are available from the corresponding author on reasonable request.