The rapid advancement of emotional artificial intelligence (AI), specifically affect recognition technology (ART), raises significant ethical, legal, and societal concerns. Unlike other AI applications, ART harvests data from an individual's non-conscious psychophysical state, often without their knowledge or consent. This poses unique challenges, especially concerning privacy and potential misuse. ART's increasing deployment in various contexts – from in-cabin personalization systems to public security – necessitates a thorough understanding of its acceptance among the general public. This study focuses on the behavioral determinants of attitudes toward ART, particularly concerning the non-conscious data harvesting practices of both governmental and private sector actors. The research aims to address the gap in existing theoretical models, such as the Technology Acceptance Model (TAM), which doesn't fully capture the nuanced impact of cultural and personal values on the adoption of such invasive technologies. By integrating insights from the mindsponge model, which considers the human mind as a filtering mechanism for new information, this study provides a more comprehensive framework for analyzing the behavioral determinants of ART acceptance. This is crucial for policymakers seeking to develop ethical guidelines and regulations for the responsible development and deployment of ART, ensuring that its benefits outweigh its potential risks.

Existing models like the Innovation Diffusion Theory, Theory of Reasoned Action, Theory of Planned Behavior, TAM, and Unified Theory of Acceptance and Use of Technology have attempted to understand technology adoption. However, these models have limitations in explaining the unique aspects of ART. TAM, for example, focuses on perceived usefulness and ease of use, neglecting the influence of cultural values and the non-conscious nature of data collection in ART. The mindsponge model provides a more comprehensive framework, considering the individual's core values, cultural environment, and the filtering process of the human mind when accepting new technologies. The integration of TAM and the mindsponge model aims to provide a more robust framework for understanding ART acceptance, considering both perceived utility and ease of use as well as the impact of cultural and personal values.

This study employs a three-step methodology. First, variables relevant to ART acceptance, based on the original TAM, are identified: perceived utility, perceived familiarity with AI, and attitude toward AI systems. Second, the mindsponge model is incorporated to expand TAM by adding personal core values (religiosity), cultural factors (region), political factors (political regime of home country), and income level. Third, Bayesian multi-level analysis is used, treating these personal, political, and environmental factors as varying intercepts in Bayesian network models. A survey was administered to 1015 young adults (aged 18-27) from diverse backgrounds, collected via a Google Form distributed through online classes at Ritsumeikan Asia Pacific University (APU) in Japan. The dataset includes socio-demographic information, social media usage patterns, perceived utility and familiarity with AI technologies, and attitudes towards non-conscious data collection by both the private and public sectors. The Bayesian multi-level analysis allows for the comparison of the plausibility of different models and provides robust estimates of the effects of predictor variables on attitudes toward ART, accounting for the non-random nature of the online survey data. Model evaluation criteria included Rhat values, effective sample sizes, Pareto-Smoothed Importance Sampling (PSIS-LOO) tests for goodness-of-fit, and Bayesian model weights (pseudo-BMA without and with Bayesian bootstrap, and Bayesian stacking) to determine the relative plausibility of different models.

The Bayesian model weight comparison revealed that models incorporating cultural factors (region) and personal core values (religiosity) significantly outperformed the traditional TAM models. For both private and public sector data harvesting, models using regions as varying intercepts were the best-performing. The analysis identified several key predictors of attitudes toward non-conscious data harvesting: * **Positive predictors:** Public social media posting, less frequent participation in heated online debates, greater familiarity with AI technologies, and higher perceived utility of AI technologies were all positively associated with a more accepting attitude towards non-conscious data harvesting by both private and public sectors. * **Negative predictor:** Increased time spent on social media negatively correlated with attitudes toward government data collection, suggesting greater distrust in government surveillance compared to private sector data collection. This effect was not as pronounced or even ambiguous in relation to the private sector. Consistent across the models, perceived utility and familiarity with AI technologies emerged as strong positive predictors of acceptance, regardless of the data collector (government or private sector). This underscores the importance of user experience and understanding of AI technologies in shaping attitudes towards non-conscious data harvesting. The results suggest that a heightened sense of digital self-efficacy and control over social media behaviors are crucial factors in mitigating concerns over non-conscious dataveillance.

The findings of this study directly address the research questions by highlighting the limitations of traditional TAM models when applied to the context of ART. The integration of the mindsponge model, incorporating cultural and personal factors, significantly improved the explanatory power of the models. The strong positive relationship between perceived utility and familiarity with AI technologies, and a more accepting attitude toward ART, supports the hypothesis that individual agency and control are key factors in shaping these attitudes. The negative correlation between social media engagement time and attitudes towards government data collection, but not private sector data collection, suggests a significant difference in public trust in these two entities regarding data privacy and surveillance. These results provide valuable insights for policymakers involved in regulating ART, emphasizing the importance of promoting digital literacy, transparency, and user control in the design and deployment of these technologies.

This study offers a significant contribution by extending TAM with the mindsponge model, providing a more nuanced understanding of ART acceptance. The findings highlight the significant role of culture and personal values, alongside technological familiarity and perceived utility. The differences in trust between governmental and private sector data collection warrant further research. Future studies could focus on exploring the specific mechanisms underlying the identified relationships, investigating the impact of awareness of data collection practices, and considering the long-term impacts of ART on individual well-being and societal values.

The study's limitations include the non-random sampling of students from a single university, the use of self-reported data, and the potential for response bias in online surveys. The age range of the participants (18-27) also limits the generalizability to other demographic groups. Future research should aim to address these limitations through more representative samples and diverse data collection methods. Despite these limitations, the findings provide valuable insights into the complexities of ART acceptance and offer a framework for future research in this rapidly evolving field.