Understanding Public Perceptions of Artificial Intelligence in China in Relation to Advanced Air Mobility ^†

Abstract

This study explored the public perceptions of advanced air mobility (AAM) in relation to artificial intelligence (AI), focusing on openness, usefulness, practical use, and trust. A survey of 93 participants was conducted using a five-point Likert scale to assess these dimensions. The results show that, while the respondents viewed AI Use favorably, openness and usefulness were rated lower, indicating hesitancy about adopting AI technologies in AAM. Trust in AI was moderate, with diverse opinions regarding fully autonomous versus human-assisted AI systems. A factor analysis using a principal components approach was conducted to further investigate these variables, but the results did not yield clear or distinct factors with trust, openness, and usefulness. This suggests that public perceptions of AI in AAM are complex and interconnected, with trust playing a key role, highlighting the need for further research to address these concerns and promote the broader acceptance of AI in AAM.

1. Introduction

Finding better ways to achieve the goal of delivering passengers and goods quickly and safely has always been an interesting research topic. Therefore, Advanced Air Mobility (AAM), a rapidly emerging technology that offers fast, safe, convenient, and sustainable transportation to both people and products, has attracted attention. However, the concept of urban air mobility has not yet been fully developed.

In the 1940s, Los Angeles adopted helicopter transportation operations in urban areas [1] as a pilot attempt toward air mobility. Although this attempt stopped due to safety reasons, the city had already proved AAM’s potential value to passengers. In recent years, notable developments in eVTOL brought AAM onto the stage. With the potential to revolutionize not just aviation but also mobility systems, AAM holds the ability to extend future travel [2]. Based on this quickly emerging technology with great potential, Goyal et al. [3] predicted that the annual AAM market valuation of the United States will reach 2.5 billion in the coming years, and AAM might bring new possibilities and consumer preferences in the future.

Since AAM will benefit society more if it is more widely accepted due to its speed and safety, which people expect, public awareness of AAM is essential to its commercialization, particularly in the AAM-related sector, which may be able to uncover vital details that could improve AAM’s development [2]. For AAM to become commercially viable, widespread public adoption is a necessary first step toward securing public funding.

Although the definition of AAM still has a long way to go, the description of AAM is accepted as an electric aircraft or an electric vertical take-off and landing aircraft [4]. Furthermore, AAM is defined as Unmanned Aerial Vehicles (UAV), and artificial intelligence is at the core of AAM [5]. When AI enables autonomous navigation, air traffic management, and real-time decision-making, this will raise both hopes and concerns regarding safety, trust, and reliability.

Estimating the rapid increase in autonomous aviation poses challenges and significant impacts on air transportation management. Especially for end users, in the field of AI-guided unmanned aviation, the public has placed higher demands on AI systems [6]. Therefore, one of the most vital phases in developing AAM is to investigate how the public perceives artificial intelligence in the context of AAM; this is the purpose of this article as well.

This study explores the public’s perception of advanced air mobility from various artificial intelligence aspects, including openness, usefulness, use, and trust factors, to help stakeholders gauge public awareness. More specifically, this study identifies the key challenges and concerns by reviewing the literature and creating a questionnaire based on validated questionnaires. To gain a thorough understanding of AAM’s public acceptance in the AI context, this study adapts questions from these surveys and administers them to the public. The survey questions were rated on a five-point Likert scale, ranging from strongly agree to strongly disagree. Because AAM is not a widely used technology, the survey asked participants several questions to gauge their propensity for adoption. After administering the survey, the response data were collected and subsequently analyzed using statistical software.

This paper is composed of the following sections: Introduction, Research Methods, Results and Analysis, Factor Analysis, and Conclusions. The Introduction Section presents the context of AAM, and the Research Methods Section introduces the research ideas and methods used in this paper. The Results and Analysis Section presents the descriptive statistics, the Factor Analysis Section continues digging into the data by using a rotating principal component analysis (PCA), and the Conclusions Section summarizes the key findings of this article.

2. Research Methods

Based on previous research on AAM public acceptance, a survey was developed and distributed through the internet. This survey consisted of six questions rated on a 5-point Likert scale to measure public attitudes toward AAM adoption, namely, openness, usefulness, use, and trust. The survey generated 93 responses, and results completed within 150 s (2.5 min) were considered irresponsible answers; the average survey completion time was 7.6 min.

To identify perceptions of AI in the context of AAM, Kelly et al. [7] illustrated that attitude, perceived usefulness, and trust have a positive impact on use behaviors of AI. Scholars have also developed a theory model called AIDUA to evaluate the predictors of consumers’ AI adoption behaviors, including attitudes [8]. In addition, use frequency is another parameter to measure public perception of AI.

Other research has found that the frequency of AI use has a considerable influence on the intention of AI adoption behaviors, forming a positive feedback loop: the higher the AI use frequency, the higher the score of AI use intention [9]. Therefore, according to the prior literature theory, this study measures public perceptions of AI in the context of AAM from 4 factors, with 7 refined questions to investigate participants’ openness to, perceived usefulness of, use frequency of, and trust of AI. The questions used in this study were the following:

• Openness: I am open to the application of emerging networks and electronic technology (such as artificial intelligence) in my daily life.
• Usefulness: I think the development of artificial intelligence technology will improve my future travel experience.
• Use: I often use voice assistants (such as Xiaoai classmates, Siri, etc.) to assist in daily life.
• Binary: I have used the autonomous driving function of the car (such as Tesla or the intelligent auxiliary driving system of other brands).
• Trust1: (Even if I have not used it) I trust the autonomous driving function of the car.
• Trust2: The driverless model of advanced air traffic (AAM) is reliable.
• Trust3: If AAM is driven by professionals instead of fully autonomous driving, I will feel more assured.

All of the questions listed above are five-point Likert scales, except for the fourth, which is a binary question. The first question measures the openness of AI, the second question is set to measure the usefulness of AI, and the third question examines the use frequency of AI. The binary question is only relevant to the way the latter question is described and not to the content being measured. From the fifth to the seventh, the questions concentrated on trust in autonomous driving. All questions remained for the following analysis, except for the binary question.

3. Results and Analysis

3.1. Demographic Results

The proportions of respondents were 44% male and 56% female, which did not indicate an obvious gap. People aged 19–24 years old are the largest age group, which might be because the sample is students; the second largest age group is from 25 to 34, which is 25% of the whole sample, and the age group from 45 to 54 accounts for 17% of the sample. This survey did not target individuals under 18 years and over 65 years of age.

Half of the participants were from Beijing, China, and participants from Hunan, China, accounted for approximately a quarter of the total. In third place were participants from abroad. It should be noted that location refers only to the current residence rather than the native place.

Educational attainment was notably high among the sample, with 59.09% holding a bachelor’s degree and 30.68% having postgraduate qualifications. Only 1.14% reported high school as their highest level of education, and none had less than a high school education.

Regarding second language proficiency, the sample demonstrated a diverse range of abilities. The largest groups were those with advanced skills (26.14%) and proficiency (27.27%), while 26.14% were unfamiliar with a second language. Only 3.41% of the sample were monolingual.

3.2. Descriptive Result of AI Perceptions

A descriptive statistical analysis was conducted to better understand the sample’s perception distribution.

Table 1. Descriptive statistics of AI questions.

Variables	Mean	Standard Deviation	Skewness	Kurtosis
Openness	1.904	0.734	0.485	0.015
Usefulness	1.915	0.876	0.659	−0.318
Use	2.649	1.224	0.203	−0.904
Trust1	2.596	0.954	−0.166	−0.202
Trust2	2.553	0.863	0.294	0.296
Trust3	2.202	0.85	0.134	−0.189

lists these details.

The analysis of statistical measures provided significant insights into how respondents rated the various categories. The mean scores revealed that the use category, with an average of 2.649, was rated the highest, indicating that the participants found this aspect to be favorable. In contrast, the categories of openness and usefulness had the lowest mean scores of 1.904 and 1.915, respectively, suggesting that respondents perceived these aspects as less satisfactory, potentially reflecting a perception of limited openness and usefulness. The categories Trust1 and Trust2 had mean scores in the middle range at 2.596 and 2.553, respectively, indicating that trust was viewed moderately positively, though with room for improvement.

The standard deviation data provided further insights, particularly regarding the variability of the responses. The usefulness category, with a standard deviation of 1.224, showed the greatest variability in responses, suggesting that while some respondents rated the use aspect highly, others rated it lower. This contrasts with the lower standard deviations observed in openness (0.734) and Trust2 (0.863), which suggests that responses in these categories were more consistent, with respondents agreeing with their assessments of openness and trust.

The skewness values offer another interpretation layer. Both openness and usefulness exhibit positive skewness at 0.485 and 0.659, respectively, indicating that more respondents gave lower scores (closer to 1), although a few respondents rated these categories higher, causing the distribution to shift slightly to the right. Conversely, Trust1 exhibited negative skewness at −0.166, implying that respondents tended to give higher ratings for this category, demonstrating a more positive inclination toward trust. The skewness for use is minimal (0.203), reflecting a symmetrical distribution of responses without a strong tendency toward either lower or higher ratings.

The kurtosis values further clarified the nature of the distributions. The use category, with a kurtosis of −0.904, indicated a flatter distribution of responses, suggesting that values were more evenly spread out with fewer extremes. In comparison, Trust2 (0.296) and openness (0.015) have kurtosis values close to zero, indicating normal distributions with neither extreme outliers nor excessive clustering around the mean. Usefulness (−0.318) shows a slightly flatter distribution, which aligns with the higher standard deviation, reflecting more dispersed responses.

Overall, the data revealed important differences between the categories. Openness and usefulness were rated low and demonstrated positive skewness, indicating that the respondents gave lower ratings, although a few outliers provided higher scores. In contrast, the use category exhibited a higher mean and greater variability in responses, with a more balanced distribution.

The trust categories, particularly Trust1, had moderate ratings, with a tendency towards more positive evaluations, as reflected by its negative skewness. The kurtosis values indicate that the responses to use were more spread out, while the trust categories exhibited more normal distributions. In summary, the analysis suggests that while the use category was rated more favorably with greater variability, openness and usefulness were perceived as lacking, and trust was rated moderately, with respondents leaning toward neutral or slightly positive trust levels. Figure 1 shows the distribution of answers to these six questions, making them more visualizable.

As the histogram shows, most participants had positive perceptions of AI regarding the factors of openness, usefulness, and trust-related questions. None of the questions showed a negative trend for AI perception.

An analysis of the histograms revealed distinct patterns across various categories of participant responses. In terms of the openness distribution, most respondents rated this aspect as a 2, indicating a neutral or slightly closed attitude, with very few high ratings (4 or 5), reflecting overall low openness. Similarly, in the usefulness category, responses clustered around 2, with only a small number of higher scores, suggesting that participants did not perceive the subject as particularly useful.

In contrast, the use distribution displayed a wider spread, peaking around 3, with a greater number of high ratings (4 and 5), indicating more variability and that some respondents found the use aspect favorable. The Trust1 and Trust2 categories peaked at 3, signifying moderate trust, although Trust2 had slightly lower ratings (1 and 2). The Trust3 distribution also peaked at 3 but exhibited a broader spread, with both lower and higher ratings, reflecting more diverse opinions on this aspect of trust.

The key findings highlight low ratings for both openness and usefulness, with most responses concentrated in the 1–2 range, implying limited openness and perceived usefulness. In contrast, use shows a more favorable distribution, with higher ratings suggesting some appreciation of this aspect. Responses related to trust indicate general neutrality, particularly with Trust1, Trust2, and Trust3, which reflect a more varied range of opinions.

Overall, the results suggest that while the use aspect is positively received by some, openness and usefulness are viewed as insufficient, and trust remains moderate, with caution or neutrality prevailing among the respondents.

3.3. Correlation Analysis

The heatmap of Pearson correlation coefficients, combined with significance testing (p ≤ 0.05), provides detailed insight into the relationships between the survey categories. By categorizing the correlations into strong (over 0.5), moderate (between 0.3 and 0.5), weak (less than 0.3), and no correlation, we can more precisely interpret the relationships between various aspects of the survey.

Figure 2 shows the correlations among these variables.

3.3.1. Strong Correlations (Over 0.5)

Openness and usefulness displayed the highest correlation, with a coefficient of 0.69, indicating a strong positive relationship. This suggests that respondents who rated openness highly were also more likely to find the subject to be useful. The strength of this correlation points to the consistent perception that openness and usefulness are linked constructs in respondents’ minds. The overlap in these two dimensions could reflect a broader trend in how openness toward the subject correlates with its perceived utility, meaning that individuals who felt more open to the subject were also more inclined to find it beneficial.

Similarly, Trust1 and Trust2 had a Pearson correlation coefficient of 0.63, which is another strong positive relationship. This demonstrates a relatively high level of consistency in how respondents evaluated trust across these two categories. Trust in one area of the survey was highly predictive of trust in another, which could imply that trust is a global perception for respondents, extending across different facets of the subject. This strong correlation between the trust variables highlights that trust, once established in one domain, tends to propagate across other related areas.

3.3.2. Moderate Correlations (Between 0.3 and 0.5)

Several moderate correlations were found between variables. First, Openness and Trust1 (0.50) showed a moderately positive relationship, indicating that participants who were more open to the subject tended to express more trust. While this correlation is not as strong as between openness and usefulness, it nonetheless suggests that openness fosters a moderate level of trust.

Similarly, usefulness and Trust1 (0.54) also exhibited a moderate-to-strong relationship, further supporting the idea that participants who found the subject useful were more likely to trust it. These moderate correlations between trust and both openness and usefulness suggest that the more positively respondents perceived the subject in terms of its utility and openness, the more trust they extended toward it.

Another moderate correlation was observed between usefulness and use (0.42). This relationship indicated that those who found the subject useful were moderately more likely to appreciate its application or usage. However, the correlation does not suggest that all respondents who found the subject useful saw its use positively, implying that usefulness and practical application, while related, were distinct enough in the respondents’ perception that other factors may influence how use is evaluated.

Openness and use (0.26) indicate a weaker end of the moderate category. This suggests that, while openness and use are related, openness does not strongly predict how respondents feel about the practical use of the subject. This weak correlation implies that, while some overlap exists between respondents’ openness to the subject and their evaluation of its use, these are independent dimensions in terms of how they are perceived.

3.3.3. Weak Correlations (Less than 0.3)

The category of use demonstrated weak correlations with the other categories. The relationship between use and Trust1 (0.29) reflects that, while there is some connection between how respondents rated use and their level of trust in the subject, it is weak. This weak relationship suggests that the practical application of the subject, as perceived by the respondents, does not heavily influence their level of trust in it.

Other weak correlations can be observed between openness and Trust2 (0.37) and usefulness and Trust2 (0.39). These figures suggest that openness and usefulness moderately influence trust in the second trust-related category, but the relationship is not as strong as in Trust1. This distinction may indicate that the second trust category reflects a slightly different aspect of trust that is not directly linked to openness and usefulness.

3.3.4. No Correlation

The blank areas in the heatmap, especially involving Trust3, indicate that there was no significant correlation between Trust3 and any of the other variables, such as openness, usefulness, use, or Trust1 and Trust2. This lack of correlation suggests that Trust3 measures a distinct aspect of trust that is unrelated to the other dimensions evaluated in the survey. This may indicate that respondents viewed this trust-related question as capturing a separate or more nuanced aspect of trust that does not align with their perceptions of openness, usefulness, or use.

Similarly, the blank space between use and Trust2 highlights no significant relationship between these two variables. This suggests that the respondents’ perceptions of the use aspect and their trust in the subject (as measured by Trust2) are independent. The practical application of the subject does not appear to influence respondents’ trust in the subject, at least in terms of how Trust2 was measured.

3.3.5. Overall Interpretation

The heatmap provided valuable insights into the relationships between different survey categories. The strongest correlations were observed between openness and usefulness, as well as between Trust1 and Trust2, indicating close alignment between openness and perceived usefulness and consistency in trust-related responses across different trust dimensions. Moderate correlations between usefulness and both trust and use aspects suggest that, while usefulness influences trust and use, these relationships are not as dominant.

Weak correlations involving use imply that practical application is not a key driver of trust or openness in the respondents’ minds, and the blank areas in the heatmap reveal that Trust3 and use in relation to Trust2 are independent of other survey dimensions. This lack of correlation suggests that respondents may view Trust3 and some aspects of trust as distinct from their overall evaluations of openness, usefulness, and use.

In conclusion, the analysis shows that while certain dimensions, such as openness, usefulness, and trust, are strongly interrelated, practical application (use) and more nuanced trust aspects (Trust3) are less connected to these dimensions. This suggests that improving respondents’ perceptions of openness and usefulness could enhance their overall trust, but distinct strategies may be needed to address more practical aspects of use and other specific trust-related concerns.

4. Factor Analysis

This section describes the factor analysis of AI perception and explores the main exploratory factors using varimax-rotated principal component analysis. Principal component analysis is widely applied in the social sciences as a method of explanatory analysis [10]. Other researchers have also applied PCA to query social acceptance on a certain topic, which shows the possibility of using PCA in public perception [11]. In conducting the factor analysis, the aim was to explore the underlying dimensions of AI perception using two, three, and four components through varimax-rotated PCA.

The four-factor solution resulted in items that did not load onto the fourth factor. The three-factor solution produced significant loadings ranging from 0.45 to 0.9. If we consider a cut-off loading point of 0.4, openness does not load on any factor, and one factor contains only one item, namely, trust3. In the two-factor solution, we obtain similar results such that use does not load on any factor, and one factor still has one loading, namely, trust3.

For the two-component analysis, openness (0.47), usefulness (0.51), Trust1 (0.49), and Trust2 (0.42) had significant loading on PC1. Based on previous studies, these values were considered moderate. This suggests that these variables reflect how respondents understood the questions. This means that these questions seem to measure the same factor, forming a collective perception of respondents’ interpretation, which encompasses usefulness, trust, and openness as one construct measuring some perceptions of AI in AAM. The loadings of items on PC1, combining their correlation results (p = 0.69, significant < 0.05), may indicate that respondents who were open to AI technology may find it useful.

Moreover, it seems that trust plays a key role in this component. However, this overlap between trust, openness, and usefulness introduces interpretive challenges because these dimensions are typically conceptually distinct. Their loadings on the same factor reduce the clarity of PC1, as it suggests that respondents may not perceive trust, openness, or usefulness as independent factors when evaluating AI.

The second principal component (PC2) was dominated by Trust3, with a strong factor loading of 0.91. This suggests that Trust3 represents a distinct dimension that is potentially linked to specific experiences or perceptions of AI technology in practice. However, no other variables loaded strongly on PC2, thereby isolating Trust3 from a broader set of variables. This isolation limits the interpretability of PC2 as it fails to reveal meaningful connections between Trust3 and other aspects of AI perception, such as openness or usefulness.

The limitations of the data stem from the survey’s inability to capture the distinct factors. While Trust3 forms a distinct dimension, PC1 encompasses four constructs (openness, usefulness, and trust); hence, we cannot conclude the effectiveness of the survey. Other studies have shown that these dimensions exhibit distinct factor loadings, reflecting their theoretical underpinnings.

Factor analysis using three principal components resulted in two factors with two item loadings, one factor with one item loading, and one item (openness) that did not load on any factor. However, this solution did not provide sufficiently clear results. Trust1 and Trust2 loaded strongly on the first component, suggesting that general trust in AI is a critical factor in the respondents’ perceptions. The second component, dominated by Trust3 only, reflects a specific experiential trust dimension that is distinct from general trust. The third component captured the practical use and usefulness of AI, with use and usefulness loading onto this component, indicating that real-world applications play a significant role in shaping views on AI.

While better results were obtained from the two-component solution, the three-component analysis suggests that trust has two perspectives: experience and general. Moreover, two factors have two item loadings that are significantly indicative of three potential factors that can describe and explain the perception of AI in AAM. This preliminary exploratory study reveals the potential existence of three factors in our context, namely, perceived use benefit, experiential trust, and general trust, and provides a clear direction for future research on AAM.

5. Conclusions

This study explored public perceptions of advanced air mobility (AAM) in relation to artificial intelligence (AI), focusing on key dimensions such as openness, usefulness, practical use, and trust. By conducting a literature review and developing a comprehensive questionnaire, this study aimed to understand how the public perceives the adoption of AI-driven AAM technology. The survey gathered data using a five-point Likert scale and open-ended questions to assess public attitudes, resulting in 93 valid responses.

The analysis indicated that, while respondents generally viewed the use of AI favorably, there were notable gaps in perceived openness and usefulness. A statistical analysis of the data revealed that trust in AI (particularly in autonomous driving functions) was moderate, with more diverse views observed for different aspects of trust, particularly Trust3, which focuses on the use of AI with human operators. The results highlight a stronger positive correlation between openness and usefulness, suggesting that individuals who are open to AI tend to recognize its benefits more readily. However, practical experience with AI and trust in AI systems did not show a significant relationship, indicating that more efforts need to be made to build public trust in the practical application of AI in air mobility.

Factor analysis was also performed using two, three, and four principal components to further explore these dimensions. However, none of these models provided sufficiently clear results. The two-component analysis showed an overlap between trust, openness, and usefulness, while the three-component analysis, despite better differentiation, still revealed considerable overlap, particularly among trust dimensions. Even the four-component solution failed to produce clear, interpretable distinctions between factors. Overall, while the study identified key areas of public perception, the complexity and interrelated nature of trust, openness, and practical use suggest that further research is needed to better understand how these perceptions influence AI adoption in AAM technology.