AI literacy enables individuals to critically evaluate artificial intelligence technologies and use them effectively, ethically, and safely in educational, work, and personal environments, as well as in various aspects of life. In trying to understand these aspects, different measures of AI literacy have been used, as some of them focused on measuring these competencies in a specific category of the public, such as university students (Kong et al., 2021, 2023). Other studies developed measures for non-experts in computer science (Laupichler et al., 2023a). In contrast, the technical dimension clearly dominated some scales (Pinski and Benlian, 2023). Carolus et al. (2023) provided a broader range of measurements that include the known dimensions of AI literacy and do not neglect the psychological and social aspects as well as self-development and work readiness.

The present study sought to comprehensively review the existing body of literature pertaining to AI literacy scales within the context of higher education. This review was undertaken to identify the key components and commonalities among these measures, ultimately facilitating the selection of the most suitable instrument for a large-scale international survey.

It is clear from the above that AI literacy is a new research area, and the studies that dealt with enhancing AI literacy in higher education focused on developing measures and self-assessment studies for students from different disciplines through a survey or quasi-experimental studies. The studies provided meta-reviews of previous literature in the field of MIL measures such as Schofield et al. (2023), concluding that few of them addressed large populations and demographic and social factors were also minimized in those measures. The study showed that “the goal of measuring the ‘entire’ MIL area seems to be quite ambitious and challenging.” This represents a notable weakness in the studies. It is recommended that several focused issues be highlighted in this regard, including AI.

Montag et al. (2024) confirmed that “the investigation of attitudes toward AI together with AI literacy/competency is critical, timely, and needs to be started in large representative samples around the world.”

Therefore, this study aims to bridge this scientific gap through a large-scale geographical survey to measure the levels of a specific aspect of information literacy, specifically AI literacy, among a sample of university students from four countries in East Asia, the Middle East, and North Africa (India, Malaysia, Saudi Arabia, and Egypt). The Global AI Index 2023, revealed by Stanford University, ranks 62 countries according to their capabilities in AI. India ranked 14th globally on the general scale, Saudi Arabia ranked 31st, Malaysia ranked 44th, and Egypt ranked 52nd among the 62 countries in the index. The ranking of the four countries varied according to the different criteria of the index, as shown in Table 1. Countries are ranked by their AI capacity at the international level. This is the fourth iteration of the Global AI Index, published on 28 June 2023.

The global ranking highlights India’s strengths across several dimensions: talent (second place), operational environment, development, and trade. Saudi Arabia leads globally in governmental strategy (first place) and also performs well in infrastructure and the operational environment. Malaysia stands out for its robust infrastructure, while Egypt has achieved an advanced position in governmental strategy. The findings of this study can provide valuable insights into the extent to which the progress reflected in the overall AI index for the four countries impacts AI literacy among university students.

This study aims to identify the levels of AI literacy as part of MIL among a sample of university students in four countries (Egypt, Saudi Arabia, India, and Malaysia) to determine the extent to which students keep pace with the development in this field, especially since this type of literacy has become seen as a human right. One of the requirements and conditions of sustainable development and the sub-objectives of the study are:

The study tests two main hypotheses:

The study adopted a survey approach to collect data from a sample of university students in four countries in East Asia, the Middle East, and North Africa (India, Malaysia, Saudi Arabia, and Egypt). After obtaining approval from the King Saud University Research Ethics Committee, the online questionnaire was distributed across 10 universities in the four countries from April 20th to May 20th, 2024.

Prior to commencing the questionnaire, participants were required to provide informed consent by agreeing to the terms outlined in the information letter. The letter explicitly stated the voluntary nature of participation, the anonymity of all responses, the exclusive use of data for research purposes, and the participant’s right to withdraw from the study at any time.

To ensure data quality and adherence to study objectives, responses were subjected to a rigorous cleaning process. Those who failed attention checks (n = 241) were excluded from the analysis. Data collection in each country was terminated once the target sample size of 450 respondents was reached. Consequently, the final dataset comprised 1,800 responses from 10 universities across the four participating countries.

The study adopted the Meta AI Literacy Scale (MAILS) developed by Carolus et al. (2023) for the following reasons:

A slight modification was made to the scale. The research team found that using the 11-point Likert scale (0–10) employed in the study by Carolus et al. (2023) would potentially increase the time needed to respond and reduce focus. Therefore, a 6-point scale (0–5) was adopted, starting from zero, meaning “Cannot at all,” and ending at five, meaning “very capable.” Statistically, the outcome will be the same in terms of measuring respondents’ responses.

The scale included 33 statements, with only one omitted, which was intended for attention checks in the original scale. This was done because three other statements served the purpose of attention checks (statements 21, 27, and 29); each statement has the same meaning as the previous statement. After these slight modifications to the scale, reliability and stability tests were applied, confirming the scale’s validity.

The authors conducted Cronbach’s alpha analyses to assess the reliability of the AI literacy scale, as illustrated in Table 2. Cronbach’s alpha values for our modified 6-point scale demonstrated high internal consistency reliability for the AI literacy dimensions, which ranged from 0.732 to 0.961 and were very close to the original 11-point scale.

Furthermore, to evaluate the internal consistency of the AI Literacy scale, the authors employed a Pearson correlation to investigate the linear relationships between the dimensions and the total scale. As illustrated in Table 3, the Pearson correlation results revealed significant positive correlations (r > 0.6). Moreover, a range of coefficients of determination (r²) from 0.37 to 0.79 indicates that even the least correlated dimension accounts for more than one-third of the variation in AI Literacy scores. The robust correlation coefficients suggest that each component contributes to AI Literacy while forming a cohesive entity. Robust internal consistency assures the scale’s reliability and construct validity, emphasizing that the dimensions are closely interrelated and may evaluate similar or overlapping constructs.

The demographic profile, as presented in Table 4, provides a comprehensive summary of the study sample (N = 1,800), emphasizing key features relevant to the research. The sample demonstrates a relatively equal distribution of genders, with a slight majority of female individuals at 51.6%. Significantly, the sample consists of an equal distribution of individuals from four different countries: Egypt, Saudi Arabia, Malaysia, and India, with each country accounting for 25% of the sample. The age distribution is predominantly skewed toward younger individuals, with almost 75% of participants being 25 or younger. Most participants (66.7%) have an academic background in human and social sciences, while a smaller proportion are in sciences and engineering (21.4%) and medical sciences (11.9%). The distribution of study levels ranges from undergraduate to doctoral studies, with the majority (27.7%) in the early stages of their bachelor’s degree. As measured by their grades, the participants’ academic achievement demonstrates a notable focus on the higher categories, with 69.2% of them attaining either “Very Good” or “Good” marks. The demographic makeup of the study’s participants offers a solid basis for examining the factors within various cultural and academic settings.

Data show that the average of the overall scale is 2.98 out of 5, which falls within the category of “Moderately able (Figure 1).” The results of the one-way ANOVA, as shown in Table 5, reveal significant differences in AI literacy scores among the four countries and respondents’ academic grades and specializations. All comparisons show statistically significant results (p < 0.001). Nevertheless, the analysis did not reveal a significant difference based on respondents’ age in artificial intelligence literacy scores (p = 0.179).

Further examination of the country-specific data demonstrates a noteworthy impact on artificial intelligence literacy scores, F(3, 1,796) = 20.33, p < 0.001, η² = 0.033, implying that about 3.3% of the observed variance in these scores may be attributed to national differences. Furthermore, the study revealed that the academic grades F(4, 1,795) = 8.87, p < 0.001, η² = 0.019, indicating that approximately 1.9% of the score variance could be explained by the respondents’ academic grades. This correlation does not necessarily indicate that grades determine AI literacy proficiency.

The most significant impact was detected in Academic Specialization, F(2, 1,797) = 38.74, p < 0.001, η² = 0.041; this effect demonstrated 4.1% of the variation in AI literacy scores.

Further analyses using the Bonferroni post-hoc test for multiple comparisons (Table 6) revealed significant differences among the variables investigated.

Concerning country-specific differences, Malaysian respondents demonstrated significantly higher AI literacy scores than participants from all other countries (p < 0.001). Saudi Arabian and Indian participants scored significantly higher than their Egyptian counterparts (p < 0.01). However, no statistically significant difference was seen in AI literacy scores between Saudi Arabian and Indian respondents. Regarding university students’ academic performance, students classified as having “Poor” grades consistently demonstrated higher levels of AI literacy compared to their peers with “Good,” “Very Good,” and “Excellent” grades (all p < 0.05). Individuals graded “Decent,” “Very Good,” and “Excellent” achieved significantly higher scores than individuals with “Good” grades (p < 0.01) (Figure 2).

Within the variable of Academic Specialization, individuals enrolled in Sciences and Engineering programs had significantly greater proficiency in AI literacy when compared to those studying Human and Social Sciences or Medical Sciences (p < 0.001).

The results of the one-way ANOVA in Table 7 indicate significant variations in AI perception (optimistic/pessimistic views toward AI) among respondents based on country, academic grade, Academic Specialization, and age (all p < 0.05). The effect sizes (η²) indicate that the respondents’ country has the most significant influence on AI perception (η² = 0.106), followed by academic grade with a lesser influence (η² = 0.035). Academic Specialization and age have the most negligible impact. These findings suggest that various demographic and intellectual characteristics may influence individuals’ perceptions of AI, with nationality being the most influential factor.

After conducting post-hoc analyses with Bonferroni correction, the results in Table 8 revealed several significant differences in AI perception across research variables. Regarding respondents’ countries, there were substantial variations among most pairs of countries, with Malaysian respondents reporting the highest perception of AI, followed by Egyptian and Saudi Arabian students. Concerning academic performance, students with “Excellent” grades had significantly higher AI perception than their counterparts who received very good grades, indicating a positive association between academic grades and AI perception. Academic specialization substantially impacted students’ AI perception; students from the humanities/social sciences and science/engineering disciplines had a higher perception of AI than those from medical sciences. Additionally, age differences were significant; respondents aged 25 and younger and individuals over 30 had a greater AI perception than those aged 25–30.

The results of Table 9 indicate significant differences in the desire to use AI based on respondents’ country, F(3, 1,796) = 32.952, p < 0.001, η² = 0.052, suggesting that around 5.2% of the variation in AI desire may be attributed to disparities among countries. Academic grades, F(4, 1,795) = 14.204, p < 0.001, η² = 0.031; this suggests that approximately 3.1% of the variation in respondents’ desire to use AI can be explained by academic performance differences. Furthermore, Academic Specialization, F(2, 1,797) = 14.480, p < 0.001, η² = 0.016, indicating 1.6% of the variability. Lastly, age, with an F-value of 4.35 and a p-value of 0.013. This suggests that around 0.5% of the variation in the desire to use AI can be attributed to differences in respondents’ age.

The post-hoc test in Table 10 revealed significant differences in the desire to use AI based on the respondents’ countries. Malaysian students were significantly more willing to utilize AI than their Egyptian and Saudi Arabian counterparts. Indian individuals exhibited the lowest level of desire (all p < 0.001). Academic grade positively correlates with the desire to use AI; students who achieved “Excellent” grades demonstrated a significantly higher level of desire compared to those who had “Good” grades (p < 0.001) and “Very Good” grades (p = 0.007). In the context of Academic Specialization, students in Sciences and Engineering confirmed a greater inclination toward utilizing AI, surpassing those in the Human/Social Sciences (p = 0.001) and Medical Sciences (p < 0.001) by a wide margin. In addition, respondents from the Human/Social Sciences field expressed a considerably greater willingness than those from the Medical Sciences field (p = 0.004). Regarding the age of students, the group aged 30 and above showed a substantially higher desire to employ AI than those aged 25–30 (p = 0.010).

The findings in Table 11 suggest a slight and insignificant difference in AI Literacy between males (M = 96.86, SD = 32.60) and females (M = 99.71, SD = 30.32), t(1,798) = −1.920, p = 0.055, d = 0.091. As measured by Cohen’s d (d = 0.091), the effect size indicates a minimal practical difference between the groups. Females had a marginally higher average score in AI Literacy, defying certain traditional assumptions in technology-related domains. In terms of AI Perception, there was an insignificant difference between males (M = 4.97, SD = 1.28) and females (M = 5.08, SD = 1.16), t(1,798) = −1.855, p = 0.064, d = 0.087. The effect size is minimal. Furthermore, there was no statistically significant difference in the Desire to Use AI between males (M = 3.55, SD = 1.33) and females (M = 3.60, SD = 1.27), t(1,798) = −0.736, p = 0.462, d = 0.035. The effect size is negligible, indicating that gender has little impact on the willingness to use AI technologies.

Data suggests that Malaysian respondents scored significantly higher than those from other countries. This finding aligns with the Global AI Index 2023, which emphasizes Malaysia’s robust infrastructure, including “basic electricity and internet access, as well as supercomputing capabilities, deep databases, and increasing AI adoption” (tortoisemedia.com, 2023). The Global AI Index also suggests higher AI readiness in Saudi Arabia and India compared to Egypt. However, a country’s overall AI ranking may not directly translate to higher AI literacy within a specific demographic, such as university students, when compared to their counterparts elsewhere.

Therefore, to enhance AI literacy amongst university students specifically, it is essential to develop targeted educational curricula and methodologies. Integrating AI literacy into courses across all disciplines is a promising approach, as advocated by numerous studies in the field (Kong et al., 2021, 2022, 2023; Laupichler et al., 2023a,b,c). These findings offer valuable insights for policymakers and curriculum developers in the educational sector. As Mansoor (2023) emphasizes, approaching information literacy as a “comprehensive and complex concept to achieve educational and development goals” necessitates broader consideration by researchers, professors, and policymakers across media, education, and professional development spheres.

This study relied on the self-assessment method for respondents, which many studies have used to measure AI literacy, such as (Laupichler et al., 2023c; Schofield et al., 2023). However, this does not mean closing the door on other methodological approaches used to measure levels of AI literacy, such as experimental and quasi-experimental methods. The best measurements allow researchers to combine more than one method, mainly if applied to large and diverse samples of respondents.

In summary, the study results should be interpreted within its methodological boundaries, as it was limited to four countries’ university students. This might explain why the age variable could have been more impactful due to the proximity of age levels within the research community. Moreover, the level of AI literacy required from all individuals necessitates surveys and scales encompassing diverse community segments from different ages and educational, economic, and social backgrounds.