Exploring AI literacy, attitudes toward AI, and intentions to use AI in clinical contexts among healthcare students in Korea: a cross-sectional study

Abstract

Background

As artificial intelligence (AI) becomes increasingly integrated into healthcare, AI literacy has emerged as an essential competency for future healthcare professionals. However, research exploring healthcare students’ AI literacy remains limited. This study aimed to assess AI literacy among healthcare students and examine its associations with attitudes toward AI, intention to use AI in clinical contexts, and students’ characteristics (interest in AI and previous AI training) within a Korean context.

Methods

A cross-sectional survey was conducted at a Korean medical school. Of 690 invited students, 91 (67 medical, 24 nursing) participated. Variables were measured using the Scale for Non-Expert AI Literacy (SNAIL), the General Attitudes toward Artificial Intelligence Scale, and intention to use AI items adapted from the Technology Acceptance Model. Analyses included confirmatory factor analysis, descriptive statistics, t-tests, and correlation analyses.

Results

After removing seven items, the validated 24-item SNAIL-KR was used for analysis. Students demonstrated slightly below-average AI literacy, with technical understanding scoring the lowest. AI literacy was positively correlated with positive attitudes toward AI and intention to use AI. Interest in AI was more strongly correlated with AI literacy than prior AI training. No significant differences were found by gender, year of study, or major.

Conclusion

The findings highlight the urgent need to incorporate AI literacy into healthcare curricula in a consistent and systematic manner. Fostering healthcare students’ positive attitudes toward and interest in AI is also crucial for enhancing AI literacy.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12909-025-07766-8.

Introduction

The use of artificial intelligence (AI) in healthcare is significantly transforming patient care, medical research, and health systems [1]. The recent release of ChatGPT by OpenAI has further expanded access to AI technologies, accelerating their integration across various healthcare domains. AI is now being applied in numerous medical disciplines, including nursing, to support clinical decision-making and enhance healthcare delivery [2–4]. As AI becomes increasingly embedded in healthcare systems, it is essential that future healthcare professionals develop a certain level of AI competencies that enable a critical appraisal of the capabilities and limitations of AI applications [3, 5, 6]. These competencies are often referred to as AI literacy [6]. Although the importance of AI literacy for healthcare students is widely acknowledged, empirical efforts to investigate their current level of AI literacy remain limited.

AI literacy is defined as “a set of competencies enabling individuals to critically evaluate AI technologies, communicate and collaborate effectively with AI, and use AI as a tool online, at home, and in the workplace“ [7]. In other words, similar to traditional literacies such as reading, writing, and numeracy, AI literacy refers to the competencies required by non-experts—such as individuals who are not AI specialists or application developers—to understand, use, monitor, and critically reflect on AI technologies in the digital world [7, 8]. With growing interest in AI, an increasing number of studies related AI have been conducted in healthcare education. However most existing research has focused on conceptual discussions [9–12], or learners’ attitudes and emotional responses toward AI [3, 5, 13–16]. Only a few studies have assessed competency-based AI literacy among healthcare students.

Assessing AI literacy can provide valuable insights for developing effective AI curricula in healthcare education. Several initiatives have recently been undertaken in this area. For example, Karaca et al. [17] evaluated Turkish medical students’ AI readiness using the Medical Students’ Artificial Intelligence Readiness Scale (MAIRS), reporting a mean score of 3.92 out of 5. This 22-item scale includes four subscales: cognition, ability, vision and ethics. Similarly, Abolshamat et al. [18] and Tung and Dong [19] applied the same scale to in the contexts of Saudi Arabia and Malaysia respectively, reporting lower mean scores of 2.26 and 3.05. In nursing education, Sumengen et al. [20] explored Turkish nursing students’ AI literacy using the Artificial Intelligence literacy Scale developed by Wang et al. [21], and reported a mean score of 4.87 out of 7. This 8-item scale includes four subscales: awareness, usage, evaluation and ethics. However, these scales present AI -related knowledge, skills and attitudes in relatively generic terms, which may limit their usefulness in developing targeted educational interventions [6]. In contrast, Laupichler et al. [6] assessed German medical students’ AI literacy using the scale for the assessment of non-experts’ AI literacy (SNAIL), reporting a mean score of 3.76 out of 7. This scale comprises three subscale-technical understanding, critical appraisal, and practical application-each reflecting distinct domains of AI competencies. These findings collectively suggest that healthcare students possess relatively low levels of AI literacy.

In Korea, while the integration of AI into medical education is gaining momentum, no studies to date have specifically examined AI literacy among healthcare students. Assessing AI literacy in this population can serve as a needs assessment to identify competency gaps and inform educational planning. However, research providing such foundational data remains scarce. Therefore, further investigation is urgently needed to evaluate the current state of AI literacy among healthcare students and to better prepare them to become competent and critical users of AI in clinical practice.

In addition to exploring AI literacy, examining students’ attitudes toward AI and their intention to use AI in clinical contexts may provide deeper insights into their readiness for AI-integrated healthcare environments. Attitudes toward AI have been shown to influence AI acceptance [13, 14, 20], and emerging evidence suggests a positive association between AI literacy and AI acceptance [6]. Accordingly, students with higher AI literacy are expected to hold more positive attitudes and exhibit a stronger intention to use AI in clinical contexts. While some studies have explored the relationship between attitudes toward AI and intention to use it [3], limited research has explored how these factors specifically relate to AI literacy, particularly in the context of healthcare education.

Thus, this study aimed to explore AI literacy, attitudes towards AI, and the intention to use AI in clinical contexts among medical and nursing students at a Korean medical school. It was conducted as a preliminary study prior to a larger-scale study involving a more diverse sample. AI literacy was assessed using SNAIL developed by Laupichler et al. [6]. This study had five objectives. First, it examined the validity of the SNAIL instrument, which remains in the early stages of development and has not yet undergone extensive validation in diverse educational settings, including Korea. Second, it investigated students’ overall AI literacy levels as well as their scores across the three subscales: technical understanding, critical appraisal, and practical application. Third, it assessed students’ attitudes toward AI, their intention to use AI in clinical settings and individual characteristics such as interest in AI and prior AI training. Additionally, it examined how AI literacy and attitudes toward AI varied depending to the extent of participants’ previous AI training. Fourth, it explored the relationships among AI literacy, attitudes toward AI, intention to use AI and individual characteristics. Finally, it analyzed differences in these variables based on gender, year of study, and major. By identifying current levels of AI literacy and examining the relationships among these variables, the findings from this study aim to inform curriculum development and educational policy, ultimately contributing to the effective integration of AI into healthcare education and practice.

Methods

Participants and procedure

This cross-sectional study was conducted at a university in Korea, employing a convenience sampling method. An invitation link to participate in the survey was distributed to all students enrolled in the College of Medicine and the Department of Nursing through each class’s official social media communication channels. The target population comprised 290 students from the College of Medicine, ranging from first-year pre-medical students to fourth-year medical students, and 400 students from the Department of Nursing across all four academic years. An a priori power analysis was conducted using G*Power (version 3.1.9.7) to determine the minimum required sample size for correlation analysis. Assuming a two-tailed test, an alpha level of 0.05, a power of 0.80, and a medium effect size (ρ = 0.30), the minimum sample size was calculated to be 84. However, a larger number of students were invited to participate, considering the generally low response rates for online surveys and the ongoing collective leave of absence among Korean medical students. Participation was voluntary and anonymous. The online survey was administered over a two-week period, and no financial incentives were provided. In Korea, the college of medicine follows a six-year curriculum consisting of two years of pre-medical education followed by four years of medical education. The Department of Nursing operates a four-year undergraduate program. Both programs offer elective courses related to AI, which focus on conceptual understanding of AI and statistical techniques for data analysis.

Tools

The questionnaire consisted of 60 items divided into four sections. The first section collected demographic and background information, including gender, age, major, year of study, interest in AI, and prior AI training. Interest in AI was measured using a 7-point Likert scale, with higher scores indicating greater interest. Prior AI training was assessed through a single item with four response options: (1) hardly any training, (2) less than 30 h of coursework or learning through other sources, (3) more than 30 h of coursework or learning through other sources, and (4) more than 120 h of intensive AI study or coursework. The second section measured AI literacy using the Scale for Non-Expert AI Literacy (SNAIL) developed by Laupichler et al. [6, 22]. This scale is designed for individuals without formal education in AI or computer science and comprises three subscales: technical understanding (T, 14 items), critical appraisal (C, 10 items), and practical application (P, 7 items). The technical understanding subscale focuses on theoretical and technical AI competencies, such as understanding machine learning models, distinguishing between narrow and strong AI, and recognizing the interaction between computer sensors and AI. The critical appraisal subscale focuses on the recognition of the importance of data privacy, security, and ethical issues related to AI use. The practical application subscale focuses on examples of AI-supported technical applications and the use of AI in daily life. All items in this section were rated on a 7-point Likert scale ranging from 1(strongly disagree) to 7 (strongly agree).

The third and fourth sections measured general attitudes toward AI and the intention to use AI in clinical contexts. General attitudes were assessed using the General Attitudes toward Artificial Intelligence Scale (GAAIS) developed by Schepman and Rodway [14], which includes 12 items measuring positive attitudes (e.g., perceived benefits, positive emotions) and 8 items measuring negative attitudes (e.g., concerns, negative emotions). Although originally presented on a 5-point scale, all items were adapted to a 7-point Likert scale for consistency. Negative items were reverse-scored so that higher scores reflected more positive attitudes. This scale has been validated in Korean contexts [13, 23]. The intention to use AI in clinical contexts was measured using three items adapted from the Technology Acceptance Model (TAM) [24]. TAM is a widely recognized models for predicting users’ willingness to adopt new technology. It was originally developed to identify the cognitive and psychological factors that influence users’ behavioral intention regarding technology adoption. In this study, the original items (e.g., “I intend to use the system in the near future”) were modified by replacing “system” with “AI” and specifying the contexts as “in clinical contexts.” Responses were recorded on a 7-point Likert scale. These adapted items have been used in previous healthcare studies to assess the intention to use AI [3, 13]. Thus, the second, third, and fourth sections of the survey employ scales that have been published in prior research.

The SNAIL scale was translated into Korean through a multi-step process. Initial translation was performed by a professor with a degree from an English-speaking country and over five years of residency in that country. Back-translation was conducted by a bilingual individual with more than ten years of education in an English-speaking country and a master’s degree. The translated items were reviewed by the study’s author and a professor teaching AI-related courses in the university. Two medical students participated in a pilot test to ensure clarity, after which the final version was confirmed. The GAAIS and the intention to use AI in clinical contexts scales had already been translated into Korean in other studies [3, 13, 23], and these versions were used in the present study.

Statistical analysis

Statistical analyses were conducted using IBM SPSS Statistics and AMOS version 27. A significance level of p <.05 was applied to all tests. First, a confirmatory factor analysis (CFA) was performed to assess the validity of SNAIL. Model fit, convergent validity, and discriminant validity were evaluated. Discriminant validity was assessed using the heterotrait–monotrait (HTMT) ratio of correlations, as proposed by Henseler et al. [25]. The HTMT values were calculated using an online calculator provided by the authors (https://www.henseler.com/htmt.html). Internal consistency for all scales was assessed using Cronbach’s alpha. Participant characteristics were analyzed using frequency analysis and the means and standard deviations of all variables were calculated using descriptive statistics. Independent t-tests were conducted to compare variables across gender, major, and year of study. To examine group differences based on year of study, medical students were divided into two groups: low (pre-med 1 to med 1) and high (med 2 to med 4). For nursing students, group comparisons by year of study were not conducted due to the small sample size. Pearson’s correlation coefficients were calculated to examine associations among variables. Additionally, Kendall’s tau coefficient was used to assess the correlation between AI training (an ordinal variable) and other variables.

Results

Participant characteristics

Of the 690 students invited to participate, 103 medical students and 53 nursing students responded to the survey. However, 36 medical students and 29 nursing students were excluded due to incomplete responses. Consequently, data from 91 participants (67 medical and 24 nursing students) were included in the final analysis. Among the medical students, there were 33 males and 34 females, with ages ranging from 19 to 33 years (M = 23.53, SD = 2.36). The distribution by year of study was as follows: 13 first-year pre-medical students, 3 s-year pre-medical students, 18 first-year medical students, 6 s-year medical students, 15 third-year medical students, and 12 fourth-year medical students. The mean interest in AI among medical students was 3.81 out of 7 (SD = 1.65). In terms of prior AI training, 29.9% (n = 20) reported having hardly any prior experience, while 50.7% (n = 34) reported less than 30 h of coursework or learning through other sources. Additionally, 12% (n = 8) had more than 30 h, and 1.5% (n = 1) reported more than 120 h of AI study.

For the nursing students, participants ranged in age from 18 to 29 years (M = 22.46, SD = 2.43), with 4 males and 20 females. Year of study distribution was as follows: 4 first-year, 7 s-year, 8 third-year, and 5 fourth-year students. Their mean level of interest in AI was 4.13 (SD = 1.33). Regarding prior AI training, 37.5% (n = 9) reported hardly any prior experience, while 62.5% (n = 15) indicated having received less than 30 h of coursework or learning through other sources.

Figures 1 and 2 present students’ interest in AI and their previous AI training levels among all participants. The overall mean interest in AI among the 91 participants was 3.89 (SD = 1.57), and 85.17% reported having received less than 30 h of coursework or learning through other sources.

Fig. 1.

Students’ interest in AI (N = 91)

Fig. 2.

Students’ previous AI training levels (N = 91)

SNAIL validation

Prior to data analysis, SNAIL underwent validation for internal consistency and construct validity using confirmatory factor analysis (CFA). Examination of internal consistency revealed that removing items P4 and T13 improved reliability. Additionally, CFA indicated that the factor loadings for these items were 0.499 and 0.522, respectively. Although T13 slightly exceeded the commonly accepted threshold of 0.50, its removal was justified due to its low contribution to internal reliability. Therefore, both P4 and T13 were excluded from further analysis. Moreover, modification indices greater than 10 and cross-loading issues were observed for items T1, P7, C2, C5, and C8. These items were also removed. After excluding these seven items (Appendix 1), the final version of the scale (SNAIL-KR) consisted of 24 items: 5 for practical application (P), 7 for critical appraisal (C), and 12 for technical understanding (T). All subsequent analyses were conducted using this revised version.

The final model demonstrated acceptable fit indices: χ²/df = 1.716 (df = 242, p <.001), satisfying the criterion of < 3.0 [26]. The standardized root mean square residual (SRMR) was 0.067, and the root mean square error of approximation (RMSEA) was 0.089, both meeting the acceptable threshold of < 0.10 [27]. The comparative fit index (CFI) was 0.907, exceeding the recommended minimum of 0.90. Convergent validity was assessed using standardized factor loadings, significance levels, composite reliability (CR), and average variance extracted (AVE) [28, 29]. All standardized factor loadings exceeded the 0.50 threshold. The unstandardized coefficients were above 0.196 and statistically significant (Appendix 2). CR values for all constructs exceeded 0.70, and AVE values were above 0.50, indicating adequate convergent validity. Discriminant validity was evaluated using the Heterotrait–Monotrait (HTMT) ratio of correlations, and all HTMT values between constructs were below the 0.85 threshold, indicating adequate discriminant validity. Table 1 presents the CR, AVE, and HTMT values for each construct.

Table 1.

CR, AVE and HTMT values of the revised model

			HTMT
	CR	AVE	T	C	P
T	0.951	0.620
C	0.933	0.666	0.473
P	0.877	0.588	0.535	0.823

Internal consistency and descriptive statistics

Table 2 presents the means, standard deviations, and internal consistency (Cronbach’s α) of all variables. All scales demonstrated good internal consistency, with Cronbach’s α values exceeding 0.83. No items were identified that would improve reliability if removed. Both medical and nursing students exhibited slightly below-average levels of AI literacy, with mean scores of 3.76 and 3.89, respectively. Among the subscales of AI literacy, technical understanding yielded the lowest scores, while critical appraisal had the highest scores in both groups. Regarding general attitudes toward AI, both medical and nursing students showed higher scores for positive attitudes compared to negative attitudes. The intention to use AI in clinical contexts was moderately high in both groups.

Table 2.

Means, standard deviations, and internal consistency of variables

	T	C	P	AL	PR	NR	AT	INT	IN
MS (N = 67)	3.23 (0.17)	4.55 (0.16)	3.93 (0.15)	3.76 (0.14)	4.65 (0.11)	4.20 (0.99)	4.47 (0.66)	4.61 (0.16)	3.81 (1.65)
NS (N = 24)	3.07 (0.25)	4.90 (0.23)	4.43 (0.22)	3.89 (0.19)	4.76 (0.16)	4.21 (1.08)	4.54 (0.59)	4.72 (0.24)	4.13 (1.33)
M (N = 37)	3.22 (1.22)	4.51 (1.11)	3.97 (1.18)	3.75 (0.95)	4.79 (0.97)	4.24 (1.09)	4.57 (0.76)	4.65 (1.24)	3.92 (1.89)
F (N = 54)	3.17 (1.42)	4.73 (1.35)	4.12 (1.27)	3.82 (1.17)	4.60 (0.81)	4.18 (0.97)	4.43 (0.55)	4.63 (1.26)	3.87 (1.33)
LG (N = 34)	3.48 (1.38)	4.66 (1.20)	4.06 (1.28)	3.95 (1.11)	4.69 (0.84)	4.25 (0.90)	4.51 (0.63)	4.66 (1.32)	3.76 (1.76)
HG (N = 33)	2.97 (1.36)	4.43 (1.38)	3.79 (1.25)	3.57 (1.15)	4.61 (0.99)	4.16 (1.09)	4.43 (0.71)	4.56 (1.24)	3.85 (1.56)
Total (N = 91)	3.19 (1.34)	4.64 (1.26)	4.01 (1.23)	3.79 (1.08)	4.68 (0.88)	4.21 (1.01)	4.49 (0.64)	4.64 (1.25)	3.89 (1.57)
Cronbach’α	0.95	0.93	0.88	0.95	0.87	0.83	0.83	0.83

MS medical students, NS nursing students, T technical understating, C critical appraisal, P practical application, M male students, F female students, LG low group, HG high group, AL artificial intelligence literacy, PR positive attitude, NR negative attitude, AT attitude in total, INT intention to use AI in clinical contexts, IN interest in AI

In addition, the mean AI literacy scores and positive attitudes toward AI varied according to participants’ levels of prior AI training as follows: those with hardly any prior experience reported a mean AI literacy score of 3.24 (SD = 1.06) and a means positive attitude score of 4.48 (SD = 0.86); those with less than 30 h of coursework or learning through other sources reported a mean AI literacy score of 3.92 (SD = 0.89) and a mean positive attitude score of 4.65 (SD = 0.83); those with more than 30 h reported a mean AI literacy score of 4.34 (SD = 1.07) and a mean positive attitude score of 5.11 (SD = 0.91). In the case of one participant who reported more than 120 h of AI study, the AI literacy score was 7.00, and the positive attitude score was 6.25.

Correlation analyses

Table 3 presents the results of the correlation analyses. AI literacy and its subscales were all significantly and strongly correlated, suggesting that they all measure a common underlying construct: AI literacy. AI literacy and its subscales were significantly correlated with positive attitudes toward AI. However, AI literacy did not show a significant correlation with negative attitudes. Among the subscales, only critical appraisal demonstrated a significant negative correlation with negative attitudes toward AI. In addition, AI literacy and each of its subscales were significantly associated with the intention to use AI in clinical contexts, interest in AI, and prior AI training. Positive attitudes toward AI were also significantly correlated with the intention to use AI in clinical contexts, interest in AI, and prior AI training. In contrast, negative attitudes were not significantly associated with any of these variables. Finally, the intention to use AI in clinical contexts was significantly correlated with interest in AI but not significantly associated with prior AI training.

Table 3.

Results of correlation analyses

	T	C	P	AL	PR	NR	AT	INT	IN	AIT
T
C	0.447**
P	0.493**	0.746**
AL	0.885**	0.790**	0.792**
PR	0.298**	0.406**	0.514**	0.443**
NR	− 0.106	− 0.210*	− 0.117	− 0.164	− 0.069
AT	0.178	0.201	0.348**	0.260*	0.777**	0.575**
INT	0.254*	0.348**	0.428**	0.375**	0.597**	− 0.036	0.467**
IN	0.325**	0.304**	0.510**	0.424**	0.267*	0.000	0.219*	0.265*
AIT	0.254**	0.320**	0.291**	0.325**	0.186*	− 0.021	0.183*	0.142	0.065

T technical understating, C critical appraisal, P practical application, AL artificial intelligence literacy, PR positive attitude, NR negative attitude, AT attitude in total, INT intention to use AI in clinical contexts, IN interest in AI, AIT prior AI training

**>0.01, *>0.05

Group comparisons

Group comparisons (medical vs. nursing students, male vs. female students, and low vs. high groups) revealed no statistically significant differences across any of the measured variables. The mean differences in Al literacy between medical vs. nursing students, male vs. female students, and low vs. high groups were t(89) = 0.496, p =.621, d = 0.123, t(89)=−0.284, p =.777, d = 0.066, and t(65) = 1.381, p =.172, d = 0.337, respectively.

Discussion

As AI literacy increasingly becomes a fundamental competency for future healthcare professionals, it is imperative that healthcare students are adequately prepared to navigate technological advancements in their future workplaces. To support this goal, assessing the current level of AI literacy among healthcare students is a necessary first step in designing targeted and effective educational interventions. In line with this objective, the present study explored healthcare students’ AI literacy and examined its associations with their attitudes toward AI, their intention to use AI in clinical contexts, and individual characteristics such as prior AI training and interest in AI.

To address this aim, SNAIL was first validated through confirmatory factor analysis in the Korean context. During this process, seven items were excluded, resulting in a final version consisting of 24 items (SNAIL-KR), which was used for the analysis. The results revealed that students demonstrated slightly below-average levels of AI literacy. These findings are consistent with those of Laupichler et al. [6], who reported a comparable level of AI literacy among German medical students, with a total mean score of 3.76 (3.79 in the present study). In their study, critical appraisal also showed the highest mean score (M = 4.89, M = 4.64 in this study) and technical understanding the lowest (M = 2.63, M = 3.19 in this study), demonstrating the same pattern observed in the present study. The relatively higher scores in critical appraisal may reflect increased exposure to media coverage on issues such as data privacy, ethics, and AI-related risks. However, given that the highest subscale score was only marginally above the midpoint of 4, the findings collectively underscore the need for more comprehensive AI literacy education. This interpretation is further supported by the fact that 85.17% of participants reported receiving less than 30 h of AI-related training. Moreover, systematic reviews by Kimiafar et al. [30] and Mousavi Baigi et al. [2] similarly reported that although healthcare professionals and students generally expressed motivation to adopt AI, they had received insufficient training and tended to possess limited knowledge and skills in using AI. These findings reinforce the present study’s results, highlighting the urgent need to strengthen AI literacy education to better prepare future healthcare professionals.

In addition, participants overall exhibited positive attitudes toward AI. However, the mean score for negative attitudes (reverse scored) was also slightly above the neutral midpoint. This suggests a balance between favorable and cautious perspectives—students recognized the potential benefits of AI while also acknowledging its current limitations. This result is consistent with the findings of Seo and Ahn [23], who examined the attitudes of 230 Korean nursing students using the same scale. On a 5-point Likert scale, the mean for positive attitudes was 3.69, while that for negative attitudes was 3.07, reflecting a pattern similar to that observed in the present study. Similarly, Gordon et al.’s [1] scoping review on attitudes toward the application of AI in clinical medicine reported the coexistence of support for and concerns about AI among healthcare learners. Such concerns may influence learners to adopt a cautious stance toward AI applications. Collectively, these findings highlight the importance of directly addressing both the opportunities and concerns related to AI in future education and training programs.

Furthermore, the findings indicate that students with higher levels of AI literacy are more likely to hold positive attitudes toward AI—and vice versa. Among the subcategories, practical application showed the strongest correlation with positive attitudes, suggesting that students who view AI favorably are more inclined to engage with it in daily life. Critical appraisal was positively associated with positive attitudes and negatively associated with negative attitudes, with a notably stronger correlation for positive attitudes. This implies that the more students understand both the benefits and potential concerns of AI, the more likely they are to adopt a positive perspective. In addition, a positive association was observed between positive attitudes toward AI and prior AI training. Collectively, these results highlight the importance of structured AI literacy education that addresses both the advantages and potential risks associated with AI.

Regarding the intention to use AI in clinical contexts, participants reported scores slightly above the neutral midpoint. This intention was significantly associated with both AI literacy and positive attitudes toward AI. These findings demonstrate that AI literacy is a critical factor influencing students’ intention to use AI in clinical practice, while also confirming previous research showing a positive association between positive attitudes toward AI and the intention to use it [1, 13].

Additionally, both AI literacy and positive attitudes toward AI were significantly correlated with students’ interest in AI and their prior AI training, with stronger correlations observed for interest in AI than for prior training. Interest in AI was also significantly associated with the intention to use AI, whereas prior AI training did not show a significant correlation with this variable. Furthermore, among the subcategories of AI literacy, practical application exhibited a notably stronger correlation with students’ interest in AI than with their prior AI training. This suggests that students with higher interest in AI are more likely to engage with AI technologies in their daily lives, thereby reinforcing their AI literacy and fostering positive attitude AI. The influential role of interest in AI on AI literacy aligns with findings by Laupichler et al. [6], who reported that students’ interest in AI was more strongly association with AI literacy than previous training experience. Moreover, there was a notable difference in AI literacy between students with hardly any training and those with less than 30 h training or more than 30 h of training, with mean differences of 0.68 and 1.1 point, respectively. These findings highlight the potential impact of AI education and suggest that programs providing more than 30 h of training may be necessary to meaningfully enhance students’ AI literacy. While the lack of significant association between prior AI training and intention to use AI may be attributable to the small sample size, further research is needed to clarify this relationship. Overall, the findings suggest that fostering interest in AI may be particularly effective in promoting AI literacy and that educational interventions should aim to stimulate engagement and curiosity about AI especially through programs offering more than 30 h of training.

No significant differences were observed by gender, year of study, or major for any of the measured variables in this study. These findings differ from those reported by Laupichler et al. [6], who found that male students scored higher in AI literacy and that academic semester was significantly associated with AI literacy levels. Similarly, Hashish and Alnajjar [5], in a study assessing nursing students’ digital health literacy—defined as perceived competence in using digital health tools—found that senior students demonstrated higher literacy levels. In addition, Kwak et al. [13] reported that senior students exhibited significantly higher positive attitudes toward AI, and Sumengen et al. [20] found that male nursing students showed more positive attitudes toward AI, as both measured by the GAAIS.

This discrepancies between these previous findings and the current results cannot be fully explained in this study. However, one possible explanation for the lack of significant gender difference is that participation in this survey was voluntary, and response rate is low. Thus, students who chose to participate—regardless of gender—may have had a greater interest in AI than the general student population. In addition, the absence of significant differences by year of study may reflect the limited integration of AI-related content into the curriculum. As previously noted, AI education in Korea is typically offered only as elective courses during pre-medical phase, which may contribute to a lack of variation in AI literacy and attitudes across academic years. Alternatively, these nonsignificant results may be attributed to the relatively small sample size. Regarding the lack of significant differences by major, further investigation is needed to explore whether and how AI literacy varies across academic disciplines within healthcare education. Future studies with larger and more diverse samples across multiple disciplines in healthcare education are needed to confirm and extend these findings.

The findings of this study highlight the urgent need to systematically integrate AI literacy into healthcare curricula. Current training opportunities appear insufficient, with most students reporting minimal exposure to AI education. Given the observed increase in AI literacy among students with more than 30 h of AI training, this study highlights the value of structured programs of sufficient length. Training that exceeds 30 h may be especially effective in enhancing both AI literacy and positive attitudes toward AI. In addition, integrating AI content into core curricula—rather than offering it solely as electives—can promote more equitable and comprehensive learning experiences. Educators should also leverage students’ existing interest in AI as a foundation for enhancing AI literacy. Instructional programs should be deliberately designed to spark and sustain students’ interest in AI, for example, by incorporating hands-on activities using AI tools, and interdisciplinary projects that apply AI to solve real-world healthcare problems. Furthermore, given that students may approach AI with a cautious stance toward AI applications, AI literacy programs should explicitly address their concerns. Promoting open dialogue, encouraging critical reflection, and providing exposure to practical applications can help learners develop a balanced and informed perspective on the role of AI in clinical practice.

Despite its contributions, this study has several limitations. First, the sample size was relatively small. The response rate of online survey was very low likely due to an ongoing two-year collective leave of absence among Korean medical students, triggered by political issues. Additionally, the sample was drawn from a single institution in one geographic region, which limits the generalizability of the findings. However, although the participants may not be fully representative of all Korean healthcare students, medical students in Korea tend to form a relatively homogeneous group due to the highly standardized and competitive admission process. Moreover, given the curricular similarities across Korean medical schools, significant variation in AI literacy between institutions may be limited. Future studies with larger and more diverse samples from multiple institutions are essential to validate and expand upon these results. Nonetheless, the present study offers preliminary insights that may serve as a valuable baseline for further investigation into AI literacy and attitudes among healthcare students. In addition, multinational studies could provide useful comparative perspectives on how AI literacy varies across educational and cultural contexts. Longitudinal or intervention-based research examining the development of AI literacy over time may also inform the design of future curricula.

Second, the study relied on self-reported data, which may be subject to response bias and subjective interpretation. To enhance the validity of future findings, objective measures—such as knowledge-based assessments or evaluations of practical AI-related skills—should be incorporated alongside self-report instruments. Finally, the adapted SNAIL-KR scale demonstrated both feasibility and internal consistency. The results obtained using SNAIL-KR were comparable to those reported by Laupichler et al. [6], despite the exclusion of seven items from the original version. Nevertheless, further validation with larger and more diverse samples is needed to confirm its reliability and applicability.

Conclusion

AI literacy is rapidly becoming a core competency for future healthcare professionals, underscoring the urgent need for targeted educational interventions. This study assessed healthcare students’ AI literacy, attitudes toward AI, and intentions to use AI in clinical contexts within a Korean setting. While further research is needed to reinforce and generalize these findings, the results highlight the importance of actively integrating AI literacy—particularly technical understanding—into healthcare education. Fostering students’ positive attitudes toward and interest in AI also appears essential for enhancing AI literacy. These insights provide a foundational basis for developing educational programs that enhance AI literacy and promote competent use of AI technologies in clinical practice. Although this was a preliminary study with a limited sample, it offers a valuable starting point for developing valid and reliable approaches to AI literacy training in healthcare education, especially given the current scarcity of research in this area.

Abbreviations

AI

Artificial intelligence

MAIRS

Medical Students’ Artificial Intelligence Readiness Scale

SNAIL

Scale for the assessment of Non-experts’ AI Literacy

GAAIS

General Attitudes toward Artificial Intelligence Scale

Authors’ contributions

JS designed the study, collected and analyzed data, wrote and revised the manuscript.

Funding

This work was supported by the Dong-A University research fund.

Data availability

The datasets used during the current study are available from the corresponding author on reasonable request.

Declarations

Ethical approval and consent to participate

This study was conducted in accordance with the Declaration of Helsinki, and approved by the Dong-A University Institutional Research Board (IRB approval no. 2-1040709-AB-N-01-202407-HR-040-04). Informed consent was obtained from all participants before they participated in the survey online.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.