AI literacy and ethical attitudes toward AI in medical students: evidence from a structural model

Abstract

Background

The rapid integration of artificial intelligence (AI) into healthcare has raised ethical concerns regarding autonomy, beneficence, non-maleficence, justice, accountability, and human oversight. While AI literacy is increasingly being adopted as a fundamental competency area in medical education, there is limited empirical evidence regarding its relationship with ethical attitudes toward AI. This study was conducted to examine the relationship between AI literacy and ethical attitudes toward AI among medical students and to explore how AI literacy is associated with ethical preparedness.

Methods

This cross-sectional correlation study was conducted with 461 medical students at a state university. Data was collected using a Demographic Information Form, AI-Literacy Scale, and AI-Ethics Scale. Descriptive statistics, independent samples t-tests, one-way ANOVA, Pearson correlation analysis, simple linear regression, and Partial Least Squares Structural Equation Modeling (PLS-SEM) were used to analyze relationships and associations between variables.

Results

A significant positive correlation was found between AI literacy and ethical attitudes toward AI (r = 0.648, p < 0.001). Regression analysis showed that AI literacy was significantly associated with ethical attitudes and explained 42% of the variance (R² = 0.420). Structural model analysis confirmed a strong positive path effect (β = 0.655, p < 0.001; R² = 0.430; f² = 0.753). Students who received formal AI training achieved significantly higher scores in AI literacy and ethical attitudes, while no significant differences were observed between AI usage frequency and academic year. Although AI literacy did not differ by gender, female students reported higher ethical attitude scores.

Conclusions

AI literacy was significantly associated with medical students’ ethical attitudes toward AI. No significant association was observed between frequency of AI use and ethical preparedness in the current study. These findings suggest that AI literacy should be conceptualized not only as a technical skill but as an important component associated with ethical preparedness in AI-enabled healthcare systems.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12909-026-10142-9.

Introduction

Artificial Intelligence (AI) has caused a significant transformation in healthcare with the proliferation of large language models (Large Multimodal Models, LMM). Within this transformation, AI systems play a critical role in medicine in processes such as clinical decision support systems, radiological imaging analysis, genetic research, and the processing of complex data to create personalized treatment plans [1–3]. The rapid integration of AI into medicine has also brought with it ethical issues such as autonomy, justice, beneficence, and non-maleficence, which are fundamental principles of medical ethics [3–5].

The World Health Organization (WHO) defined the ethical governance framework for AI use in healthcare in its 2021 report, “Ethics and Governance of Artificial Intelligence for Health”; its 2023 guide, “Regulatory Considerations on AI for Health,” details the legal and safety dimensions [6, 7]. The “LMM Guidance” document published in 2025 also addressed the new risks and opportunities brought about by generative AI models; it recommended adopting a human-centered, accountable, fair, and safe approach in education, clinical practice, and research [8]. However, various international organizations have emphasized the need for healthcare professionals to be trained in AI ethics [9]. These documents emphasize the need to adhere to ethical principles in the use of AI in healthcare.

AI literacy is a set of competencies that enable individuals to critically evaluate AI technologies, establish effective communication and collaboration with these systems, and use AI as a tool in various contexts [10, 11]. The increasing use of AI highlights the importance of AI literacy [12, 13]. From the early years of medical education, students encountered technology-based applications such as AI-supported information systems, decision support mechanisms, clinical simulations, and digital learning environments. Therefore, AI literacy is not merely a technical skill for medical students but an ethical and professional responsibility. It is recommended that AI competencies in medical education be made measurable and integrated into program outcomes [14].

Current medical education curricula have not been able to adapt sufficiently to the rapid advancement of AI in medical education [2, 15]. For AI to be used safely, fairly, and ethically in healthcare, it is important to integrate both AI literacy and ethical awareness components into educational programs. Current studies indicate a lack of research examining the ethical dimension of AI and students’ attitudes toward this issue at a sufficient level [16–18]. There are limited studies in literature investigating the relationship between ethical attitudes toward AI and AI literacy. Therefore, this study aims to determine the relationship between medical students’ AI literacy level and their attitudes toward AI ethics.

The purpose of this study is to examine the relationship between medical students’ AI literacy and AI ethics; to reveal how these variables differ in the context of various demographic factors; and to examine the extent to which AI literacy explains variance in AI ethics. In this context, answers were sought to the following research questions:

• RQ1: What are the levels of AI literacy and ethical attitudes toward AI among medical students?

• RQ2: What are the patterns of AI tool use among medical students, including frequency, type of tools, and purposes of use?

• RQ3: Do AI literacy and ethical attitudes toward AI differ according to demographic and educational variables (academic year, gender, prior AI training, and frequency of AI use?

• RQ4: Is there a significant relationship between AI literacy and ethical attitudes toward AI among medical students?

• RQ5: To what extent does AI literacy explain variance in ethical attitudes toward AI among medical students?

Method

Study design

The study design adopted for the purpose of this study is correlational study. In this design, the direction and degree of the relationship between two or more variables are examined statistically [19, 20]. Because the study used a cross-sectional correlational design, the findings reflect statistical associations observed at a single point in time and do not allow causal inferences. The data for the study were collected between November and December 2025. Detailed study methodology is shown in Fig. 1.

Fig. 1.

The study methodology

Participants

Convenience sampling was used to select the sample for this study. This sampling method relies on reaching volunteer participants whom the researchers can easily access. This approach saves both time and money [21]. The study population consisted of 1,707 medical students enrolled across all six academic years at the medical faculty during the 2025–2026 academic year. Data were collected through an online survey distributed voluntarily via class WhatsApp groups with the assistance of student representatives. The survey link included information about the purpose of the study, voluntary participation, confidentiality, and informed consent procedures. A total of 461 medical school students from all grade levels participated in the study, corresponding to a response rate of approximately 27.0%; 41.2% of the students were first-year students. This distribution is partly consistent with the overall student population of the faculty, in which first-year students also constitute the largest academic group. Of the participants, 234 of the students were female, and 227 were male. No incomplete responses were identified; therefore, all submitted questionnaires were included in the analyses. A post-hoc power analysis (α = 0.05, medium effect size, f² = 0.15) indicated that the sample size (n = 461) provided excellent statistical power (> 0.99) for regression analyses. There is no course related to AI in the faculty’s teaching programs. Students have acquired the knowledge they declare from their own individual sources.

Data collection tools

Demographic information form

Consists of questions regarding variables such as class, gender, prior AI training, frequency of AI use, AI tool used, and purpose of AI use

Artificial Intelligence Ethics Scale (AI-Ethics)

The scale aims to measure individuals’ tendency to evaluate artificial intelligence applications in accordance with ethical principles. The scale was adapted into Turkish in 2025 (Cronbach Alpha = 0.95) [22].

The 5-point Likert scale consists of 17 items and five dimensions: transparency, non-maleficence, privacy, responsibility, and fairness. High scores indicate that participants have a higher awareness and more positive attitudes toward the ethical use of artificial intelligence technologies.

Artificial Intelligence Literacy Scale (AI-Literacy)

The scale was adapted into Turkish in 2025 to measure individuals’ competence in terms of knowledge, awareness, evaluation, and ethical dimensions related to artificial intelligence (Cronbach’s α coefficient 0.85), and a validity-reliability study was conducted on healthcare professionals and student [23]. The 7-point Likert-type scale consists of 12 items and four dimensions: awareness, usage, evaluation, and ethics.

Data analysis

In the scope of the study, normality analysis and descriptive statistics were performed to determine the levels of AI literacy and AI ethics among medical school students. Descriptive statistics were performed to determine what types of AI tools students used, their purposes, and frequency of use. An independent samples t-test was performed to determine the extent to which the scores obtained from the data collection tools differed according to gender and AI training status. A one-way analysis of variance was applied to examine the extent to which the scores obtained from the data collection tools differed according to the students’ class level and AI usage frequency. A Pearson correlation analysis was performed to determine the direction and strength of the relationship between AI literacy and AI ethics. A simple linear regression analysis was conducted to examine whether AI literacy was statistically associated with AI ethics and to determine the extent to which it explained variance in AI ethics scores. A simple linear regression analysis was initially conducted to examine the association between AI literacy and AI ethics based on composite scores. PLS-SEM was employed to investigate the relationship at the latent construct level. The use of PLS-SEM was considered appropriate because both AI literacy and AI ethics represent multidimensional constructs composed of several underlying dimensions [24, 25]. Compared with conventional regression analysis based on observed variables, PLS-SEM enables the simultaneous estimation of measurement and structural models while accounting for measurement error [24, 26]. Furthermore, this approach provides a comprehensive assessment of construct reliability, convergent validity, discriminant validity, and the structural relationships among latent variables [24, 27]. Accordingly, regression analysis was used to examine the association between composite observed scores, whereas PLS-SEM was employed to evaluate whether this association remained at the latent construct level while simultaneously assessing measurement quality. The two analyses were therefore considered complementary rather than redundant.

Results

Preliminary analysis and descriptive statistics

The normality of the data obtained from 461 medical students within the scope of the study was analyzed by examining skewness and kurtosis values. In this regard, as shown in Table 1, it was determined that the skewness and kurtosis values were within the ± 2 range and that the data were normally distributed.

Table 1.

Normality analysis and scale averages

Variables	Min-Max	± SS	Skewness	Kurtosis
AI Literacy	1.00–7.00	5.21 ± 1.39	-1.314	1.514
AI Ethics	1.00–5.00	3.76 ± 0.80	-1.455	1.860

Most of the students, 41.2%, were first-year students (see Appx.1). The gender distribution was found to be balanced (women 50.8%; men 49.2%). It was determined that 75.3% of participants had not received any training related to AI, while 79.1% of participants used AI tools every day or several times a week. It was determined that the AI tool most frequently used by participants was ChatGPT at 94.1%, followed by Gemini at 48.4%. The most prominent usage purposes were accessing information and literature review at 92%, followed by personalized learning at 55.7%. Learning-supported applications followed these usage purposes.

Measurement model assessment

AI literacy and AI ethics scales used as data collection tools in the study were analyzed in terms of reliability, convergent validity, discriminant validity, and multicollinearity. Internal consistency reliability was evaluated using Cronbach’s alpha and composite reliability (CR). As shown in Table 2, the Cronbach’s alpha values ranged between 0.587 and 0.932, while CR values ranged between 0.777 and 0.953 (must be above 0.70 [28]), indicating acceptable to high levels of internal consistency reliability. In particular, the higher-order constructs of AI Ethics and AI Literacy demonstrated strong reliability coefficients (α = 0.920 and 0.932, respectively). However, the Responsibility subdimension demonstrated a relatively low Cronbach’s alpha value (α = 0.587), which falls below the commonly recommended threshold of 0.70. Although its composite reliability value remained acceptable (CR = 0.777), findings related to this subdimension should be interpreted with some caution. Convergent validity was assessed using average variance extracted (AVE). All AVE values exceeded the recommended threshold of 0.40 [24], ranging from 0.501 to 0.871, supporting adequate convergent validity for all constructs. In addition, factor loadings were generally above the recommended 0.70 threshold (see Appx 2), further supporting indicator reliability. Discriminant validity was examined using the Fornell–Larcker criterion. As reported in Appx 3, the square roots of the AVE values were greater than the inter-construct correlations for all constructs, indicating satisfactory discriminant validity. Furthermore, multicollinearity was assessed using variance inflation factor (VIF) values. As shown in Appx 4, all VIF values were below the critical threshold of 5, suggesting that multicollinearity was not a concern in the measurement model [24]. Detailed item loadings and additional validity statistics are presented in the appendices. Although the AI Literacy Scale includes an ethics-related subdimension, discriminant validity analyses supported the conceptual distinction between AI literacy and AI ethics constructs. The ethics dimension within AI literacy focuses on awareness and understanding of ethical considerations related to AI use, whereas the AI Ethics Scale evaluates broader ethical attitudes and value-based judgments regarding AI applications. In addition, Fornell–Larcker analyses demonstrated adequate discriminant validity between the constructs.

Table 2.

Reliability and convergent validity results

Variables	Cronbach’s alpha	CR	AVE
AI Ethics	0.920	0.932	0.501
AI Literacy	0.932	0.944	0.655
Fairness	0.672	0.820	0.604
Evaluation	0.926	0.953	0.871
Ethics	0.726	0.879	0.784
Awareness	0.719	0.877	0.780
Privacy	0.734	0.881	0.788
Usage	0.837	0.924	0.860
Responsibility	0.587	0.777	0.540
Harmlessness	0.872	0.922	0.797
Transparency	0.761	0.863	0.677

Group comparisons based on demographic variables

Analyses were conducted to compare the scale scores of medical school students participating in the study with various demographic characteristics. As shown in Table 3, an independent samples t-test was applied to examine whether the participants’ gender and prior AI training differed significantly in terms of scale scores; a one-way analysis of variance was applied to examine whether the students’ class levels and frequency of AI use differed significantly in terms of scale scores.

Table 3.

Comparison of participant scale scores with demographic characteristics

Variables		AI Literacy	AI Ethics
		± SS	± SS
Academic Year	1st year	5.07 ± 1.45	3.68 ± 0.88
	2nd year	5.30 ± 1.27	3.72 ± 0.78
	3rd year	5.47 ± 1.18	3.88 ± 0.69
	4th year	5.24 ± 1.35	3.94 ± 0.56
	5th year	5.21 ± 1.53	3.74 ± 0.76
	6th year	5.28 ± 1.51	3.70 ± 1.00
	Test value	0.970***	1.553***
	p	0.436	0.172
Gender	Female	5.29 ± 1.23	3.88 ± 0.73
	Male	5.13 ± 1.54	3.65 ± 0.86
	Test value	1.261**	3.082**
	p	0.208	0.002*
Prior AI training	Yes	5.53 ± 1.25	3.94 ± 0.64
	No	5.11 ± 1.42	3.70 ± 0.84
	Test value	2.838**	2.805**
	p	0.005*	0.005*
Frequency of AI tool usage in the last 30 days	Every day	5.26 ± 1.44	3.76 ± 0.82
	A few times a week	5.31 ± 1.32	3.78 ± 0.73
	Once a week	4.66 ± 1.33	3.53 ± 0.82
	A few times a month	4.92 ± 1.29	3.86 ± 0.85
	Once a month	5.13 ± 1.57	3.63 ± 1.01
	I have never used it	5.08 ± 1.73	3.82 ± 1.14
	Test value	1.366***	0.642***
	p	0.236	0.668

*p<0.05, ** Independent t-test, *** One-way analysis of variance

There is a statistically significant difference in AI ethics scores based on participants’ gender (p < 0.05). Female participants’ AI ethics scores are higher than those of male participants. However, no significant difference was observed in terms of AI literacy based on gender. There is a statistically significant difference in AI literacy and AI ethics scores based on whether participants prior AI training (p < 0.05). Participants who received AI-related training had higher AI literacy and AI ethics scores than those who did not. No significant differences were observed in either scale based on grade level or frequency of AI tool use.

Correlation and regression analysis

A Pearson correlation analysis was conducted to determine the relationship between AI literacy and AI ethics. As shown in Table 4, a statistically significant and positive relationship was found between AI literacy and AI ethics (r = 0.648, p < 0.05). This relationship, which has a medium-high correlation coefficient, indicates that as students’ AI literacy levels increase, their awareness levels regarding AI ethics also increase.

Table 4.

The relationship between variables

Variables	AI Ethics
	r	p
AI Literacy	0.648	< 0.001*

A simple linear regression analysis was conducted to determine the association between variables of AI literacy on AI ethics. As shown in Table 5, the model tested was found to be statistically significant (F = 332.309, p < 0.05). In this regard, it was determined that AI literacy significantly explained variance in AI ethics in a statistically significant and positive direction (β = 0.374, t = 18.229, p < 0.05) and explained 42% of the variance in AI ethics. The Durbin–Watson coefficient of 1.789 found in the analyses indicates that there is no autocorrelation problem in the model of this study setting.

Table 5.

Regression analysis results

Depen-dent Variable	Predictor	β	Std. Error	Beta	t	p	F	Model p	R 2	Durbin Watson
AI Ethics	AI Literacy	0.374	0.021	0.648	18.229	< 0.001*	332.309	< 0.001*	0.420	1.789

Structural model assessment

While the regression analysis examined the association between observed composite scores, the structural model provided additional evidence that the relationship remained significant when measurement error and latent construct structure were taken into account. Partial least squares structural equation modeling (PLS-SEM) path analysis was conducted to test the main model of the study. In this regard, as shown in Table 6, it was determined that AI literacy significantly explained AI ethics in the model (R² = 0.430). The positivity of the Q² value also indicated that the model’s predictive power was sufficient.

Table 6.

Structural model results

Path	R²	Q2	β	Std. Error	t	p	f²
AI Literacy → AI Ethics	0.430	0.426	0.655	0.043	15.154	< 0.001	0.753

When examining the path coefficient results, it was found that AI literacy was significantly and positively associated with AI ethics (β = 0.655, t = 15.154, p < 0.05). This indicates that higher AI literacy scores were associated with higher AI ethics scores. Looking at the effect size value, it is understood that the strength of the association between AI literacy and AI ethics was high (f² = 0.753). When examining the explanatory power of the model, it was determined that AI literacy explains 43% of the variance in AI ethics. This indicates that the model is moderately explanatory.

As shown in Fig. 2, the factor loadings related to the model are presented as the awareness, usage, evaluation, and ethical dimensions of AI literacy; while the AI ethics structure is presented as the dimensions of transparency, non-maleficence, privacy, responsibility, and justice. All factor loadings were found to be statistically significant and high (p < 0.001). Thus, it can be said that the measurement model has sufficient representativeness.

Fig. 2.

Structural model

Discussion

AI has evolved beyond being a futuristic concept in today’s medical world and has become an active force directly impacting areas such as clinical practice, decision-making processes, and healthcare management [3, 29]. The integration of this technology into medical education is a critical point in the training of future healthcare professionals. AI technologies in medical education have contributed to personalized learning, simulation-based clinical education [30], diagnosis and treatment practice, and the development of clinical reasoning skills [31, 32]. However, AI’s contributions to medical education also bring serious challenges that affect its ethical principles of autonomy, justice, beneficence, and non-maleficence [3, 33]. Integrating AI tools into medical education requires more than just technical knowledge; students must also possess AI ethics literacy [11, 30, 34]. AI ethics education teaches students to question AI outputs, understand the limitations of algorithms, explain the use of AI to patients in a transparent manner, and consider all these ethical issues when using AI [3, 15, 30].

Association between AI literacy and ethical readiness

This study found that AI literacy was significantly and positively associated with ethical attitudes toward AI use. The positive association identified in this study suggests that higher levels of AI literacy may be related to greater ethical awareness regarding the responsible use of AI in healthcare and medical education. The literature also shows that AI literacy is positively related to readiness for AI use and responsible use tendencies [2, 11, 35].

It should also be acknowledged that the AI Literacy Scale includes an ethics-related dimension. Although discriminant validity analyses supported the conceptual distinction between AI literacy and AI ethics, some degree of conceptual overlap may remain between the constructs. Therefore, the observed association should be interpreted with caution, as this overlap may have partially contributed to the strength of the relationship identified in the present study.

When evaluated within the framework of the principles of autonomy, beneficence, non-maleficence, and justice outlined by Beauchamp and Childress [36], AI literacy may be considered a fundamental component of ethical practice in AI-supported clinical settings. First and foremost, respect for patient autonomy requires the clinician to understand how AI systems work and the limitations of their outputs, and to be able to explain this transparently to the patient. A physician lacking sufficient AI literacy may be unable to meaningfully explain the role and uncertainties of AI-supported decisions, thereby weakening informed consent and shared decision-making processes. Therefore, the necessity of ethical training that includes critical evaluation of AI outputs and understanding the limitations of algorithms is emphasized [3, 15, 30].

The principles of beneficence and non-maleficence are similarly directly related to AI literacy. The literature frequently highlights risks associated with AI systems, such as hallucination risk, data bias, out-of-context suggestions, and overconfidence [32, 37–39]. Ethical concerns such as automation bias and the weakening of clinical reasoning have also been reported [9, 30, 40]. In this context, it can be said that students with high AI literacy may have a more developed capacity to recognize the potential risks of AI outputs and critically evaluate them. Indeed, it has been noted that students who lack sufficient knowledge tend to use the information provided by AI without critical evaluation [41].

The principle of justice requires the recognition of algorithmic bias and data-driven inequalities. Ethical concerns regarding fairness, privacy, and non-maleficence in AI use have been reported to be prevalent among students [38, 42]. However, basic AI literacy is required for these concerns to translate into conscious ethical evaluation. In this regard, AI literacy may support future physicians in recognizing algorithmic injustices and evaluating these risks in clinical contexts. The association identified in our study suggests that technical knowledge and ethical sensitivity may be related. Furthermore, AI literacy may support physician candidates in exercising their professional moral agency. Professional moral agency refers to the capacity for reflective decision-making, taking responsibility, and maintaining human control in technological environments. Although the literature shows that students report positive attitudes toward AI, they experience gaps in ethical practice [10, 35]. The regression findings of this study indicate that AI literacy was significantly associated with ethical positioning.

Additionally, the uncertainty regarding whether responsibility for complications arising from AI use lies with the physician or the AI system is a significant area of ethical debate [43]. Furthermore, concerns about the mechanization and dehumanization of the patient-physician relationship are also present in the literature [1, 5, 44]. These risks demonstrate that AI literacy must be integrated not only with technical skills but also with an awareness of ethical responsibility. In short, AI literacy is not merely a technical skill; it may be considered an important cognitive and ethical component in applying the principles of autonomy, non-maleficence, beneficence, and justice. AI literacy may be considered an important component that contributes to future physicians assuming ethical responsibility in clinical decision-making processes by avoiding technological dependency, conducting critical evaluations, and maintaining human oversight.

Beyond its pedagogical importance, AI literacy may contribute to ethically informed and responsible engagement with AI-enabled healthcare services. In healthcare systems where algorithmic decision support systems can influence clinical judgment, insufficient literacy may give rise to risks such as automation bias, opacity, and diffusion of responsibility. From an AI perspective, literacy acts as an intermediary layer that enables transparency, accountability, and human oversight. Therefore, these findings extend beyond medical education to address the ethical governance of AI-integrated healthcare systems.

Gender and ethical sensitivity in AI-mediated healthcare/medical education

This study found that AI literacy scores did not vary by gender, but ethical attitude scores toward AI were significantly higher among female students compared to male students. The literature reports that female students are more sensitive and cautious toward ethical dimensions such as fairness, privacy, and non-maleficence in AI use [38, 42]. Similarly, it has been reported that female students have a higher level of concern about data security, privacy, and ethical risks [38]. In contrast, some studies have indicated that male students have a higher level of technical knowledge regarding AI [45] and that male medical students are more inclined to use AI [10]. However, there are also studies reporting that gender has no significant effect on AI usage [11, 15]. The findings of this study suggest that, despite similar levels of technical knowledge between genders, differences in ethical attitudes may exist. This may be related to differences in risk perception and ethical sensitivity levels, particularly in the context of the principles of non-maleficence and justice. Considering the potential harms of AI systems, algorithmic biases, and data security risks, individuals with high ethical sensitivity levels may be expected to take a more critical and cautious approach to these technologies. On the other hand, the fact that ethical attitudes can develop independently of technical competence suggests that AI literacy alone may not be sufficient to explain ethical sensitivity; value-based approaches and professional socialization processes may also be decisive. Therefore, it is important to address gender differences in AI ethics education not as a one-dimensional issue of technical capacity, but in the context of ethical risk perception, sense of responsibility, and understanding of professional roles.

The role of formal AI education in responsible use

This study found that students who had previously received AI-related training had significantly higher scores in both AI literacy and ethical attitudes toward AI. This finding suggests that simple frequency of AI use, as measured in the current study, was not significantly associated with higher AI literacy or ethical attitude scores. However, because the study did not assess the depth, quality, or context of AI use, these findings should be interpreted cautiously. The literature reports that although most students are familiar with AI tools, only a limited proportion have received formal AI training [35]. A study on the duration of AI training showed that students who received 30 h or more of training had a significant increase in their AI literacy scores [11]. Similarly, studies conducted on healthcare professionals and students have revealed a positive and significant relationship between AI literacy and the level of readiness to use AI [2, 35]. In a study conducted with nursing students, it was reported that a three-day structured AI ethics training program significantly increased students’ ethical awareness and their intention to use AI ethically [46]. When these findings are evaluated together, it can be said that students who reported prior AI training also demonstrated higher ethical attitude scores, suggesting a possible association between structured education and ethical awareness. Responsible AI use requires understanding not only how algorithms work, but also their limitations, potential errors, and contextual risks. It is particularly important for students to be able to critically evaluate the accuracy of AI outputs to ensure the sustainability of human oversight in clinical decision support systems.

In this study, the fact that frequency of use did not create a significant difference in AI literacy and ethical attitudes suggests that informal AI use was not significantly associated with differences in ethical competence in the current study. Although it has been determined that students use AI most frequently for accessing information and literature review, it can be predicted that this usage may remain at a superficial level of familiarity when not supported by critical evaluation and ethical reasoning. Indeed, the literature reports that despite students’ high level of knowledge about AI, they experience gaps in practical ethical application [10].

In this context, formal AI education may contribute to the cognitive and ethical foundation of responsible AI use. Educational programs should not be limited to technical modules; they must also cover ethical dimensions such as algorithmic bias, data privacy, responsibility sharing, and the impact on the patient-physician relationship [3, 15, 30]. Such an approach can contribute to students moving beyond being passive consumers of AI and positioning themselves as critically thinking and responsible professionals.

However, because the study was cross-sectional, the observed differences between students with and without prior AI training should be interpreted cautiously. It is possible that students who were already more interested in AI or more ethically sensitive were more likely to seek AI-related training. Therefore, the observed differences between students with and without prior AI training should not be interpreted as evidence of a causal effect of training itself. Pre-existing interest in AI, ethical sensitivity, or motivation toward technology-related learning may also have influenced students’ likelihood of seeking AI-related education.

Frequency of AI use and the knowledge–practice gap

This study shows that the frequency of AI use does not create a meaningful difference in either AI literacy or ethical attitudes toward AI. Similarly, no significant difference was observed based on grade level. These findings indicate that no statistically significant association was observed between frequency of AI use and ethical competence in the current study. However, because the study measured only frequency of use and not the quality or depth of engagement with AI tools, these findings should be interpreted cautiously. The literature reports that AI usage duration and frequency vary. For example, a study with nursing students found that 22.7% of students used AI for more than 1 h per day [41], while a study among medical students in Europe found that half of the students used AI chat applications for 1–3 h per week [5]. However, it is emphasized that increased usage time does not always reflect effective and conscious use; a significant portion of students do not fall into the “effective user” category [41]. This highlights the need to distinguish between quantitative use and qualitative and ethical use.

This study found that students most frequently use AI for accessing information and conducting literature reviews. However, it is thought that without critical evaluation and ethical reasoning, this usage may remain at a superficial level of familiarity. Indeed, different studies have reported that although students’ AI knowledge scores are high, they score lower in practical application and behavioral dimensions [10]. This situation points to a divergence that can be defined as a knowledge-application gap. Superficial familiarity with AI technologies can bring with it the risks of automation bias and overconfidence. The literature indicates that students’ express concerns about the accuracy of information provided by AI [35] and that individuals lacking sufficient technical knowledge tend to use AI outputs without subjecting them to critical scrutiny [41]. This situation has been associated with risks of weakened clinical reasoning and dependence on technology [9, 30, 40].

The absence of a significant difference based on class level is also noteworthy. The assumption that ethical awareness increases with clinical experience is not supported by the findings of this study. This situation demonstrates that professional advancement does not automatically develop AI-specific ethical competencies. Similarly, some studies have reported that although students in the clinical stage have more positive attitudes toward AI, their knowledge and behavior scores are lower [10]. These findings suggest that ethical maturity in the context of AI develops through structured education and critical awareness rather than experience. The knowledge-practice gap can be considered one of the major barriers to responsible AI use. The technical accessibility of AI tools does not imply that they will be used in an ethically conscious and responsible manner. Therefore, structured educational strategies that strengthen critical thinking and ethical reasoning are needed, rather than an exposure-based approach. Again, the reason for the lack of a difference in this study may be due to students’ familiarity with AI tools and the fact that they have not yet been adapted to medical education.

Educational implications for ethical AI integration in medical curricula

Because these implications are based on cross-sectional associations, they should be interpreted as evidence-informed educational considerations rather than causal recommendations. The findings of this study indicate that AI literacy was significantly associated with ethical attitudes and that formal education was positively associated with both AI literacy and ethical sensitivity. However, the fact that frequency of use and class level did not create a significant difference indicates that frequency of AI use alone was not significantly associated with ethical preparedness in the current study. These results suggest that medical curricula may benefit from more structured approaches to AI integration rather than relying primarily on random or indirect learning experiences. The literature emphasizes that AI education should not be limited to the transfer of technical knowledge; it should also cover ethical principles, legal responsibilities, and algorithmic bias [3, 30, 38, 47]. It is recommended that case-based discussions and simulation exercises addressing AI-based ethical dilemmas be integrated into the curriculum [39, 46, 48]. This approach can enable students to not only learn how to use AI tools, but also to critically evaluate the role of these tools in clinical decision-making processes. In our study, the fact that students who received formal AI training had higher ethical attitude scores supports the idea that structured training may be associated with higher levels of ethical awareness. Taken together with previous literature, the present findings support considering AI education as an important component of medical curricula rather than solely an elective topic. Indeed, medical students have been reported to support the mandatory integration of AI and AI ethics education into the curriculum [5, 44]. Furthermore, medical schools should collaborate with computer scientists, ethics experts, lawyers, and policymakers in developing AI-related curricula [39, 48]. Universities should prepare guidelines that include ethical policies regarding the use of AI in assignments and exams [37, 47, 49, 50].

From an educational design perspective, AI integration can be approached in three key dimensions. First, technical literacy: understanding the logic behind algorithms, data sources, and their limitations. Second, ethical reasoning: evaluating AI use in the context of principles such as autonomy, non-maleficence, beneficence, and justice. The third is professional responsibility and human oversight: instilling the awareness that AI outputs do not replace the final decision and that the clinician’s clinical and ethical responsibility continues. This multidimensional approach can contribute to students moving beyond being passive consumers of AI and becoming critical thinkers and responsible healthcare professionals. Furthermore, the rapid development of AI technologies indicates that this education should not be limited to the pre-graduation period. It is recommended that AI ethics and responsible use modules be offered as part of continuing professional development programs [46, 48]. This approach can contribute to the formation of a professional culture that can adapt to the pace of technological transformation and keep its ethical reflexes up to date.

These findings also have implications at the institutional level. Medical schools have an ethical responsibility to ensure that students develop a critical perspective on AI integration. Developing institutional AI policy frameworks, establishing ethical oversight mechanisms, and incorporating AI literacy into accreditation standards can ensure that the technology is aligned with ethical principles and responsible AI practices. Without such governance structures, the burden of responsible AI use may fall disproportionately on individual learners.

Conclusion

This study demonstrates that AI literacy was significantly associated with ethical attitudes toward AI. The findings suggest that technical knowledge and ethical sensitivity may be related.

No significant association was observed between AI use frequency and ethical competence in the current study. However, because the study did not evaluate the quality or context of AI use, these findings should be interpreted cautiously. In contrast, the fact that students who have received formal AI education have higher literacy and ethical attitude scores supports the idea that the findings suggest a possible association between structured education and responsible AI use.

These findings indicate that AI integration should be approached not merely as a matter of technological adaptation, but as a process of developing ethical capacity. The preservation of the principles of autonomy, non-maleficence, beneficence, and justice in AI-supported clinical decision-making processes depends on students’ ability to understand algorithmic limitations, maintain human oversight, and assume professional responsibility. Therefore, AI literacy should be considered not only a technical competency for future physicians but also an important component associated with ethical preparedness. The inclusion of structured, ethics-based AI education in the medical curriculum may contribute to the sustainability of human-centered and responsible healthcare in the technological transformation process. These findings reposition AI literacy not merely as a pedagogical competency but as an important component of ethically responsible AI use in healthcare.

Limitations

This study has some limitations. First, due to its cross-sectional design, the relationship between AI literacy and ethical attitudes cannot be interpreted causally. Although statistically significant associations were identified the direction and continuity of this relationship should be supported by longitudinal studies. Data were collected using self-report scales, which may carry the risk of social desirability bias. In addition, because both predictor and outcome variables were measured using self-reported questionnaires administered at a single time point, the possibility of common-method bias cannot be completely excluded. Consequently, part of the observed association may reflect shared measurement method rather than the true magnitude of the relationship. The extent to which reported ethical attitudes reflect actual clinical behavior is unknown. Furthermore, this study was conducted at a single public university within a specific national and curricular context; this may limit the generalizability of the findings to medical students in different educational systems, institutional structures, and cultural contexts. Additionally, although students from every class level were included, it is noteworthy that first-year students constituted a larger proportion of the sample than initially planned. This distribution may have influenced the overall representation of AI literacy and ethical attitudes within the sample. However, this distribution partially reflects the structure of the overall student population, in which first-year students also constitute the largest academic cohort. Moreover, although proportional representation across academic years was initially intended, the final sample was obtained through voluntary participation via class WhatsApp groups and therefore reflects a convenience sampling approach. Consequently, selection bias cannot be completely ruled out. Furthermore, participants had not received standardized AI training as part of the official medical curriculum; instead, they acquired their AI-related knowledge primarily through individual experience and self-directed learning. Finally, the nature of AI use has not been analyzed in detail. Future studies are recommended to deepen these findings using behavioral measures and multi-center designs.

Because data were collected at a single point in time, temporal ordering between variables could not be established. Therefore, although statistically significant associations were identified, it cannot be concluded that AI literacy leads to changes in ethical attitudes. Reverse or bidirectional relationships may also be possible. In addition, because the AI Literacy Scale includes an ethics-related dimension, some degree of conceptual overlap with the AI Ethics Scale cannot be completely ruled out. Although discriminant validity analyses supported the distinction between the constructs, this overlap may have partially inflated the observed association. Additionally, although gender and prior AI training were examined through separate group comparisons, adjusted regression models including these demographic variables were not conducted. Therefore, the independent association between AI literacy and ethical attitudes should be interpreted with caution, as potential confounding by demographic characteristics cannot be ruled out. Future studies should employ multivariable regression models that simultaneously adjust for gender, prior AI training, and other relevant demographic variables to better estimate the independent contribution of AI literacy to ethical attitudes.

Additionally, students who had previously received AI-related training may already have differed from other participants in terms of interest in AI, motivation, or ethical sensitivity. Therefore, the observed group differences cannot be interpreted as direct effects of training.

Abbreviations

AI

Artificial Intelligence

Appx

Appendix

LMM

Large Multimodal Models

RQ

Research Question

WHO

World Health Organization

Authors’ contributions

HHÜ and FT conceived the idea for the study and designed the survey. HHÜ collected the data and performed the initial literature review. FT managed the survey. HHÜ and FT performed the statistical analysis and wrote the manuscript. All authors read, contributed to, and approved the final version submitted for publication.

Funding

The authors did not receive any financial support for the research, writing, and/or publication of this article.

Data availability

The datasets used in the present study can be obtained from the corresponding author upon reasonable request.

Declarations

Ethics approval and consent to participate

This study was approved by the Non-Interventional Ethics Committee of the Faculty of Medicine, Aydın Adnan Menderes University (Prot.No: 2025/313 − 18). Before data collection, participants were provided with information regarding the voluntary basis of their involvement and the use of their responses solely for research purposes. Informed consent was subsequently obtained, and participants were assured that all data would remain confidential. The study was conducted in accordance with the Helsinki Declaration. The study procedures adhered strictly to applicable ethical standards and regulatory requirements.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.