Navigating the integration of artificial intelligence in the medical education curriculum: a mixed-methods study exploring the perspectives of medical students and faculty in Pakistan

Abstract

Background

The integration of artificial intelligence (AI) into medical education is poised to revolutionize teaching, learning, and clinical practice. However, successful implementation of AI-based tools in medical curricula faces several challenges, particularly in resource-limited settings like Pakistan, where technological and institutional barriers remain significant. This study aimed to evaluate knowledge, attitudes, and practices of medical students and faculty regarding AI in medical education, and explore the perceptions and key barriers regarding strategies for effective AI integration.

Methods

A concurrent mixed-methods study was conducted over six months (July 2023 to January 2024) at a tertiary care medical college in Pakistan. The quantitative component utilized a cross-sectional design, with 236 participants (153 medical students and 83 faculty members) completing an online survey. Mean composite scores for knowledge, attitudes, and practices were analyzed using non-parametric tests. The qualitative component consisted of three focus group discussions with students and six in-depth interviews with faculty. Thematic analysis was performed to explore participants’ perspectives on AI integration.

Results

Majority of participants demonstrated a positive attitude towards AI integration. Faculty had significantly higher mean attitude scores compared to students (3.95 ± 0.63 vs. 3.81 ± 0.75, p = 0.040). However, no statistically significant differences in knowledge (faculty: 3.53 ± 0.66, students: 3.55 ± 0.73, p = 0.870) or practices (faculty: 3.19 ± 0.87, students: 3.23 ± 0.89, p = 0.891) were found. Older students reported greater self-perceived knowledge (p = 0.010) and more positive attitudes (p = 0.016) towards AI, while male students exhibited higher knowledge scores than females (p = 0.025). Qualitative findings revealed key themes, including AI’s potential to enhance learning and research, concerns about over-reliance on AI, ethical issues surrounding privacy and confidentiality, and the need for institutional support. Faculty emphasized the importance of training to equip educators with the necessary skills to effectively integrate AI into their teaching.

Conclusions

This study highlights both the enthusiasm for AI integration and the significant barriers that must be addressed to successfully implement AI in medical education. Addressing technological constraints, providing faculty training, and developing ethical guidelines are critical steps toward fostering the responsible use of AI in medical curricula. These findings underscore the need for context-specific strategies, particularly in resource-limited settings, to ensure that medical students and educators are well-prepared for the future of healthcare.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12909-024-06552-2.

Background

Artificial intelligence (AI) has emerged as a transformative force across various sectors, including education, and is defined as the ability of machines to carry out tasks that typically require human intelligence, such as reasoning, learning, and problem-solving [1]. In recent years, the use of AI in educational settings has drawn a lot of attention, driven by advancements in avenues such as natural language processing, deep learning, and machine learning. The 2016 Horizon Report highlighted AI as a critical advancement poised to profoundly impact higher education [2]. In the context of medical education, AI holds the promise of revolutionizing teaching and learning in manners that not only personalize learning experiences but also integrate AI-driven technologies, including virtual patients and predictive analytics, into the experience, thus potentially impacting educational outcomes up to the clinical practice level [3].

However, despite its potential, a successful integration of AI into medical curricula requires solutions for various challenges, specifically those conserning privacy and ethical considerations, necessitating the development of robust regulatory frameworks [4]. Additionally, a major obstacle continues to be teachers’ and students’ lack of readiness to apply AI tools [5]. Globally, medical institutions are at varying stages of integrating AI into their curricula. While studies from developed regions such as the USA, China, and Europe have demonstrated the benefits of AI in improving medical competencies [6–8], resource-limited settings, including many developing countries, lag behind. These disparities risk widening the gap between medical graduates from developing and developed regions, leaving the former underprepared for an AI-driven healthcare landscape where AI is increasingly critical for clinical decision-making and diagnostics [9].

Previous studies have assessed a general readiness among healthcare professionals regarding the utilisation of AI, revealing a generally positive attitude [9]. Students view AI mainly as a tool to help a physician, not substitute for them, even though many have not had formal curricular exposure to AI. Furthermore, even as there is great enthusiasm regarding the potential of AI, concerns remain regarding job displacement and its effect on the doctor-patient.

In the context of Pakistan, these challenges are even more pronounced. The current undergraduate medical education curriculum lacks formal coverage of AI-related topics, and systemic barriers such as limited access to technology, inadequate institutional support, and a lack of faculty training further hinder progress. Recent studies have assessed the general readiness of healthcare professionals regarding AI, reporting positive attitudes but minimal exposure to formal curricular integration [10]. While some studies have explored healthcare professionals' perceptions of AI, the specific attitudes, knowledge, and practices of medical students and faculty—key stakeholders in education and training—remain underexplored in Pakistan [11]. Understanding these stakeholders' readiness to integrate AI into the curriculum is essential for designing context-appropriate AI training programs.

The present study aims to address these gaps by providing a comprehensive evaluation of both faculty’s and students’ self-perceived knowledge levels, attitudes and practives regarding AI and the integration of AI in medical education. The study also aims to explore the socio-cultural factors affecting our key stakeholders’ perceptions regarding a formal integration of AI-based courseworks in the present medical curricula, as well as identify their perceived hindrances and challenges towards such an integration, thus laying the groundwork for curriculum development and policy implementation.

Methods

A concurrent mixed methods study was conducted over a period of six months (July 2023 to January 2024) at the medical college of a tertiary care center in Karachi, Pakistan.

Quantitative component

The quantitative component employed a cross-sectional design. Convenience sampling was used to recruit medical students and faculty members via email invitations to participate in the online survey. The invitation to participate was thus extended to all students and faculty, and those that agreed to completion of the survey were included in the study. At the time of data collection, first-year medical students had not yet been inducted into the medical program and were therefore excluded from the study population. Eligible participants included students in their second, third, and fourth years of the program (n = 300) and all faculty members employed at the medical college (n = 178). All participants were required to provide informed consent through a web-based form prior to accessing the survey. Individuals who did not consent were excluded from the study.

The survey tool collected demographic information from the participants such as gender, age group, year of study, and years of teching experience. This was followed by a questionnaire which was developed based on validated questionnaires from two prior cross-sectional surveys conducted in Pakistan and Syria [12, 13]. The initial 34-item questionnaire underwent content validation by five experts, including three medical educationists with more than 10 years of teaching experience, one associate director of digital and blended learning, and a PhD student involved in the study. Each expert independently rated the relevance of each item on a five-point Likert scale, ranging from 0 (not relevant) to 4 (highly relevant).

The content validity ratio (CVR) for each item was calculated using the following formula:

$$CVR=(N_e-N/2)/(N/2)$$

where ​N_e represents the number of experts rating the item as "highly relevant" and N is the total number of experts. A threshold of 0.6 was used, meaning that only items with a CVR of at least 0.6 were retained in the final questionnaire. As a result, 10 out of the 34 items were removed (the initial questionnaire, along with the CVR values for each item, is provided in Supplementary Table 1).

Items of the finalized questionnaire was categorized into three domains: self-perceived knowledge, attitude, and practice. Responses were recorded using a five-point Likert scale (1 = strongly disagree to 5 = strongly agree) with a higher rating indicating a higher self-perceived knowledge level, more positive attitudes, or a higher liklihood of practicing each respective item. Composite scores for each domain were calculated by summing the Likert-scale responses for all items in a domain and dividing by the number of items, resulting in a mean score per respondent for that domain. To ensure clarity and appropriateness of the items, the questionnaire was piloted on a subset of the population (10% of the study sample: 15 students and 8 faculty members). These individuals were excluded from the main survey. As no significant changes were required following the pilot study, the finalized survey was disseminated via email to the full study population. No identifiable information was collected, ensuring participant anonymity and promoting unbiased, honest responses.

Sample sizes for medical students and faculty were calculated based on the total population sizes of 300 and 178, respectively, using a 90% confidence interval and a hypothesized frequency of 50% for good knowledge and positive attitudes regarding AI integration, based on the formula for estimating a proportion in a finite population [14].

$$n=\frac{N{Z}^{2}p(1-p)}{d^2\left(N,-,1\right)+Z^{2}p(1-p)}$$

where:

• N = total population size,

• Z = Z-score corresponding to the desired confidence level (1.645 for 90%),

• p = hypothesized frequency of the outcome (set at 50% to maximize the sample size),

• d = margin of error (set at 5%).

Using this formula, the required sample sizes were determined to be 143 medical students and 108 faculty members.

Qualitative component

For the qualitative component, focused group discussions (FGDs) with medical students and in-depth interviews (IDIs) with faculty members were conducted. A comprehensive interview guide was developed based on a thorough literature review on AI, AI-based tools in medical education, and the benefits and barriers associated with integrating AI into medical curricula. This interview guide was utilized to facilitate the discussions with the students and faculty (the interview guide is available as Supplementary Material 1).

Purposive sampling was employed to invite medical students from each of the three eligible years of study, ensuring adequate representation from each level of education. Three FGDs were held, each comprising of eight participants. Similarly, purposive sampling was employed to recruit faculty members who were year leads of the medical program or were medical educationists, due to their direct role in the curriculum development and delivery process. Six faculty members—three year leads, one from each academic year, and three medical educationists—were interviewed. Qualitative data collection concluded when thematic saturation was achieved, defined as the point in data analysis beyond which no new themes emerged [15]. This iterative process ensured comprehensive exploration of perspectives.

The FGDs and IDIs were facilitated by a trained member of the research team (MAN) using the interview guide. The facilitator was not assisted by other team members during data collection. Transcription of the audio recordings was performed verbatim by MAN and revised by MGRM, ensuring the accuracy of participants’ responses. Field notes or journaling were not undertaken during the discussions, and data analysis was based solely on participants' verbal responses.

To protect participant privacy and confidentiality, the IDIs were conducted in the respective faculty members' offices, while the FGDs with students were held in designated campus meeting rooms. Participants were informed about the interview process and the audio recording of the sessions, and written informed consent was obtained before each discussion.

The recorded interviews were transcribed verbatim by a member of the research team and subsequently underwent thematic analysis. As saturation of themes was reached after the three FGDs and six IDIs, no further interviews or discussions were conducted.

Triangulation of findings

Triangulation was performed to enhance the rigor and validity of this concurrent mixed-methods study at the levels of data, methods and researchers. Quantitative survey data and qualitative findings from FGDs and IDIs were compared and cross-referenced to identify consistencies and divergences in participants' knowledge, attitudes, and practices regarding AI integration. Methods triangulation was performed through the integration of quantitative and qualitative data, which provided complementary perspectives, with the quantitative component offering measurable trends and the qualitative component providing contextual depth. Finally, researcher triangulation was ensured through data coding independently by two researchers of the team, and differences in interpretation were resolved through consensus, ensuring reliability in thematic analysis.

Data analysis

Quantitative data were analyzed using STATA version 15 [16]. Categorical variables are reported as frequencies and percentages, while mean composite scores for the three domains of the survey questionnaire (knowledge, attitude, and practices) are presented as means ± standard deviations (SD). Due to the non-parametric distribution of the data, univariate analyses were conducted using Mann–Whitney U and Kruskal–Wallis H tests to compare mean scores across various participant demographics. Bonferroni Correction was applied to calculate the adjusted alpha values for multiple comparisons across the students and faculty subgroups. Internal consistency of the questionnaire domains was assessed using Cronbach’s α. A p-value of less than 0.05 was considered statistically significant.

For the qualitative component, thematic analysis was performed on the transcribed interviews [17]. Each significant word or sentence from the responses was treated as a unit of analysis during the coding process. Through an iterative process of coding, evaluating, and discussing the data among the research team, similar codes were combined to develop key themes which were refined until consensus was reached. To enhance the rigor of the analysis, two researchers independently coded the data. Similar codes were grouped together, allowing the main themes to emerge. NVivo software was used to facilitate data management and organization throughout the qualitative analysis [18].

Clinical trial number

Not applicable.

Results

Quantitative results

A total of 236 participants completed the survey, comprising 83 faculty members (35.2%) and 153 medical students (64.8%). Among faculty, majority were aged 31 to 40 years (41%, n = 34), held positions as Assistant Professors (50.6%, n = 42) and had 1 to 5 years of teaching experience (25.3%, n = 21). Majority of the students belonged to the 20 to 21 year-old age group (59.5%, n = 91) and were in their 3rd year of study (47.1%, n = 72) (Table 1).

Table 1.

Demographics details of survey participants (n = 236)

Population	Variable	Frequency	Percetage
Faculty Members	Age Group
	• 21–30 years	16	19.3
	• 31–40 years	34	41
	• 41–50 years	15	18.1
	• 51–60 years	10	12
	• Older than 60 years	8	9.6
	Gender
	• Male	30	36.1
	• Female	53	63.9
	Teaching Position
	• Lecturer/Demonstrator	19	22.9
	• Senior Registrar/Senior Lecturer	11	13.3
	• Assistant Professor	42	50.6
	• Associate Professor	4	4.8
	• Professor	7	8.4
	Teaching Experience
	• Less than 1 year	15	18.1
	• 1–5 years	21	25.3
	• 6–10 years	19	22.9
	• 11–15 years	15	18.1
	• More than 15 years	13	15.7
Students	Age Group
	• 18–19 years	32	20.9
	• 20–21 years	91	59.5
	• 22–23 years	28	18.3
	• 24–25 years	2	1.3
	Gender
	• Male	44	28.8
	• Female	109	71.2
	Year of Study
	• 2nd year	59	38.6
	• 3rd year	72	47.1
	• 4th year	22	14.5

The internal consistency of both surveys, measured using Cronbach's alpha, was 0.94 for the 24-item faculty survey and 0.94 for the 22-item student survey, indicating excellent reliability. Responses to the surveys revealed key insights into the respondents’ self-perceived knowledge and current practices related to AI, as well as their attitudes towards AI’s integration into medical education. Faculty members had significantly higher mean composite attitude scores compared to medical students (3.95 ± 0.63 versus 3.81 ± 0.75, p = 0.040). Analysis of mean composite knowledge scores showed that both groups perceived themselves as knowledgeable, with faculty scoring 3.53 ± 0.66 and students 3.55 ± 0.73, though the difference was not statistically significant (p = 0.870). Similarly, mean composite practice scores related to AI use were also similar, with no statistically significant difference between faculty (3.19 ± 0.87) and students (3.23 ± 0.89, p = 0.891) (Table 2).

Table 2.

Mean composite scores of participants’ self-perceived knowledge, attitudes and practices related to AI

Questionnaire Component	Mean score ± SD		p-value
	Faculty	Students
Self-perceived knowledge	3.53 ± 0.66	3.55 ± 0.73	0.870
Attitude	3.95 ± 0.63	3.81 ± 0.75	0.040
Practices	3.19 ± 0.87	3.23 ± 0.89	0.891

SD: standard deviation

Evaluation of individual item responses by faculty (Fig. 1) and students (Fig. 2) revealed that while 79.6% (n = 66) of faculty and 79.1% (n = 121) of students believed that they possess a general understanding of AI, only 42.1% (n = 35) of faculty and 41.2% (n = 63) of students were aware of AI subsets, such as machine learning or deep learning. Additionally, approximately only half of the faculty members (50.6%, n = 42) and 42.5% (n = 65) of students were aware of the AI tools currently being utilised in medical education. Similarly, while a vast majority of faculty (84.3%, n = 70) and students (74.5%, n = 114) believed that the integration of AI in medical education could improve the teaching and learning experience, only 45.7% (n = 38) of faculty and 45.1% (n = 69) of students reported to be confident while using AI-based tools in their teaching and learning practices. Less than half of the respondents (45.8% of both faculty members and students, n = 38 and n = 70, respectively) reported to have personally used AI tools or technologies in their medical education practices. However, when asked about their interest in learning more about the utility of AI in medical education, majority of the respondents expressed a positive attitude, with 84.3% (n = 70) of faculty members and 66.0% (n = 101) of students stating that they would like to actively seek out opportunities such as workshops, webinars, or conferences to further enhance their understanding of AI and its utility.

Fig. 1.

Responses of the faculty members to the administered survey (n = 83). x-axis: percentage, y-axis: survey item

Fig. 2.

Responses of the students to the administered survey (n = 153). x-axis: percentage, y-axis: survey item

Subgroup analyses of the mean composite scores of knowledge, attitude and practices across the demographic variables revealed that, among medical students, age was significantly associated with mean knowledge and mean attitude levels, with older students reporting greater self-perceived knowledge of AI (p = 0.010), and more positive attitudes towards AI’s utilitiy and integration in medical education (p = 0.016). Additionally, male students were found to have significantly higher mean knowledge scores than their female counterparts (3.69 ± 1.02 versus 3.49 ± 0.57, p = 0.025), however this difference was not significant after adjusting for multiple comparisons. Among faculty respondents, no significant differences in mean knowledge, attitude, or practice scores were observed across age, gender, teaching position, or teaching experience (Table 3).

Table 3.

Subgroup analysis of mean composite scores of the survey domains by participant demographics

Respondent group	Grouping variable		Mean knowledge score	p-value	Mean attitude score	p-value	Mean practices score	p-value
Faculty	Gender	Male	3.44 ± 0.65	0.577	3.85 ± 0.81	0.909	3.08 ± 0.85	0.331
		Female	3.58 ± 0.68		4.01 ± 0.50		3.25 ± 0.90
	Age group	21–30 years	3.59 ± 0.88	0.692	3.85 ± 0.79	0.782	3.16 ± 0.87	0.214
		31–40 years	3.57 ± 0.64		4.04 ± 0.38		3.41 ± 0.78
		41–50 years	3.58 ± 0.52		4.02 ± 0.63		3.15 ± 1.04
		51–60 years	3.42 ± 0.70		3.93 ± 0.73		2.90 ± 1.02
		Older than 60 years	3.27 ± 0.57		3.70 ± 0.99		2.75 ± 0.69
	Teaching position	Lecturer	3.65 ± 0.82	0.662	3.88 ± 0.72	0.297	3.24 ± 0.83	0.666
		Senior Lecturer	3.46 ± 0.68		4.02 ± 0.47		3.52 ± 0.79
		Assistant Professor	3.50 ± 0.63		4.03 ± 0.55		3.12 ± 0.95
		Associate Professor	3.33 ± 0.83		3.14 ± 1.15		2.94 ± 1.16
		Professor	3.62 ± 0.43		4.08 ± 0.44		3.11 ± 0.53
	Teaching experience	Less than 1 year	3.66 ± 0.52	0.658	3.87 ± 0.39	0.512	3.13 ± 0.62	0.514
		1–5 years	3.36 ± 0.74		4.00 ± 0.70		3.15 ± 0.95
		6–10 years	3.71 ± 0.67		4.15 ± 0.40		3.50 ± 0.85
		11–15 years	3.42 ± 0.77		3.84 ± 0.80		3.07 ± 1.13
		More than 15 years	3.51 ± 0.55		3.83 ± 0.78		3.00 ± 0.74
Students	Gender	Male	3.69 ± 1.02	0.025	3.78 ± 1.00	0.481	3.42 ± 0.97	0.072
		Female	3.49 ± 0.57		3.82 ± 0.62		3.15 ± 0.85
	Age group	18–19 years	3.20 ± 0.66	0.010	3.53 ± 0.73	0.016	2.99 ± 0.75	0.324
		20–21 years	3.63 ± 0.75		3.82 ± 0.74		3.30 ± 0.92
		22–23 years	3.63 ± 0.68		4.08 ± 0.73		3.29 ± 0.95
		24–25 years	3.75 ± 0.11		3.66 ± 0.28		3.00 ± 0.95
	Year of study	2nd year	3.46 ± 0.75	0.310	3.68 ± 0.71	0.071	3.14 ± 0.87	0.621
		3rd year	3.58 ± 0.74		3.88 ± 0.77		3.27 ± 0.92
		4th year	3.67 ± 0.67		3.89 ± 0.78		3.32 ± 0.89

Adjusted alpha for Faculty comparisons: p < 0.013

Adjusted alpha for Student comparisons: p < 0.017

Values in bold are statistically significant after adjusting for multiple comparisons

Qualitative results

In the qualitative component, three FGDs were conducted, with eight medical students per group, and six IDIs were held with faculty members Several themes emerged from these discussions, providing insights into the understanding, opportunities, challenges, and ethical considerations associated with integrating AI into medical education (Table 4).

Table 4.

Themes, subthemes and quotes from the IDIs with faculty memebers and FGDs with students

Theme	Subtheme	Verbatim Quote
AI in Medical Education	Understanding of AI	"My understanding is that artificial intelligence is something computer-based that can stimulate human intelligence. For example, we ask various questions, and they assimilate by gathering the data. They simulate intelligence and try to give you a holistic picture of your question." -IDI_Faculty_3 "It is computer with human thoughts, actually." -FGD_1_Student_5
	Functional Capacity	"It is the development and application of the algorithm for mimicking human thinking and decision-making." -IDI_Faculty_5 "This is a software that has very vast information for everything." -FGD_1_Student_3
	Use of AI in Education	"Artificial intelligence is being used particularly in areas where we have to plan something in medical education… for lesson planning… students are also using it… to search articles, to search more accurate references and more accurate information." -IDI_Faculty_6 "We use ChatGPT to get our answers within seconds. We can use it for generating mnemonics or flash cards." -FGD_1_Student_6
	AI in Research	"Yes, of course, it has a role especially in terms of research- when I am doing my research and I know about using artificial intelligence tools … They at least give us directions to go, where the important things are." -IDI_Faculty_3 "It could help you look up a topic to research on. Basically, for identification of a topic on which research could be done. If there is already an article on that topic, so we can assess that through AI." -FGD_2_Student_2
Limitations of AI in Education	Over-reliance and Misuse of AI	"They should not get dependent on it… there are few misuses as well, that's what I think." -IDI_Faculty_1 "Students might start abusing AI… they will depend too much on it. It can hinder their critical thinking." -FGD_1_Student_6 "They should be used as an additional and supplementary tool…Making these tools the first principal tool would make a generation of educators very crippled." -IDI_Faculty_3
Educational Integration of AI	Foundational Knowledge	"What students need to learn is to first get familiar with the common AI search engines and platforms… learn to play along how the AI responds." -IDI_Faculty_2 “Our undergraduate programs are already very extensive and if you will incorporate AI extensively in our curriculum students will not be able to grasp it. So I think sticking to the basics of AI and how to simply use it will be enough.” -FGD_1_Student_2
	Faculty Training	"To start with, in our context- we should educate the faculty first… then we must pursue it accordingly." -IDI_Faculty_5
	Integration Strategies	"(AI can be integrated) with the current modules, and you know there is no need of a separate course or separate thing for this AI." -FGD_1_Student_3 "It should be included in one module or some part of it… But replacing or putting in place of our previous system is not recommended." -IDI_Faculty_1 "I don't think we need to have a stand-alone (curriculum) for that, because we don't need it.” -IDI_Faculty_3
Ethical and Moral Considerations	Confidentiality and Privacy	"Data privacy will be a big problem… There is always a chance that it can be leaked or it can be approached by others." -IDI_Faculty_5 "I guess people will not be very comfortable sharing their personal information to AI… people want things to be private." -FGD_2_Student_4
	Mitigation Strategies	"Making someone accountable is the way to deal with things in a better way, whether it's a simple rule or the use of AI. So, I think the accountability and the clarity of the learners that the clear guidelines given to the learners about the legalities, the penalties they're going to face because of use of this AI or irrational use of the AI tool will make them accountable for their acts and may improve it." -IDI_Faculty_3
Institutional Factors and Support	Institutional Support	"Institutional readiness is also needed because it might raise concerns about the validation of the content." -IDI_Faculty_2 "It is important, and I think nothing is possible without the institutional support particularly when we are talking about incorporation of something new in the curriculum." -FGD_2_Student_4
	Resource Allocation	"The resources are very important, like, what kind of artificial intelligence tools your organization is going to offer. What do you think will help the learning and not hamper the learning. I think these are the most important areas or the factors that will affect the use of Artificial Intelligence." -IDI_Faculty_3
	Accessibility	"So there are paid websites of AI tools which an individual cannot buy easily. So an institute can provide free institutional access to the students." -FGD_3_Student_3
Barriers to AI Integration and Future Recommendations	Technological and Financial Constraints	"Choosing the best from the rest, it also becomes a challenge… for a third world country, buying subscriptions on a monthly basis could be hefty on pockets for some of the candidates." -IDI_Faculty_2
	Mindset and Expertise	"If I'm not ready to accept AI as an individual, as a learner, whether it's faculty or student, the biggest barrier is the individual themselves." -IDI_Faculty_2 "There's kind of a stigma that AI is not useful, or that people just misuse it. I think first of all that should be addressed that it can be useful if it is used in the right ways." -FGD_2_Student_2 "The lack of proper expertise. Someone well-versed should be there. The proper expertise should be there… Someone must be there who knows the importance of using all these things and who knows the drawbacks of using all things." -IDI_Faculty_3

Commonly emerged themes and subthemes

AI in medical education—understanding of AI

Participants’ understanding of AI varied significantly between faculty and students, highlighting disparities in exposure and familiarity with technology. Faculty members viewed AI as a sophisticated tool that could mimic cognitive abilities and enhance decision-making processes, reflecting their academic and clinical focus. In contrast, students associated AI with basic computer functionalities, indicating a more superficial understanding. Faculty perceived AI as a transformative tool capable of augmenting their teaching and research. Their advanced understanding likely stems from their involvement in curriculum development and exposure to AI-driven research tools.

"It (AI) is the development and application of the algorithm for mimicking human thinking and decision-making." -IDI_Faculty_5

Students’ limited understanding, often confined to practical tools like ChatGPT, reflects a narrower application-focused perspective. Their emphasis on tangible benefits, such as generating flashcards or mnemonics, highlights the potential of AI for simplifying routine learning tasks but also reveals a gap in understanding its broader implications.

"We use ChatGPT to get our answers within seconds. We can use it for generating mnemonics or flash cards." -FGD_1_Student_6

This gap underscores the need for differentiated AI education, tailored to the prior knowledge and professional context of each group. It also underscores the importance of foundational AI education for both students and faculty, with tailored content to address their specific needs.

AI in medical education -use of AI in education and research

Participants acknowledged AI’s potential to revolutionize both education and research, albeit in different ways. Faculty emphasized AI's role in enhancing conceptual understanding and enabling more effective lesson planning. For instance, they viewed AI as a tool to distill complex information into accessible formats. Faculty’s reliance on AI for planning reflects its potential to streamline their workload, allowing them to focus on higher-order teaching and mentoring activities. Students, on the other hand, highlighted AI’s utility for self-directed learning and research, particularly in finding resources and generating study aids. Their emphasis on efficiency and accessibility suggests that AI can democratize learning, particularly in resource-constrained settings. This dichotomy reveals an important opportunity for institutions to guide both groups in using AI responsibly, balancing efficiency with critical thinking and conceptual depth.

Ethical and moral considerations

Participants raised significant concerns about ethical issues, including data privacy and the potential misuse of AI. Faculty, in particular, emphasized the importance of accountability frameworks to safeguard against these risks.

"Making someone accountable is the way to deal with things in a better way… clear guidelines given to the learners about the legalities, the penalties they're going to face because of use of this AI or irrational use of the AI tool will make them accountable for their acts…" -IDI_Faculty_3.

Their suggestions for legal education and clear guidelines reflect a proactive approach to addressing these challenges, underscoring the importance of ethical literacy in AI education.

Role of institutional support

Both faculty and students stressed the critical role of institutional support in facilitating AI integration. This includes not only providing access to resources but also fostering an environment that encourages innovation and collaboration. Faculty emphasized the importance of validation and supervision, reflecting their concerns about maintaining academic standards and integrity. Students’ call for free institutional access to paid AI tools highlights the potential for institutions to bridge resource gaps and democratize access to technology.

Theme-specific insights: faculty perspectives

Limitations of AI in education

Faculty members raised significant concerns about over-reliance on AI, cautioning that excessive dependence could undermine students’ critical thinking and problem-solving skills. This reflects a broader global concern about AI potentially reducing the role of human judgment in education. Faculty framed AI as a supplementary tool rather than a replacement for traditional methods, emphasizing the irreplaceable value of human insight in teaching and learning. Their apprehension underscores the need for frameworks that integrate AI thoughtfully into medical education, ensuring that it complements rather than replaces critical pedagogical practices.

Future recommendations

Faculty stressed the importance of a phased approach to AI integration, beginning with foundational training for educators. This highlights their awareness of the challenges associated with incorporating new technologies into entrenched curricula. Their preference for integrating AI into existing modules rather than creating standalone courses suggests a practical approach to curriculum development, aiming to minimize disruptions while maximizing impact. Faculty also emphasized the need for institutional readiness, reflecting their recognition of systemic barriers that could hinder AI adoption, such as limited resources and resistance to change.

"To start with, in our context- we should educate the faculty first... then we must pursue it accordingly." -IDI_Faculty_5

Theme-specific insights: student perspectives

Barriers to AI integration

Students identified several barriers, including technological constraints, financial challenges, and limited accessibility to AI tools. Their concerns reflect the broader socio-economic challenges faced by students in resource-limited settings. The financial burden of accessing advanced AI tools, combined with the lack of institutional support, emerged as a significant obstacle. This highlights the need for institutions to invest in infrastructure and provide equitable access to resources.

So there are paid websites of AI tools which an individual cannot buy easily. So an institute can provide free institutional access to the students." - FGD_3_Student_3

Future recommendations

Students advocated for a practical, application-focused approach to AI education, emphasizing the importance of hands-on training. Their preference for basics reflects their current knowledge gaps and underscores the need for incremental learning pathways. This theme highlights the potential of AI to level the playing field for students, particularly in under-resourced settings. However, it also raises questions about the depth of engagement required to foster a nuanced understanding of AI's capabilities and limitations.

“Our undergraduate programs are already very extensive and if you will incorporate AI extensively in our curriculum students will not be able to grasp it. So I think sticking to the basics of AI and how to simply use it will be enough.” -FGD_1_Student_2.

Discussion

Our study aimed to address the gap in understanding the perceptions, attitudes, and practices of medical students and faculty regarding AI integration in medical education. In particular, we sought to explore these perspectives in the context of a resource-limited setting, where the integration of AI presents unique challenges, such as technological barriers and concerns around privacy. Our findings offer valuable insights into these areas and highlight key themes that emerged through our mixed methods approach.

Our results revealed a generally positive attitude toward AI among both students and faculty. However, significant knowledge gaps were identified, particularly in understanding AI subtypes like machine learning and deep learning. This lack of familiarity suggests that without foundational AI education, medical institutions in Pakistan may struggle to prepare their students for the rapidly evolving healthcare landscape. Similar findings have been reported in other developing regions, where limited access to advanced AI tools hinders both students and educators from fully integrating AI into their curricula [19].

While both students and faculty expressed generally positive attitudes toward AI integration, faculty members reported significantly higher mean composite attitude scores than students. This may reflect greater exposure to AI’s academic and research applications among faculty, and, therefore, faculty members' greater familiarity with the practical applications of AI in clinical practice and research. However, concerns about over-reliance on AI were prevalent among both groups, who stressed the importance of using AI as a complementary tool rather than a replacement for traditional methods. This aligns with global findings, where educators emphasize the need for balance to ensure that critical thinking and clinical judgment remain central to medical education [5].

Our study’s subgroup analyses revealed notable patterns, with older students reporting higher self-perceived knowledge of AI and more positive attitudes toward its integration in medical education. This could be attributed to increased maturity and a better understanding of AI’s long-term benefits. Gender differences were also evident, with male students reporting significantly higher knowledge scores than their female counterparts. Similar gender disparities in technology-related fields have been reported in previous studies [12], which could potentially stem from a historically lower level of representation and participation of women in technology-related fields, which may limit their exposure to AI and related topics. This also suggests the need for inclusive AI education strategies that address these gaps and foster equitable access to AI learning opportunities.

Ethical concerns were a recurring theme throughout the discussions. Faculty and students raised significant issues around data privacy, confidentiality, and the potential misuse of AI, reflecting concerns that have been widely reported in the literature [20, 21]. Both groups emphasized the need for institutional safeguards, including clear guidelines on AI usage, legal frameworks for accountability, and training programs to promote responsible AI integration. The consensus was that without robust ethical frameworks, AI could pose risks to academic integrity and patient confidentiality. These concerns are particularly relevant in developing countries, where regulatory oversight may be limited, making it imperative for institutions to proactively address these issues [22–24].

Furthermore, despite AI not yet formally integrated into the medical curriculum, participants highlighted its informal use in both teaching and learning practices. Students reported using AI tools to simplify learning tasks, such as generating mnemonics and flashcards, or for quick access to information. Faculty members mentioned leveraging AI for lesson planning and research, recognizing its potential to improve efficiency and accuracy. These findings demonstrate that, even in the absence of formal training, AI is becoming a valuable resource in medical education. However, this also underscores the need for structured integration to ensure the responsible use of AI, mitigate potential misuse, and maximize its educational benefits.

Our findings also highlight several barriers to AI integration, including financial and technological constraints, outdated curricula, and resistance to change. Suggested strategies to overcome these barriers included fostering openness to AI, increasing faculty readiness through targeted training, and promoting proactive attitudes toward technological adaptation [25–27].

Practical implications

While this study primarily focuses on gauging the perceptions of students and faculty, its findings offer important implications for the integration of AI in undergraduate medical curricula. A phased approach to curriculum integration could address existing gaps in knowledge and preparedness. Foundational workshops for faculty and students would help build a baseline understanding of AI concepts and tools. These workshops could later evolve into more advanced sessions as familiarity with AI increases. Faculty development remains a critical first step, as educators play a pivotal role in guiding students' engagement with AI. Training programs aimed at equipping faculty with practical skills and knowledge of AI tools would ensure effective teaching. Similarly, incremental integration of AI into existing modules, such as research methods or clinical reasoning, could complement existing curricula without overburdening students or faculty. Institutional readiness and support are paramount for successful implementation. Providing access to paid AI tools and resources, along with creating policies to standardize ethical use, would address many of the barriers highlighted in this study. Context-specific strategies, tailored to the challenges of resource-limited settings, must be prioritized to ensure equitable access to AI technologies. These findings lay the groundwork for future studies to explore and test the feasibility of these strategies. While our study highlights broad directions for AI integration, further research involving pilot programs across multiple institutions could provide a more comprehensive understanding of implementation in diverse settings.

Strengths and limitations

This study addresses a critical gap in the literature on AI integration in medical education, particularly within the South-East Asian region. The findings offer several practical implications. The mixed-methods approach of this study provided a more holistic view of the subject, allowing for a richer analysis of both qualitative and quantitative data. To our knowledge, this is the first study to conduct an in-depth qualitative exploration of both faculty and student perspectives on AI, alongside a quantitative assessment of their knowledge, attitudes, and practices. This comprehensive approach offers valuable insights into the multifaceted challenges and opportunities associated with AI integration in medical curricula.

However, several limitations must be acknowledged. The use of convenience sampling risks the introduction of selection bias, and along with the single center setting, may limit the generalizability of the findings to other institutions and settings. Regarding the survey instrument, a lack of construct validity assessment through factor analysis limits confidence in the instrument's ability to accurately measure the constructs. Additionally, the cross-sectional design limits the ability to infer causal relationships. Furthermore, a smaller sample of faculty participants responded to the survey as compared to the calculated required sample size, which may decrease the statistical power of the results. Despite these limitations, the mixed-methods design offers a well-rounded understanding of the perceptions and experiences of faculty and students regarding AI integration.

Future research should focus on longitudinal studies that track the evolution of knowledge, attitudes, and practices as AI becomes increasingly integrated into medical education. Expanding the scope of such research to include multiple institutions and regions would offer a broader range of perspectives. Additionally, identifying effective strategies to overcome the barriers to AI adoption, particularly in resource-constrained environments, will be crucial to advancing AI education globally.

Conclusion

This study highlights the generally positive attitudes of both medical students and faculty toward the integration of AI in medical education, while also identifying significant barriers that must be addressed for successful implementation. Although the potential of AI to enhance research and learning in medical education is well recognized, concerns about over-reliance on AI, privacy, and ethical issues persist. By exploring these perspectives in a resource-limited setting, the findings underscore the need for context-specific strategies, including foundational training, ethical frameworks, and institutional support, to enable effective integration. As AI continues to shape the future of medicine, these insights provide a foundation for further research and pilot initiatives to prepare healthcare professionals for an increasingly AI-driven landscape.

Abbreviations

AI

Artificial intelligence

CVR

Content validity ratio

FGD

Focused group discussions

IDI

In-depth interviews

Author’s contributions

MAN, SS, AA contributed to the conception and design of the study. MAN and SA were involved in the acquisition of quantitative data and conducting of the FGDs. MAN and SA contributed to the acquisition, analysis, and interpretation of the quantitative data, while MGRM and SS conducted qualitative analysis. MGRM and MAN contributed towards drafting of the manuscript. SS and AA contributed to the critical review of the work. All authors read and approved the final manuscript and agree to be personally accountable for all parts of the work.

Funding

None.

Data availability

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

Declarations

Ethics approval and consent to participate

Ethical approval was obtained from the Ethical Review Committee of Liaquat National Hospital and Medical College (App# 0936–2023-LNH-ERC). Following ERC approval, the objective of the study was communicated to both faculty and students, and separate consents were obtained for both the quantitative survey and qualitative interviews. The participants had freedom to decline participation or to drop out at any time. No personal identification was disclosed, and participant data was coded. At all times, the research team protected the privacy and confidentiality of the data.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.