Artificial Intelligence Adoption in Healthcare Practice and Research: A Cross‐Sectional Study of Knowledge, Attitudes, and Practices in Balochistan, Pakistan Authors: Zahra Ali, Humza Saeed, Tania Bahar, Muhammad Ali, Sangeen Khan Tareen, Muqaddas Bahar, Huma Ali, Saira khan, Achit Kumar Singh, Aqeel Shah Gillani, Raz Muhammad Bazai Published: 2026 ABSTRACT ABSTRACTBackground and AimsArtificial intelligence (AI) is a transformative technology in modern medicine, offering significant potential to enhance diagnostic accuracy and patient outcomes. In Pakistan, digital health integration varies across regions, with Balochistan remaining one of the most resource‐limited provinces. This study assessed knowledge, attitudes, and practices (KAP) regarding AI among healthcare professionals in Balochistan and explored sociodemographic and institutional factors associated with AI adoption.MethodsAn IRB‐approved cross‐sectional study was conducted using a KAP survey among healthcare professionals in Balochistan (N = 153). Data were collected through in‐person and online questionnaires. Descriptive statistics summarized KAP levels. Bivariate logistic regression analysis was performed to examine associations between participant characteristics and AI‐related KAP. Analyses were conducted using SPSS version 26.0, with p < 0.05 considered statistically significant.ResultsOverall, 72.5% of participants demonstrated good knowledge of AI, while 39.2% exhibited a positive attitude, and 58.8% reported good AI‐related practice. AI was most commonly used in clinical settings (60.1%), particularly in radiology (24.2%) and clinical decision support (26.1%). Participants with higher academic qualifications showed higher odds of good AI knowledge, although not statistically significant. Institutional affiliation was significantly associated with attitudes toward AI; participants from BMCH/Civil Hospital Quetta had lower odds of a positive attitude compared to those from QIMS/CMH Quetta (COR = 0.21; p = 0.008). Higher odds of good AI practice were observed among Medical Officers and participants with longer clinical experience, although not statistically significant.ConclusionDespite relatively high awareness of AI among healthcare professionals in Balochistan, positive attitudes and consistent practical use remain limited. Institutional factors influence perceptions of AI. Targeted educational initiatives, structured training programs, and institutional support are needed to strengthen trust and facilitate effective integration of AI into healthcare practice in resource‐limited settings.Keywords: artificial intelligence adoption, healthcare professionals, knowledge–attitude–practice survey, resource‐limited settings Introduction Artificial intelligence (AI) is increasingly recognized as a key component of digital health transformation and refers to computer‐based systems capable of performing tasks that typically require human intelligence [1]. In healthcare, AI applications are being integrated into medical education, diagnostic support, and clinical decision‐making [2]. Broadly, AI applications in medicine can be classified into virtual and physical domains. Virtual applications include electronic health records, clinical decision support systems, and diagnostic algorithms, while physical applications encompass robotic‐assisted surgery, smart prosthetics, and assistive technologies [1]. The largest application of AI is seen in radiology; however, its use has expanded into dermatology, ophthalmology, psychiatry, cardiology, oncology, neuroscience, pathology, and internal medicine [2, 3]. These developments highlight AI's growing role in improving diagnostic accuracy, workflow efficiency, and clinical decision‐making. Existing literature suggests that AI adoption is influenced by health system readiness, including infrastructure, workforce capacity, and digital literacy [4, 5]. While high‐income countries have increasingly integrated AI into clinical practice, implementation in low‐ and middle‐income countries remains limited [6]. This uneven adoption occurs not only across countries but also across healthcare facilities, specialties, and levels of technological infrastructure [7]. Previous studies have reported insufficient knowledge among healthcare professionals regarding AI applications, ethical concerns, data privacy, and reliability, which contribute to hesitancy toward adoption [8, 9]. Although AI Initiatives have been launched under presidential programs in Pakistan, their implementation within the health sector faces numerous obstacles. These include resistance from healthcare providers, financial constraints, a shortage of trained professionals to establish the diagnostic frameworks on which AI systems rely, limited data on public attitudes towards AI, and concerns about physicians being replaced, alongside sociocultural barriers, issues of confidentiality, and medicolegal risks [10, 11, 12]. While these barriers operate at a broader system level, they ultimately influence AI adoption by affecting frontline healthcare professionals, who serve as the primary users and implementers of AI‐enabled tools in clinical settings. As such, healthcare professionals' knowledge, attitudes, and practices (KAP) represent an essential component of health system readiness for AI integration. Balochistan, the largest and least resource‐developed province of Pakistan, faces persistent challenges related to healthcare access, workforce shortages, and infrastructure limitations. Although studies from other regions of Pakistan have reported variable levels of AI awareness and use among healthcare professionals, evidence from Balochistan remains scarce [13, 14, 15]. Understanding healthcare professionals' perceptions in this setting provides insight into user‐level readiness and potential implementation challenges, rather than directly measuring structural system capacity. Therefore, this study aimed to assess the KAP regarding AI among healthcare professionals in Balochistan as an indicator of preparedness for AI adoption within a resource‐limited healthcare system, and to examine sociodemographic and institutional factors associated with AI adoption. Methods Study Characteristic An IRB‐approved cross‐sectional study was conducted to assess KAP regarding AI in healthcare practice and research among healthcare professionals in Balochistan, Pakistan. The study included medical doctors working in public and private hospitals across the province, including BMCH/Civil Hospital Quetta, Jhalawan Medical College/Teaching Hospital Khuzdar, Loralai Teaching Hospital, QIMS/CMH Quetta, and other district‐ and tehsil‐level facilities. The majority of participants were recruited from Civil Hospital Quetta and Bolan Medical Complex Hospital (BMCH), which are the largest tertiary care and postgraduate teaching hospitals in the province and serve as major referral centers managing complex, emergency, and trauma cases. These institutions represent a substantial proportion of the active medical workforce in Balochistan, thereby providing access to healthcare professionals with broad clinical exposure. Additionally, participants from district‐ and tehsil‐level hospitals were included to capture perspectives across different levels of care and geographic locations. While the distribution of participants reflects the concentration of healthcare professionals in tertiary centers within the province, this may limit the generalizability of findings to underrepresented peripheral facilities. Sample Size and Sampling Technique The study population included medical doctors of varying professional roles, including Residents, Medical Officers, Consultants, and Private Practitioners, irrespective of specialty or years of clinical experience. The sample size was 153, calculated using a precision level of 8%, a 95% confidence interval, and an assumed prevalence of 63.5% for moderate‐level AI knowledge among healthcare providers, based on available literature [13, 14]. An additional 10% was added to account for potential non‐response. Data were collected using a nonprobability convenience sampling approach, with questionnaires distributed electronically via WhatsApp and through in‐person contact to maximize participation across institutions. Validation of an Instrument The questionnaire is derived from previously published studies and a thorough literature review [10]. The validity and reliability of the questionnaire were rigorously analyzed and assessed by two medical educationists from Bolan Medical College, Quetta. It verified that the content, criteria, relevancy, and validity were appropriate for assessing AI knowledge, attitudes, practices, and among medical students. Minor changes in the questionnaire were made to ensure better validity, as the core content was derived from Pakistani and Jordanian studies [11, 16]. Data Collection and Management A structured survey‐based questionnaire was used for data collection. The questionnaire consisted of four sections [1]: sociodemographic characteristics, including age, gender, years since graduation, years of clinical experience, current professional role, specialty, and hospital or institutional affiliation [2]; knowledge assessing basic awareness of AI and its integration in healthcare practice and research [3]; attitudes toward AI assessed using Likert‐scale items addressing perceived benefits, challenges, and ethical considerations; and [4] practice assessed through frequency‐based questions evaluating the use of AI in healthcare practice and research. A systematic approach to data management was employed to ensure data integrity, accessibility, and long‐term preservation. Each outcome variable (knowledge, attitude, and practice) was predefined and dichotomized for analysis. Knowledge was categorized as good (scores 6–11) or poor (scores 0–5) using a 14‐point scale, with one point assigned to each item. Attitude was assessed using a 15‐point scale and categorized as positive (scores 8–15) or negative (scores 0–7). Practice was evaluated using a 9‐point scale, with scores of 6–10 indicating good practice and 0–5 indicating poor practice. Details of item‐level scoring are provided in Supporting Information: Table S1, while the score thresholds used to classify each domain into binary outcomes are summarized in Supporting Information: Table S2. Statistical Analysis Descriptive statistics were used to summarize participant characteristics and KAP levels, presented as frequencies and percentages. Bivariate logistic regression analysis was performed as a prespecified analytical approach to assess the association between sociodemographic and institutional variables (age, years since graduation, professional role, workplace type, and institutional affiliation) and the outcomes of good knowledge, positive attitude, and good practice. Results were reported as crude odds ratios (CORs) with corresponding 95% confidence intervals (CIs). All statistical tests were two‐sided, and a p‐value < 0.05 was considered statistically significant. Due to the limited sample size and the small number of statistically significant predictors, multivariable regression analysis was not performed. All analyses were conducted using IBM SPSS Statistics (Version 26.0; IBM Corp., Armonk, NY, USA), in accordance with SAMPL reporting recommendations. Results Baseline Characteristics of Participants A total of 153 healthcare professionals participated in the study. The majority of respondents were aged 25–30 years (85/153, 55.6%), followed by 31–35 years (29/153, 19.0%). Male participants constituted 78/153 (51.0%), while 75/153 (49.0%) were female. More than half of the participants (85/153, 55.6%) had graduated within the last 5 years, and 95/153 (62.1%) reported less than 5 years of clinical experience. Regarding educational qualifications, 71/153 (46.4%) held an MBBS degree, 65/153 (42.5%) had FCPS qualifications, and 11/153 (7.2%) held MCPS/MRCP/FRCS degrees. A smaller proportion (6/153, 3.9%) possessed higher academic qualifications, including MS, MPhil, PhD, MPH, or equivalent. Residents comprised the largest professional group (56/153, 36.6%), followed by Medical Officers (47/153, 30.7%) and Consultants (28/153, 18.2%). Most participants were employed in teaching hospitals (71/153, 46.4%), followed by public hospitals (54/153, 35.3%). Over half (83/153, 54.2%) were affiliated with BMCH/Civil Hospital Quetta, while 20/153 (13.1%) were from QIMS/CMH Quetta (Table 1). Table 1Sociodemographic characteristics of participants.VariableCategory n (%)Age (years)20–2418 (11.8)25–3085 (55.6)31–3529 (19.0)36–409 (5.9)41–507 (4.6)> 505 (3.3)GenderMale78 (51.0)Female75 (49.0)Years since graduation< 585 (55.6)5–1045 (29.4)11–1512 (7.8)16–206 (3.9)> 205 (3.3)Highest level of educationMBBS71 (46.4)FCPS65 (42.5)MCPS/MRCP/FRCS11 (7.2)Othera 6 (3.9)Years of clinical experience< 595 (62.1)5–1033 (21.6)11–1515 (9.8)16–206 (3.9)> 204 (2.6)Current professional roleResident Doctor56 (36.6)Medical Officer47 (30.7)Consultant28 (18.2)Private Practitioner14 (9.2)Otherb 8 (5.2)SpecialtyInternal Medicine56 (36.6)Surgery38 (24.8)Radiology9 (5.9)Pathology5 (3.3)Otherc 45 (29.4)Workplace typePublic hospital54 (35.3)Private hospital21 (13.7)Teaching hospital71 (46.4)Clinic7 (4.6)InstitutionBMCH/Civil Hospital Quetta83 (54.2)QIMS/CMH Quetta20 (13.1)Loralai Teaching Hospital24 (15.7)Jhalawan Medical College/TH Khuzdar4 (2.6)Otherd 22 (14.4) a Other education includes MS, MPhil, PhD, MPH, BDS, and subspecialty FCPS. b Other roles include associate professors and students. c Other specialties include gynecology/obstetrics, pediatrics, neurology, cardiology, dermatology, and others. d Other institutions include DHQ Pishin, THQ Khanozai, Lasbela Hospital, THQ Surab, and similar facilities. Sociodemographic characteristics of participants. Other education includes MS, MPhil, PhD, MPH, BDS, and subspecialty FCPS. Other roles include associate professors and students. Other specialties include gynecology/obstetrics, pediatrics, neurology, cardiology, dermatology, and others. Other institutions include DHQ Pishin, THQ Khanozai, Lasbela Hospital, THQ Surab, and similar facilities. Knowledge of AI in Healthcare Based on predefined scoring thresholds (Supporting Information: Table S2), 111/153 (72.5%) of participants demonstrated good knowledge of AI in healthcare, while 42/153 (27.5%) had poor knowledge (Figure 1). Nearly half of the respondents (76/153, 49.7%) reported being moderately to very familiar with AI, although only 1/153 (0.7%) identified as experts. Knowledge of AI subdomains varied: 74/153 (48.4%) were aware of machine learning and deep learning, and 98/153 (64.1%) reported familiarity with key AI technologies in healthcare. A majority correctly identified machine learning as a system that learns from data and improves over time (102/153, 66.7%) and recognized AI as a supportive tool assisting medical professionals rather than replacing them (125/153, 81.7%). Awareness of AI applications was highest in radiology and imaging analysis (62/153, 40.5%), followed by medical education (40/153, 26.1%) and pathology (21/153, 13.7%). Ethical concerns related to AI, including patient data privacy, were acknowledged by 109/153 (71.2%) of participants (Table 2). Bar chart showing the proportion of participants with good and poor knowledge, positive and negative attitudes, and good and poor practice related to AI in healthcare among healthcare professionals in Balochistan (N = 153). Knowledge, attitude, and practice levels were categorized using predefined scoring criteria (see Supporting Information: Table S2). Values are presented as percentages of participants in each category. Table 2Knowledge of artificial intelligence in healthcare.VariableResponse n (%)Familiarity with AI in healthcareNot familiar at all14 (9.2)Slightly familiar62 (40.5)Moderately familiar57 (37.3)Very familiar19 (12.4)Expert1 (0.7)Knowledge of machine learning (ML) and deep learning (DL)Yes74 (48.4)No79 (51.6)Awareness of key AI technologies in healthcareYes98 (64.1)No55 (35.9)Understanding of machine learningLearns from data and improves over time102 (66.7)Fixed programmed instructions34 (22.2)Used only in robotics10 (6.5)Other/unsure7 (4.6)Best description of AI in healthcareAssists clinicians in diagnosis and decision‐making125 (81.7)Fully replaces doctors7 (4.6)Performs surgeries independently9 (5.9)Limited to medical imaging12 (7.8)Awareness of AI‐assisted robotic surgeryYes111 (72.5)No42 (27.5)AI use in medical imaging (X‐ray, CT, MRI)Yes121 (79.1)No32 (20.9)AI efficiency in disease predictionYes93 (60.8)No60 (39.2)AI use in drug discovery and personalized medicineYes112 (73.2)No41 (26.8)Awareness of ethical concerns (e.g., data privacy)Yes109 (71.2)No44 (28.8)Most widely implemented AI specialtyRadiology/imaging62 (40.5)Medical education40 (26.1)Clinical diagnosis28 (18.3)Pathology21 (13.7)Patient–doctor communication2 (1.3) Knowledge of artificial intelligence in healthcare. Attitudes Toward AI Using predefined attitude score criteria (Supporting Information: Table S2), 60/153 (39.2%) of respondents demonstrated a positive attitude toward AI, while 93/153 (60.8%) exhibited a negative attitude (Figure 1). Less than half of participants agreed that AI could improve diagnostic accuracy and treatment planning (65/153, 42.5%), and 71/153 (46.4%) believed AI could reduce the workload of healthcare professionals. Confidence in trusting AI tools for interpreting medical imaging was limited, with only 50/153 (32.7%) expressing agreement. Support for regulatory oversight was common, with 78/153 (50.9%) agreeing that AI should be regulated before widespread implementation. A majority supported the inclusion of AI education in medical training (78/153, 51.0%) and agreed that the ethical implications of AI should be understood by physicians (85/153, 55.6%). Importantly, 96/153 (62.7%) disagreed that AI could fully replace doctors in complex clinical decision‐making (Table 3). Table 3Attitudes toward artificial intelligence in iealthcare.StatementStrongly agree n (%)Agree n (%)Neutral n (%)Disagree n (%)Strongly disagree n (%)AI can improve diagnostic accuracy and treatment planning8 (5.2)57 (37.3)52 (34.0)11 (7.2)25 (16.3)AI can reduce the workload of healthcare professionals10 (6.5)61 (39.9)45 (29.4)26 (17.0)11 (7.2)AI can increase patient trust in healthcare decisions4 (2.6)32 (20.9)62 (40.5)38 (24.8)17 (11.1)AI will make healthcare more cost‐effective6 (3.9)41 (26.8)56 (36.6)29 (19.0)21 (13.7)I trust AI tools for interpreting medical images (x‐ray, CT, MRI)7 (4.6)43 (28.1)65 (42.5)29 (19.0)9 (5.9)AI should be regulated before widespread clinical implementation23 (15.0)55 (35.9)47 (30.7)18 (11.8)10 (6.5)Doctors should receive formal training in AI tools18 (11.8)60 (39.2)37 (24.2)23 (15.0)15 (9.8)AI will be required in future medical practice9 (5.9)62 (40.5)49 (32.0)21 (13.7)12 (7.8)Physicians must understand the ethical implications of AI24 (15.7)61 (39.9)50 (32.7)16 (10.5)13 (8.5)Clinical AI will be more accurate than physicians4 (2.6)22 (14.4)50 (32.7)53 (34.6)24 (15.7)AI threatens job security in some medical specialties11 (7.2)42 (27.5)45 (29.4)38 (24.8)17 (11.1)AI will increase diagnostic errors10 (6.5)34 (22.2)59 (38.6)35 (22.9)15 (9.8)AI can improve hospital administrative efficiency15 (9.8)55 (35.9)53 (34.6)20 (13.1)10 (6.5)Medical curricula should include AI training19 (12.4)45 (29.4)47 (30.7)32 (20.9)10 (6.5)AI can fully replace doctors in complex decision‐making0 (0.0)23 (15.0)34 (22.2)51 (33.3)45 (29.4) Attitudes toward artificial intelligence in iealthcare. Practice of AI in Healthcare According to predefined practice score thresholds (Supporting Information: Table S2), 90/153 (58.8%) of participants demonstrated good AI‐related practice, while 63/153 (41.2%) reported poor practice (Figure 1). More than half (92/153, 60.1%) reported using AI‐based tools in clinical practice. The most commonly reported applications were AI‐powered clinical decision support systems (40/153, 26.1%) and radiology or imaging‐based AI tools (37/153, 24.2%). However, regular use remained limited; only 20/153 (13.1%) reported frequent use, while 50/153 (32.7%) reported occasional use. In research settings, 107/153 (69.9%) of respondents reported using AI tools, and 104/153 (68.0%) considered AI very or somewhat useful for enhancing research quality and efficiency. Despite this, 99/153 (64.7%) had not received any formal training in AI, and only 33/153 (21.6%) reported having access to AI‐based tools in their workplace. The most frequently cited barriers to AI adoption were lack of AI training programs (39/153, 25.5%), lack of awareness (38/153, 24.8%), and ethical or legal concerns (35/153, 22.9%) (Table 4). Table 4Practice of artificial intelligence in healthcare.VariableCategory n (%)Use of AI in clinical practiceYes92 (60.1)No61 (39.9)Type of AI tools used in clinical practicea Radiology/Imaging AI37 (24.2)AI‐powered clinical decision support40 (26.1)AI in pathology/laboratory diagnostics23 (15.0)AI‐based electronic health records14 (9.2)Other AI tools2 (1.3)Not applicable36 (23.5)Frequency of AI use in clinical practiceNever37 (24.2)Rarely46 (30.1)Sometimes50 (32.7)Often20 (13.1)Perceived effectiveness of AI for patient careVery effective18 (11.8)Somewhat effective80 (52.3)Neutral44 (28.8)Ineffective6 (3.9)Very ineffective5 (3.3)Use of AI for research purposesYes107 (69.9)No46 (30.1)Perceived usefulness of AI in medical researchVery useful33 (21.6)Somewhat useful71 (46.4)Neutral40 (26.1)Not very useful8 (5.2)Not useful at all1 (0.7)Barriers to AI adoptionb Lack of AI training programs39 (25.5)Lack of awareness38 (24.8)Ethical and legal concerns35 (22.9)High cost of implementation19 (12.4)Lack of facilities in hospitals14 (9.2)Resistance from medical professionals8 (5.2)Formal AI training receivedYes54 (35.3)No99 (64.7)Access to AI tools at workplaceYes33 (21.6)No95 (62.1)Not applicable25 (16.3) a Multiple responses allowed. b Percentages may not total 100% due to multiple selections. Practice of artificial intelligence in healthcare. Multiple responses allowed. Percentages may not total 100% due to multiple selections. Factors Associated with Knowledge, Attitude, and Practice of AI Age No statistically significant association was observed between age groups and good knowledge of AI (Table 5). Participants aged 31–35 years showed higher odds of having a positive attitude toward AI compared to those aged 20–24 years (COR = 1.88, 95% CI: 0.83–4.25, p = 0.13), although this association was not statistically significant. Similarly, no significant association was observed between age and good AI practice across age categories, although participants aged above 50 years showed higher odds of good practice (COR = 3.20, 95% CI: 0.30–34.59, p = 0.34). Table 5Factors associated with knowledge, attitude, and practice of artificial intelligence (bivariate logistic regression).VariableCategoryGood knowledge COR (95% CI) p‐valuePositive attitude COR (95% CI) p‐valueGood practice COR (95% CI) p‐valueAge (years)20–24Reference—Reference—Reference—25–301.15 (0.37–3.55)0.810.69 (0.36–1.33)0.270.56 (0.20–1.56)0.2731–350.54 (0.13–2.22)0.401.88 (0.83–4.25)0.130.31 (0.09–1.05)0.0636–401.30 (0.23–7.32)0.772.20 (0.52–7.86)0.310.64 (0.13–3.20)0.5941–500.43 (0.04–4.57)0.490.61 (0.11–3.23)0.560.32 (0.34–9.55)0.24> 501.73 (0.22–13.70)0.601.03 (0.17–6.38)0.973.20 (0.30–34.59)0.34Years since graduation< 5Reference—Reference—Reference—5–100.64 (0.25–1.66)0.360.94 (0.42–2.11)0.880.66 (0.29–1.51)0.3211–150.87 (0.26–2.97)0.820.53 (0.18–1.60)0.260.88 (0.29–2.66)0.8216–201.20 (0.21–6.91)0.843.05 (0.34–27.17)0.320.66 (0.12–3.77)0.64> 200.80 (0.08–8.00)0.850.61 (0.08–4.52)0.633.95 (0.40–39.38)0.24Education levelMBBSReference—Reference—Reference—FCPS1.48 (0.71–3.07)0.300.99 (0.50–1.96)0.982.56 (0.52–12.71)0.25MCPS/MRCP/FRCS0.37 (0.05–2.69)0.322.97 (0.60–14.72)0.18——Higher academic degree2.76 (0.33–23.14)0.350.22 (0.02–2.22)0.205.00 (0.42–59.0)0.20Clinical experience (years)< 5Reference—Reference—Reference—5–101.22 (0.55–2.69)0.630.86 (0.41–1.81)0.690.75 (0.36–1.59)0.4511–150.90 (0.22–3.61)0.880.57 (0.17–1.93)0.370.98 (0.29–3.31)0.9616–200.54 (0.06–4.84)0.521.15 (0.20–6.63)0.880.68 (0.12–3.92)0.68> 200.67 (0.07–6.35)0.730.38 (0.06–2.42)0.315.44 (0.58–50.72)0.14Professional roleResidentReference—Reference—Reference—Consultant1.20 (0.43–3.32)0.731.47 (0.56–3.82)0.430.78 (0.29–2.09)0.62Medical Officer1.55 (0.66–3.64)0.321.03 (0.47–2.26)0.952.03 (0.92–4.50)0.08Private practitioner0.82 (0.20–3.36)0.780.93 (0.28–3.04)0.902.60 (0.79–8.57)0.12InstitutionQIMS/CMH QuettaReference—Reference—Reference—BMCH/Civil Hospital1.38 (0.46–4.17)0.57 0.21 (0.06–0.66) 0.008 1.35 (0.50–3.66)0.56Loralai Teaching Hospital2.04 (0.56–7.45)0.280.84 (0.20–3.62)0.822.53 (0.76–8.44)0.13Jhalawan Medical College—— 0.07 (0.01–0.91) 0.04 2.14 (0.25–18.50)0.49Other institutions0.85 (0.19–3.79)0.830.67 (0.15–2.94)0.591.75 (0.50–6.17)0.38 Note: Bold values denote significant association for the variable where p < 0.05. Factors associated with knowledge, attitude, and practice of artificial intelligence (bivariate logistic regression). Note: Bold values denote significant association for the variable where p < 0.05. Years Since Graduation No statistically significant associations were found between years since graduation and good knowledge, positive attitude, or good practice related to AI (Table 5). Participants who graduated 16–20 years ago demonstrated higher odds of positive attitude toward AI (COR = 3.05, 95% CI: 0.34–27.17, p = 0.32), though this finding did not reach statistical significance. Highest Level of Education Participants with FCPS qualifications had higher odds of good knowledge of AI compared to MBBS graduates (COR = 1.48, 95% CI: 0.71–3.07, p = 0.30), although the association was not statistically significant. Those with higher academic qualifications (MS, MPhil, or PhD) also demonstrated higher odds of good knowledge (COR = 2.76, 95% CI: 0.33–23.14, p = 0.35). No statistically significant associations were observed between education level and positive attitude toward AI or good AI practice (Table 5). Years of Clinical Experience No significant associations were observed between years of clinical experience and good knowledge or positive attitude toward AI (Table 5). Participants with more than 20 years of clinical experience showed higher odds of good AI practice compared to those with less than 5 years of experience (COR = 5.44, 95% CI: 0.58–50.72, p = 0.14), although this association was not statistically significant. Current Professional Role Medical Officers demonstrated higher odds of good AI practice compared to Resident Doctors (COR = 2.03, 95% CI: 0.92–4.50, p = 0.08), though this association did not reach statistical significance. No statistically significant associations were found between professional role and good knowledge or positive attitude toward AI (Table 5). Workplace Type No statistically significant associations were observed between workplace type (public hospital, private hospital, teaching hospital, or clinic) and good knowledge, positive attitude, or good AI practice (p > 0.05 for all comparisons) (Table 5). Name of Hospital/Institution Institutional affiliation was significantly associated with attitude toward AI. Compared to participants from QIMS/CMH Quetta, those affiliated with BMCH/Civil Hospital Quetta had significantly lower odds of demonstrating a positive attitude toward AI (COR = 0.21, 95% CI: 0.06–0.66, p = 0.008). Similarly, participants from Jhalawan Medical College/Teaching Hospital Khuzdar had significantly lower odds of a positive attitude toward AI (COR = 0.07, 95% CI: 0.01–0.91, p = 0.04). No statistically significant associations were observed between institutional affiliation and good knowledge or good AI practice (Table 5). Discussion In this cross‐sectional study assessing KAP regarding AI among healthcare professionals in Balochistan, the majority of participants demonstrated good knowledge of AI (72.5%); however, positive attitudes toward AI were less prevalent (39.2%), and just over half reported good AI‐related practices (58.8%). While awareness of AI applications and ethical concerns was relatively high, access to formal AI training and the availability of AI‐based tools within institutions remained limited. Institutional affiliation was significantly associated with attitudes toward AI, whereas most sociodemographic and professional characteristics were not statistically associated with knowledge, attitudes, or practices. These findings highlight a disconnect between AI knowledge and its attitudinal acceptance and practical integration in healthcare settings. Our findings indicate that a substantial proportion of participants possessed good knowledge regarding AI in healthcare are consistent with prior studies showing a high level of interest and recognition of AI competency among healthcare professionals in clinical practice [17]. This aligns with a KAP survey conducted in South Gujarat, which reported that approximately 90% of healthcare professionals were fully or partially aware of AI [16]. In contrast, studies from Pakistan have reported more modest knowledge levels, with 56.1% of participants recognizing AI's role in diagnostics and treatment planning [18]. Similarly, research from Mardan, Khyber Pakhtunkhwa, found that although 70% of respondents were familiar with the term AI, only 19%–23% could correctly identify its medical applications [19]. A similar pattern has been observed in other low‐ and middle‐income settings. A recent study from Bangladesh reported that while technological awareness and readiness for AI are increasing, structural and perceptual barriers, such as concerns regarding risk, trust, and system‐level readiness, continue to limit effective adoption [20]. These findings are consistent with our results and suggest a broader regional trend in which growing awareness of AI is not yet matched by sufficient institutional capacity, infrastructure, and clinician confidence to support its integration into routine healthcare practice. In the present study, 71.2% of participants were aware of ethical concerns related to AI, particularly issues surrounding patient data privacy. This finding contrasts sharply with results from Gujarat, where awareness of ethical guidelines was considerably lower [16]. This level of awareness may also reflect growing recognition of broader governance challenges associated with AI in healthcare, including concerns related to transparency, accountability for AI‐assisted decisions, and responsible integration into clinical practice. However, findings from Saudi Arabia, knowledge of specific AI subtypes and their clinical implementation remained limited, likely reflecting a lack of structured, formal training in AI across medical education and professional development programs [21]. Participants with higher academic qualifications (PhD, MPhil, or MS) showed higher odds of possessing good AI knowledge compared to those with an MBBS degree, although this association did not reach statistical significance. This trend may reflect greater academic exposure, research involvement, and opportunities to engage with emerging technologies among those pursuing advanced degrees. Likewise, participants aged 36–40 years demonstrated higher odds of good knowledge compared to younger counterparts, possibly due to broader clinical experience and enhanced ability to contextualize AI applications in practice. Despite relatively strong knowledge levels, attitudes toward AI were largely cautious, with only 39.2% of participants demonstrating a positive attitude. This gap between knowledge and acceptance reflects a broader socio‐technical challenge in AI integration, where awareness alone is insufficient to drive adoption. In such contexts, clinician trust is not determined solely by knowledge but is shaped by the interaction between system‐level factors, including algorithmic transparency, accountability structures, and opportunities for experiential validation. When clinicians are unable to interpret how AI systems generate outputs (“black‐box” limitations), or when responsibility for AI‐assisted decisions remains unclear, uncertainty is introduced into clinical workflows, which may reduce trust and willingness to adopt these tools in practice [22, 23]. In our study setting, this dynamic is further compounded by limited institutional support, including restricted access to AI tools and formal training opportunities. As a result, healthcare professionals may possess theoretical awareness of AI but lack the practical exposure required to develop confidence in its clinical reliability. This suggests that the observed disconnect between knowledge and attitudes is not simply an individual‐level issue, but reflects a misalignment between technological awareness and system‐level readiness for integration. These findings align with broader socio‐technical perspectives on AI adoption, which emphasize that effective integration depends on the co‐development of technical systems, governance frameworks, and professional practices, rather than knowledge alone [24, 25, 26]. These findings can be further interpreted within emerging frameworks of responsible and context‐sensitive AI deployment in healthcare, which conceptualize AI adoption as a socio‐technical process shaped by the interaction between technological systems, institutional capacity, and governance structures rather than as a function of awareness alone [27, 28]. Within these frameworks, AI systems are not implemented in isolation but operate within interconnected domains that include technical infrastructure, organizational practices, and external governance conditions. In this context, adoption depends on how these domains align to support both functional integration and institutional legitimacy [27, 29]. Applying this perspective to our findings, the observed gap between knowledge (72.5%) and positive attitudes (39.2%) can be interpreted as a consequence of misalignment between individual‐level awareness and system‐level capacity. While participants demonstrate familiarity with AI concepts, limited access to tools, training, and governance structures constrains opportunities for experiential validation and trust formation. This interpretation suggests that increasing awareness alone is insufficient for adoption unless it is accompanied by parallel development of institutional infrastructure, governance mechanisms, and workforce capacity. Consistent with these frameworks, our findings indicate that effective AI integration requires coordinated development across technical, organizational, and regulatory domains to enable context‐sensitive and sustainable implementation in resource‐limited healthcare settings [25]. Beyond knowledge and attitudes, increasing attention has been directed toward the real‐world deployment of AI in healthcare and the ethical, professional, and governance challenges that accompany its adoption. AI systems are already being used in clinical settings for tasks such as radiologic image interpretation, risk prediction, triage, and clinical decision support. These technologies have demonstrated potential to improve diagnostic accuracy, optimize workflow efficiency, and support personalized treatment planning across multiple specialties. However, empirical studies highlight that the translation of AI tools from research environments to routine clinical practice remains limited due to concerns related to reliability, transparency, and integration into existing healthcare systems [24]. The ethical implications of AI in healthcare are particularly significant. Systematic reviews have identified key concerns, including algorithmic bias, data privacy, transparency of decision‐making processes, accountability for clinical outcomes, and patient autonomy. Algorithms trained on nonrepresentative datasets may introduce or exacerbate disparities in healthcare delivery, while the “black‐box” nature of many machine learning systems can limit clinicians' ability to interpret and trust AI‐generated recommendations [25, 30]. Furthermore, governance frameworks emphasize the need for structured oversight mechanisms, including ethics committees, regulatory evaluation processes, and continuous monitoring of AI systems throughout their lifecycle [26]. These challenges underscore the importance of developing robust institutional policies, transparent validation procedures, and interdisciplinary collaboration between clinicians, data scientists, and policymakers to ensure the safe and responsible integration of AI into healthcare practice. Institutional affiliation emerged as a significant determinant of attitudes toward AI, with participants from BMCH/Civil Hospital Quetta and Jhalawan Medical College demonstrating substantially lower odds of a positive attitude compared to those from QIMS/CMH Quetta. This difference is unlikely to reflect individual‐level variation alone and instead suggests that institutional environments actively shape clinicians' perceptions of AI. In settings with greater access to digital infrastructure, structured training programs, and exposure to technology‐supported workflows, clinicians are more likely to interact with AI tools in practice, which facilitates familiarity, perceived reliability, and ultimately trust in these systems [24, 30, 31]. In contrast, in resource‐constrained institutional environments, where access to AI tools is limited and training opportunities are scarce, AI remains largely theoretical. This lack of experiential exposure may hinder clinicians' ability to evaluate the reliability and clinical utility of AI systems, thereby reinforcing uncertainty and cautious attitudes. These findings indicate that regional differences in AI literacy and attitudes are not merely descriptive variations but reflect underlying disparities in institutional capacity and system readiness. From a socio‐technical perspective, this suggests that AI adoption is embedded within organizational contexts, where infrastructure, training, and governance mechanisms interact to shape both perception and use [25, 27]. In this study setting, the observed institutional differences likely represent a gap between awareness and implementation capacity, highlighting the importance of context‐specific capacity‐building strategies to enable effective AI integration in resource‐limited healthcare systems. Although positive attitudes were limited, more than half of the participants demonstrated good AI‐related practice. This aligns with studies from Gujarat and Saudi Arabia, where overall practice scores were also reported to be favorable [16, 21]. In contrast, studies from other regions of Pakistan, including Mardan, have documented substantially lower levels of AI utilization, highlighting notable regional differences in adoption [19]. In our cohort, AI use was primarily focused on technical applications, including clinical decision support systems (26.1%), radiology/imaging AI (24.2%), pathology/laboratory diagnostics (15.0%), and AI‐based electronic health records (9.2%), with very few using other AI tools (1.3%) and nearly one‐quarter reporting no practical engagement (23.5%). This pattern is consistent with global trends in AI deployment, where hands‐on exposure remains concentrated in a few key applications. Despite moderate practice levels, frequent use of AI tools remained uncommon, with only 13.1% reporting regular use. This pattern mirrors findings from Gujarat, where occasional rather than routine use predominated [16]. Limited access to formal training and institutional AI infrastructure likely contributes to this gap. Notably, over one‐third of participants in our study reported receiving formal AI training, which may partly explain the higher practice levels observed compared to other Pakistani studies. Medical Officers demonstrated higher odds of good AI practice compared to Resident Doctors, possibly reflecting greater autonomy in clinical decision‐making. Additionally, participants with over 20 years of clinical experience showed a trend toward higher AI utilization, which may reflect confidence in integrating AI as a supplementary tool rather than perceiving it as a threat to clinical expertise. These findings have broader implications for healthcare policy and workforce development, particularly in resource‐limited settings, and align with existing evidence emphasizing the role of system‐level readiness, governance, and workforce capacity for successful AI implementation in healthcare [30, 31, 32, 33]. The observed gap between relatively high knowledge (72.5%) and lower positive attitudes (39.2%) suggests that awareness alone is insufficient to support effective AI integration, highlighting the need for policies that prioritize not only education but also experiential training and system‐level support. Additionally, the high proportion of participants without formal AI training (64.7%) and limited access to AI tools in clinical settings (21.6%) indicates a substantial workforce readiness gap that cannot be addressed through theoretical instruction alone. From a policy perspective, these findings underscore the importance of investing in institutional infrastructure, including access to validated AI tools and the integration of AI into routine clinical workflows, to enable hands‐on exposure and build clinician confidence. Furthermore, the significant influence of institutional affiliation on attitudes toward AI suggests that targeted, institution‐specific capacity‐building strategies may be more effective than uniform national approaches [25]. Strengthening governance frameworks, including clear guidelines on accountability and ethical use, is also essential to address uncertainty and support responsible AI implementation. Collectively, these findings suggest that successful AI adoption requires coordinated efforts across workforce training, infrastructure development, and governance, emphasizing that policy interventions must be aligned with local institutional capacity to ensure equitable and context‐sensitive implementation. This study has several limitations. Its cross‐sectional design precludes causal inference, and the use of convenience sampling may limit generalizability beyond the study population. Additionally, the use of self‐reported survey data may introduce social desirability bias, potentially leading to overestimation of participants' knowledge, attitudes, or practices. Furthermore, the binary categorization of KAP measures may oversimplify responses and limit the ability to capture more nuanced perspectives, particularly among participants with moderate or neutral views. While this study identifies key barriers such as limited access to AI training and restricted availability of AI tools, it does not allow for in‐depth exploration of the underlying reasons for observed attitudes or the influence of operational and institutional factors, such as workload, infrastructure, and resource variability across settings. Future research should address these limitations by employing longitudinal designs, larger and more representative samples, and mixed‐methods or qualitative approaches to better understand these underlying dynamics. The use of multi‐level or continuous scoring systems may also provide a more detailed assessment of variations in KAP. Despite these limitations, this study provides one of the first comprehensive assessments of AI‐related KAP among healthcare professionals in Balochistan and offers valuable baseline data to inform future research and policy development in resource‐limited settings. Conclusion This study demonstrates that although healthcare professionals exhibit adequate baseline knowledge of AI, this knowledge does not consistently translate into positive attitudes or routine clinical use. Key factors such as training exposure, institutional affiliation, and professional experience influence AI‐related readiness. These findings highlight the need for structured educational initiatives and institutional support to enable effective and responsible integration of AI into healthcare practice. The results provide foundational evidence to guide future interventions and research in this setting. Author Contributions Zahra Ali: conceptualization, investigation, writing – original draft, writing – review and editing, methodology, software, formal analysis, data curation, supervision. Humza Saeed: methodology, software, formal analysis, writing – review and editing, Writing – original draft, visualization, validation. Tania Bahar: methodology, software, formal analysis, writing – review and editing, writing – original draft, visualization. Muhammad Ali: methodology, validation, visualization, writing – review and editing, writing – original draft, formal analysis. Sangeen Khan Tareen: data curation, software, methodology, writing – original draft, writing – review and editing, formal analysis. Muqaddas Bahar: software, data curation, supervision, writing – review and editing, visualization, validation, writing – original draft. Huma Ali: writing – original draft, writing – review and editing, validation, methodology, software, formal analysis. Saira Khan: software, formal analysis, writing – review and editing, writing – original draft, Visualization. Achit Kumar Singh: writing – review and editing, writing – original draft, visualization, investigation, formal analysis, software. Aqeel Shah Gillani: project administration, writing – review and editing, writing – original draft; validation, methodology. Raz Muhammad Bazai: conceptualization, writing – original draft, writing – review and editing, visualization, project administration, supervision. Funding No funding was received for this study. Therefore, no funding source had any role in the study design; data collection, analysis, or interpretation; manuscript writing; or the decision to submit the manuscript for publication. Ethics Statement Ethical approval for this study was obtained from the ethics committee of Bolan Medical College, Quetta, Pakistan (IRB Approval Number: 228‐29/BMC/IRB/25). Consent Written informed consent was obtained from all participants before they participated in the study. Conflicts of Interest The authors declare no conflicts of interest. AI Disclosure The authors declare that no artificial intelligence (AI) tools were used in the drafting, editing, analysis, or preparation of this manuscript. Transparency Statement The lead author Achit Kumar Singh affirms that this manuscript is an honest, accurate, and transparent account of the study being reported; that no important aspects of the study have been omitted; and that any discrepancies from the study as planned (and, if relevant, registered) have been explained. Supporting information Table S1: Binary Scoring of Knowledge, Attitude, and Practice (KAP) Items.Table S2: Classification of Knowledge, Attitude, and Practice (KAP) as Binary Outcomes. Acknowledgments We would like to acknowledge Farhat Ullah for his help in data collection. Written permission was obtained from the individual to be included in this section. Data Availability Statement The data that support the findings of this study are available from the corresponding author upon reasonable request. References Associated Data Supplementary Materials Table S1: Binary Scoring of Knowledge, Attitude, and Practice (KAP) Items.Table S2: Classification of Knowledge, Attitude, and Practice (KAP) as Binary Outcomes. Data Availability Statement The data that support the findings of this study are available from the corresponding author upon reasonable request.