Artificial intelligence (AI) plays an important role in transforming healthcare by enhancing diagnostic accuracy and enabling personalized and precision-based decisions. Ophthalmology, for example, has multiple AI applications in disease diagnosis, guiding treatment plans, surgical advice, and imaging interpretations, especially in retinal diseases such as diabetic retinopathy and age-related macular degeneration, where fundus photos can be read by AI-driven systems (1). In addition to computational power, it supports clinicians’ decision-making, boosts efficiency, and may reduce human errors (2).

However, integration of AI into ophthalmic practice raises many concerns around trust, accountability, and the evolving clinician role. Many previous surveys and reviews showed that eye care professionals are optimistic about these great advancements, but they have significant concerns about AI’s reliability, ethical implications, and risks of professional displacement (3–7).

In this study, we analyzed the perceptions of both medical students and practicing ophthalmologists (the current and future ophthalmologists) toward AI use in ophthalmology that may help in shaping ethical frameworks and educational strategies about the implementation of AI as a potential part of the standard of care in the field of ophthalmology.

The institutional review board (IRB) committee at King Hussein Cancer Center, Amman, Jordan, approved this cross-sectional study (25KHCC004).

The results of our study provide a comprehensive analysis of the perceptions of both medical students and ophthalmologists in a homogenous community towards the use of artificial intelligence (AI) in ophthalmology practice. Overall, our findings showed that AI is preferred as a complementary supportive tool that can help in diagnosis and recommendations for treatment, rather than a replacement for human expertise. This is expected because of the known promising AI capabilities and potential, but with a substantial limitations and ethical challenges (8–10). However, it is important to emphasize that our findings reflect perceptions and attitudes derived from hypothetical scenarios rather than observed clinical behavior.

Diagnostic reliability remains a major concern for practicing physicians when it comes to the integration of AI into clinical practice. Eighty percent of medical students and 72% of ophthalmologists still favor human clinical judgment over AI in diagnostic decision-making, which is consistent with prior literature emphasizing the importance of contextual and experimental knowledge that AI systems often lack (11). Diagnostic process in ophthalmology involves making correlations between clinical findings, patient history, and disease manifestations that current AI models may not fully capture, especially given their training on limited or homogeneous datasets. This underscores the necessity for clinicians to use AI cautiously, rather than deferring blindly. However, the perception that AI-assisted surgical techniques and recommendations could yield a better outcome was endorsed by more than half of participants in this study (55% of medical students and 56% of ophthalmologists), highlighting a growing confidence in AI’s ability to enhance procedural precision and consistency (12). Surgical applications of AI, such as robotic-assisted microsurgery and intraoperative guidance, have demonstrated promise in improving outcomes by minimizing human error, reducing fatigue-related variability, and enabling data-driven customization of surgical plans. This growing confidence in AI capabilities and in machine learning in general comes from the understanding that the AI mechanism of work depends on repetitive, high-precision tasks, and surgeons believe that these capabilities can help and enhance a surgeon’s performance and surgical outcomes. This perspective is consistent with findings from other fields, including neurosurgery and orthopedics (11, 13).

Surprisingly, we find that the trust in AI is variable according to the clinical scenarios involving a paradox between clinical human judgment and AI recommendations. When AI contradicted a clinician’s positive assessment of the expected outcome for refractive surgery, the majority (77%) deferred to AI recommendations, suggesting that participants regard AI as an important safeguard against overtreatment and potential procedural risks. This tendency was even stronger among medical students (79%) compared to ophthalmologists (72%), which indicates that less clinical experience leads to greater reliance on algorithmic caution than on human judgment. In contrast, when AI claimed that refractive surgery is safe and recommended in a situation where clinicians advised against this procedure, 91% of participants sided with the clinicians, reflecting enduring confidence in human expertise to make high-stakes, risk-averse decisions.

This asymmetry in trust distribution highlights a key psychological behavior: AI is predominantly trusted as a risk mitigation tool, especially in identifying when intervention may be unwarranted or harmful, but human clinical judgment retains primacy in approving high-risk procedures (13, 14). Such dynamics have been reported in radiology and oncology, where AI is welcomed for flagging suspicious findings or predicting adverse events but is less trusted to overrule experienced clinicians’ decisions to withhold treatment (11, 13). This phenomenon also highlights the importance of understanding AI not as an autonomous decision-maker but as a collaborator that enhances clinical safety. These interpretations should be considered cautiously, as psychological explanations for trust dynamics were not directly measured and remain speculative.

Ethical and practical concerns about AI’s implementation were prominent. The leading ethical concern in this study was focused on bias and ethics rather than healthcare disparities, as advocated by 69% of medical students and 76% of ophthalmologists. These concerns are well-founded, given that AI algorithms often derive from datasets that lack diversity, leading to potential misdiagnoses or inequitable care in underrepresented populations (15). The risk of systemic biases through AI could exacerbate existing health disparities unless human efforts are made to validate the algorithms to ensure fairness and transparency. Additionally, concerns about accountability because clinicians found it hard to trust AI models whose decision pathways are not clearly understood by humans, raising challenges in clinical validation and medico-legal liability (16, 17).

The vast majority of participants still believe that the responsibility for AI-driven clinical outcomes should be assigned to physicians rather than AI developers, and this aligns with current ethical and regulatory paradigms that place ultimate accountability on human clinicians. This finding emphasizes the importance of continuous clinical validation of all AI tools and for clear regulatory guidelines that define the boundaries of AI’s clinical role. Without such frameworks, there is a risk of legal ambiguity and erosion of trust among both practitioners and patients.

As expected, our data showed certain differences in perception toward AI within the ophthalmologist cohort. Those with fewer than 10 years of experience or who are ophthalmology residents in training were significantly more supportive of AI-assisted surgical interventions than ophthalmologists with more than 10 years of practicing experience. This trend likely reflects increased exposure to digital technologies during training and a greater openness to technological innovation in the young group (18, 19). It also suggests that early and sustained incorporation of AI literacy into medical education could foster more uniform acceptance and more effective human-AI collaboration in future clinical practice. More senior physicians are still trusting human skills more. This explains why about one-third of participants are worried that increased reliance on AI could impair the clinical judgment capabilities and worsen the surgical skills among future physicians (10, 19). The greater enthusiasm for AI-assisted surgery observed among residents and early-career ophthalmologists may reflect several factors beyond digital familiarity alone. These include differences in training environments, levels of surgical autonomy, perceived medico-legal responsibility, risk tolerance, and greater exposure to supervised or technology-augmented surgical workflows during training. In contrast, senior ophthalmologists may rely more heavily on experiential judgment developed over years of independent practice, contributing to more cautious adoption of AI-assisted interventions. Furthermore, more than half of the participants also believed that AI might negatively affect the doctor-patient relationship, possibly out of concern that it could make care feel less personal or reduce the sense of empathy. These findings underscore the importance of integrating AI in ways that uphold the human side of medicine, emphasizing communication, empathy, and ethical responsibility.

The support for incorporating AI training into ophthalmology curricula (69-71%) is encouraging, as it is a new tendency in medical education towards creating a structured, longitudinal education to improved algorithm literacy to ensure clinicians develop the skills necessary to critically appraise and effectively use AI tools (15, 19). Such educational initiatives will be crucial to avoid both over-reliance and undue skepticism and to cultivate the competence and confidence necessary for future ophthalmologists getting the maximum benefits of AI safely.

While our study provides valuable insights, it is important to consider its limitations. The use of hypothetical clinical scenarios, while methodologically necessary to control variables, may not capture the full complexity and time pressures inherent in real-world clinical decision-making. As respondents’ answers reflect idealized rather than practical behaviors, these findings should be viewed as a baseline for future qualitative study. Furthermore, the sample likely includes participants with more interest or familiarity with AI, potentially biasing results toward more favorable views. The study’s regional focus may limit generalizability, as attitudes toward AI can vary widely based on cultural, institutional, and healthcare system differences (20). Future research should explore these factors in a diverse larger population. Additionally, the regional nature of the study population may limit generalizability, as attitudes toward artificial intelligence can vary across healthcare systems, regulatory environments, and cultural contexts.

In summary, AI in ophthalmology is a promising, yet context-dependent, adjunct whose greatest value lies in complementing human clinical judgment rather than replacing it (21). The nuanced interplay of trust, risk perception, and ethical considerations revealed here should guide future AI deployment in clinical practice. We recommend integrating AI education throughout medical training, establishing clear ethical and regulatory frameworks, and designing specialty-specific models of human-AI collaboration that optimize patient outcomes without compromising the clinical role of physicians. This balanced approach will be essential as ophthalmology navigates the transformative potential of AI.

Artificial intelligence is a valuable tool in ophthalmology, with trust depending mainly on clinical context and perceived risk. AI is most trusted to avoid over-treatment, while clinician judgment remains decisive in high-stakes decisions. To maximize AI benefits, we need to integrate AI training in medical education, clarify accountability, and collaborative frameworks by making ethical guidelines that regulate human-AI interaction. Combining clinical sense and AI algorithms will be key to future ophthalmic practice.