Point-Counterpoint: An Overview of Artificial Intelligence in Eye Care

INTRODUCTION

Artificial Intelligence (AI) is a branch of computer science focused on developing programs that mimic human thinking.¹ It operates by analyzing large volumes of data and identifying patterns that support more informed decision-making.¹ It is primarily designed for problem-solving with datasets, pattern identification, and performance improvement without direct human intervention.^1,2 AI is not a single, all-encompassing technology; instead, it consists of multiple subfields that embed intelligence into applications applied across numerous sectors, including business, manufacturing, and health care, offering the potential to transform communication, streamline problem-solving, and enhance human interactions.^2,3

History

The concept of AI dates to ancient mythology and 18th-century probability frameworks, which envisioned information-collecting “idea generators” that could assist in problem-solving.² Although purely aspirational at the time, the goal of creating a nonhuman intelligence using empirical data was achieved with the invention of the computer. As computers advanced, they began collecting and processing information far faster than humans, enabling the identification of intelligible patterns and trends in data, laying the foundation for AI.² Further advances in computing led to the development of transformer architectures, enabling generative AI models to learn underlying patterns in data and produce videos, music, images, and text.³ These probabilistic models represented the first generation of AI.⁴ In health care, this early form of AI is exemplified by tools such as clinical decision trees.⁴

Second-generation AI emphasizes machine learning, particularly deep learning, in which convolutional neural network–based algorithms enable computers to interpret, recognize, and classify images.⁴ This type of AI is task-specific, meaning it is designed to perform only one task at a time.⁴ Deep learning models are trained using labeled datasets that serve as ground truth—for example, identifying an image as “a cat”.⁴ Instead of relying solely on predefined rules, these models independently learn and detect patterns within the data.⁴ A notable example is a deep learning system capable of detecting diabetic retinopathy with accuracy comparable to that of trained expert eye care physicians.⁵ The ongoing AI boom has since impacted nearly every aspect of society.³

Third-generation AI, also called generative AI, differs fundamentally from earlier forms in that it can perform a wide range of tasks without retraining and is powered by large language models.⁴ Rather than truly “thinking,” generative AI predicts the most likely next word based on patterns learned from its training data. This form of AI increasingly shapes everyday life through tools such as writing assistants, image-generation platforms, and chatbots that understand conversational context.⁴

Healthcare Applications

In a broad sense, AI enhances health care by enabling predictive analytics and creating algorithms that support data-driven decision-making.³ It aids in differential diagnosis and treatment planning, forecasts morbidity and mortality trends, and promotes more efficient and effective patient care.³ AI is also highly valuable in research, where it accelerates the development of best-practice therapies, supports drug discovery, and facilitates the development of new curative treatments.³ In precision diagnostics, AI has proven particularly effective in diagnostic imaging.³ The “quadruple aim” framework highlights AI’s role in patient care, focusing on improving population health, enhancing the patient experience, increasing health care provider effectiveness, and reducing costs.^3,6,7 AI applications facilitate connectivity among providers across various health care settings. For patients, AI has been beneficial through tools such as chatbot-integrated wearable devices, including smartwatches, that monitor vital signs in real time. Moreover, AI’s role as a virtual assistant holds significant potential to strengthen and optimize the patient–provider relationship.^3,6,7

This paper explores the advantages and disadvantages of AI by reviewing current applications in health care. Among its benefits, AI enables rapid and efficient information collection, supports informed decision-making, and helps reduce human bias during data gathering and analysis. It also automates repetitive, labor-intensive, and time-consuming tasks, freeing health care professionals to focus on more complex and creative responsibilities.¹

Nevertheless, several disadvantages exist. Implementing AI can be costly, especially when systems require specialty-specific customization. Automation may result in workforce reductions by replacing human tasks, and AI lacks the nuanced, creative thinking inherent to humans. Ethical and security concerns are also significant considerations in its use.⁸

Although AI remains somewhat of a “black box” to the public and raises ethical concerns, with the potential to disrupt professional environments, its integration into optometry practice is inevitable. It offers a valuable tool for enhancing problem-solving and supporting clinical decision-making and is expected to play an increasing role in routine care as technologies continue to advance.

POINT

Leslie Wilderson, OD, FAAO, Dipl AAO

Eye Care Applications

AI can augment clinicians’ expertise by detecting pathology in ocular structures through imaging.^9,10 Many highly anticipated tools—such as Altris AI, RetinAI Discovery, Notal Optical Coherence Tomography Analyzer, and ZEISS CIRRUS PathFinder—support clinical decision-making by detecting abnormalities in retinal images. These systems enable clinicians to focus more on diagnosis and treatment planning while reducing the time required for image interpretation.^9–11 Several sources cite studies showing that deep learning algorithms perform comparably to expert eye physicians in detecting ocular disorders from fundus photographs and optical coherence tomography scans.^5,10,12 These systems can identify subtle structural changes, prioritize high-risk cases through automated triage, and support earlier intervention.^9,10 Furthermore, AI-assisted analysis of ocular disease has shown potential to improve diagnostic consistency and accessibility of eye care, particularly in underserved or resource-limited settings where specialist availability is challenged.^5,9,12 In routine clinical settings, AI has been shown to improve workflow and reduce patient wait times.¹³

Landmark Studies

The core landmark literature in eye AI includes works by Grzybowski et al,¹⁴ Gulshan et al,⁵ Ting et al,¹⁵ De Fauw et al,¹⁰ and Abràmoff et al,¹⁶ among others. Collectively, these studies showed that deep learning-based algorithms can achieve expert-level performance comparable to human graders and, in many cases, outperform traditional image-analysis.^5,10,14–16 The work by Ting et al was supported by a large multicenter dataset and demonstrated generalizability across diverse populations and multiple retinal diseases.¹⁵ Grzybowski et al conducted a comprehensive 2024 systematic review that highlights the “performance, scope, and breadth” of AI algorithms across multiple retinal conditions and imaging tasks.¹⁴ Subsequently, the pivotal trial by Abràmoff et al reported the first FDA-cleared autonomous AI diagnostic system, IDx-DR (Digital Diagnostics), representing a significant regulatory milestone in clinical AI.¹⁶ The Gulshan et al landmark study used the Eye Picture Archiving and Communication System (EyePACS) dataset, one of the most widely used public datasets for training and evaluating AI models for diabetic retinopathy detection.⁵ The EyePACS dataset was designed to be broadly representative, including retinal images from individuals who self-identified as African, European, Asian, from the Indian subcontinent, and Indigenous American.¹⁷ This diversity is a key strength of the dataset and has contributed to its widespread adoption by researchers who develop and validate AI algorithms for retinal disease detection.¹⁷ The Kaggle EyePACS model demonstrates that deep learning systems have a transformative impact on eye care when intentionally debiased, reducing disparities.¹⁷ The EyePACS dataset was a turning point in medical AI, showing that deep learning could match human experts in image-based diagnosis.⁵

Analytical Thinking

When used as a decision support tool rather than a replacement for clinician judgment, AI can reinforce analytical thinking in diagnosing ocular disease.¹⁸ By synthesizing large datasets and identifying clinically relevant patterns, AI exposes clinicians to emerging trends that may prompt reassessment of diagnostic assumptions rooted in outdated information.¹⁹ When regarded as opportunities to elevate clinical practice, AI-supported systems encourage reflective reasoning through comparative analyses and evidence-based recommendations that require interpretation within a clinical context,²⁰ ultimately enhancing critical thinking rather than diminishing it.¹⁸

Patient–Doctor Relationship

Recent survey data indicate that the adoption of AI tools in clinical practice increased substantially in 2024, with approximately two-thirds of physicians reporting use of health care AI, up from less than 40% the previous year.^21,22 The integration of AI has facilitated more efficient care coordination among clinical teams, resulting in improved patient experiences.²³

Moreover, as diagnostic accuracy improves, AI has the potential to positively impact the patient-doctor experience. By automating time-intensive tasks such as image interpretation, data analysis, and clinical triage, AI enables physicians to devote more time to direct patient engagement, education, and shared decision-making regarding treatment options. Furthermore, AI-driven systems can reduce diagnostic variability among clinicians, fostering greater confidence in diagnoses and reinforcing patient trust.²³

In screening environments, AI-enabled tools have demonstrated expert-level diagnostic performance while expanding access to care, particularly in underserved populations where specialist availability is limited.¹² Earlier detection and referral, facilitated by AI-supported screening, may lead to more timely interventions, improved outcomes, and increased patient satisfaction and trust when guided by best-practice frameworks.^18,22

In clinical practice, transparently implemented AI systems can enhance clinician expertise and improve the quality of care while enabling a greater focus on empathetic, patient-centered interactions.²³ Although questions remain regarding optimal integration, evidence suggests that responsible implementation of AI can strengthen the patient-doctor relationship.

Accountability and Beneficence

Responsibility for the appropriate use of AI-generated diagnostic data ultimately rests with the licensed health care professional, as current legal and regulatory frameworks do not recognize AI as an autonomous decision-maker.^24,25 Professional judgment is essential to balance AI-enabled diagnostic innovations with treatment decisions grounded in established clinical protocols and evidence-based standards. In eye care, the principle of beneficence guides clinicians to integrate AI in ways that enhance diagnostic accuracy, enable early disease detection, and ensure timely intervention in the patient’s best interest. When responsibly implemented, AI-driven tools can reduce diagnostic error, standardize care, and expand access to eye care services, thereby promoting patient welfare while complementing clinical expertise rather than replacing it.^12,18,26

Conclusion

The pro-AI perspective highlights humanity’s drive to expand knowledge, innovate, and overcome limitations to advance health care. In optometry, preparations are underway for AI to play an increasingly significant role in routine care as technologies evolve. Reflecting on the growing presence of AI in optometry at the VA, Dr. Doug Rett, Chief of the Boston VA Optometry Service, stated, “Anyone who thinks their skills are superior to technology is soon humbled.” (Doug Rett, OD, FAAO, personal communication, December 7, 2025). While humans naturally seek control, we are equally driven by curiosity, exploration, and the pursuit of progress.

COUNTERPOINT

Danielle Weiler, OD, FAAO, Dipl AAO

Overreliance and Erosion of Critical Thinking

With the rapid expansion of AI applications in eye care, is it inevitable that reliance on AI will diminish clinicians’ critical thinking abilities? Critical thinking, defined as the ability to analyze, evaluate, and synthesize information to guide decision-making, is vital for managing complex patient care.²⁷ Excessive use of AI tools can promote cognitive offloading, whereby mental tasks are transferred to external aids. While this can free cognitive resources, overreliance may reduce reflective engagement and foster “cognitive laziness”.²⁷ A related phenomenon, the Google effect, refers to reliance on external sources for information rather than internalizing it.²⁸ Individuals with broader knowledge bases and higher education levels are less affected, but studies show that reliance on AI can still weaken memory and independent problem-solving.^27–29

Patient–Doctor Relationship

Despite its advantages, AI cannot replace human expertise. Optometric excellence relies on thorough ocular examinations, a majority of which require in-person evaluation.³⁰ Moreover, Beede et al found that although the AI system for diabetic retinopathy performed well in laboratory tests, it caused workflow issues, patient anxiety, and staff confusion in real-world clinics—underscoring the gap between controlled validation and everyday clinical practice.¹³ AI algorithms may not capture the nuanced clinical judgment gained through years of patient interactions and deductive reasoning.³⁰ Models for clinical use must align with existing imaging modalities and incorporate clinician-derived conclusions.³¹ Whether AI can reliably detect and differentiate pathologies independently remains uncertain, making human expertise indispensable.

AI cannot replicate human experience.^32,33 Human connection and empathy are vital components of comprehensive, patient-centered care and are deeply human skills that AI cannot emulate. Optometrists spend years cultivating the ability to build rapport with patients, allowing them to understand, share, and respond appropriately to patients’ emotions and experiences while engaging in shared decision-making.^23,34 Although AI may enhance diagnostic accuracy, it cannot provide emotional support that must remain central when communicating diagnoses to patients.²³ Additionally, patients may lack sufficient understanding of their conditions and may require guidance from an expert clinician.²³

Accountability, Non-Maleficence, and Equity

Clear lines of accountability are foundational when AI is used in health care decision-making.³⁴ For instance, if AI contributes to a misdiagnosis, responsibility may fall on the clinician, the developer, or both.^32,34 Current regulatory frameworks have not fully adapted to address AI-related medical errors.³¹ Clinicians must receive proper training in AI tools and remain accountable for accurately interpreting their outputs.³³ The consensus is that AI should serve as an adjunct, not a replacement, for clinical decision-making.^11,30,34 Ultimately, clinicians retain responsibility for final decisions.³³

From an ethical perspective, non-maleficence requires avoiding harm.³⁴ While AI can improve diagnostic and therapeutic outcomes, it may exacerbate health disparities if models are trained on small or homogeneous datasets.^{11,12,33–37} Furthermore, proprietary algorithms and limited data sharing favor organizations with greater resources, potentially leaving disadvantaged populations underserved.^33,37 A landmark study by Burlina et al investigated algorithmic bias arising from the underrepresentation of ethnic groups in training datasets.¹⁷ It showed clear performance disparities across skin pigmentation groups, with AI accuracy at 73% for lighter-skin individuals and 60.5% for darker-skin individuals.¹⁷ It was inferred that such bias may occur because variations in skin pigmentation, related to melanin concentration in uveal melanocytes, affect retinal coloration.¹⁷ The study modified the original Kaggle EyePACS database, which is neither imbalanced nor biased.¹⁷ By augmenting it with clinician-annotated labels and introducing an artificial scenario to simulate data imbalance and domain generalization, bias in the diagnostic datasets for diabetic retinopathy classification adversely affected AI performance in darker-pigment eyes.¹⁷

Finally, the “black box” nature of AI—stemming from complex deep learning algorithms—limits transparency and interpretability.^11,12,34,37 Developing systems that provide clear, clinician-understandable explanations ensures that AI supports decision-making in the patient’s best interest.³⁴

Conclusion

Current evidence indicates that AI, when used as a clinical decision support tool, can enhance multiple aspects of optometry practice. Studies show that AI systems improve diagnostic accuracy and efficiency while expanding access to eye care in certain clinical contexts. However, safe and effective implementation requires ongoing clinician oversight, careful evaluation of training data to ensure equitable representation, and continual ethical and regulatory review. The literature consistently suggests that AI in optometry augments clinician expertise rather than replacing human judgment.

Conflicts of Interest

None

Financial Disclosures

None