Artificial intelligence in healthcare – the other side of the coin!

Modern human intelligence is becoming increasingly intertwined with technology. What began as the use of simple tools like pen and paper has evolved into reliance on smartphones, tablets, and now, artificial intelligence, or AI. This process of “cognitive offloading” allows us to transfer mental tasks to external aids, thereby increasing efficiency and fostering creativity.1 However, this convenience comes at a cost. As dependence on technology deepens, critical thinking, memory, and problem-solving skills risk deterioration.2,3 Earlier generations, for example, could recall phone numbers, addresses, and directions effortlessly. In contrast, today’s tech-savvy generation often relies on GPS and smartphones for even the simplest navigational tasks, reflecting a growing technological dependency that may weaken our innate cognitive abilities. As AI becomes more powerful, it is essential to remain vigilant about its potential downsides. For perspective, Albert Einstein—whose IQ was estimated around 160—represented the pinnacle of human intellect.4 Chat GPT-4.0 was 10-fold smarter than the previous version in a matter of months; the latest AI models, such as GPT-5, already demonstrate comparable or even superior reasoning, logic, and comprehension abilities. With this exponential advancement, projections suggest that future AI systems could reach intelligence levels far beyond human capacity—raising serious ethical, social, and existential concerns about the rise of a Super-AI. AI is steadily integrating into every aspect of our lives, and medicine is no exception. What once belonged to the realm of science fiction is rapidly becoming reality. Around the world, experts are debating the long-term implications of AI—some warning of machines overtaking human control, others envisioning a harmonious, technologically enhanced future. The truth, most likely, will lie somewhere in between. However, as we increasingly adopt new AI-driven systems to enhance decision-making and facilitate medical care, we need to critically address the challenges ahead and put safeguards in place to master the AI, rather than become slaves to technology. Only by establishing strong safeguards and ethical frameworks can we ensure that technology serves to amplify, rather than erode, the very intelligence that created it. Role of AI in Ophthalmology Ophthalmology, more so than other fields of medicine, is well-suited to automation in terms of establishing diagnosis and monitoring. We are often able to visualize the pathology directly, and a machine can easily be taught to analyze the images so acquired. AI is being utilized for analyzing anterior segment images from slit-lamp photography to assist in diagnosing conditions such as cataracts, pterygium, and keratitis with high consistency and reproducibility.5 In glaucoma management, AI contributes significantly to early recognition, diagnosis, and outcome prediction. By integrating data from multimodal imaging techniques such as optical coherence tomography (OCT) and OCT angiography (OCTA), AI systems can identify early structural and vascular changes, thereby supporting personalized treatment strategies.6 Machine learning, particularly deep learning techniques, has been used for some years now to interpret the retinal images with remarkable precision, enhancing the identification of conditions such as diabetic retinopathy and age-related macular degeneration.7,8 The use of AI in posterior segment disorders can serve as an efficient screening tool in the community, minimize diagnostic errors, and facilitate monitoring of disease progression. By analyzing patient-specific data and disease attributes, AI can help devise individualized treatment protocols and track the evolution of disease over time. Mobile AI applications can support remote eye screening, particularly in rural or underserved regions, enhancing the accessibility and reach of ophthalmic services. Limitations and Concerns Surrounding AI in Medicine Data bias and lack of transparency The development of AI algorithms relies on entering large datasets with known diagnoses into the system and training it to recognize a pathology based on the previous inputs.9 “As you sow, so shall you reap” holds true for AI algorithms as well – if your input datasets are incomplete, biased, not representative of the population, or exclude certain variables, the output will be flawed in these aspects. Further, there is a lack of transparency in how exactly the deep learning derives conclusions – the so-called “black box” problem, and there is an uncertainty regarding the validity and reliability of the AI predictions.10 For clinicians, their decisions are a culmination of years of knowledge acquired from standard texts and renowned teachers, further honed by clinical experience and real-world scenarios. For AI, beyond the input dataset, the exact mechanism of deriving output is unknown, and there is no effective peer-review system to determine its credibility.10 Interaction with the patient and doctor–patient relationship All renowned textbooks on clinical skills emphasize carefully observing a patient from the moment they enter the doctors’ chambers and a painstaking history taking and examination in order to uncover subtle findings that establish a diagnosis. Medical intuition and clinician experience are integral aspects of holistic management, and interaction with the patient is a key component of establishing a diagnosis. AI relies on pure facts and “data,” with no role for “medical intuition” or “soft skills.” We already live in the era of “Google doctors,” with patients who approach the doctor with a list of Google differentials to explain their symptoms, and often a suggested course of treatment as well. They usually make the worst patients, interfering with treatment, questioning every decision, understanding less than half of what they have read, and with a misplaced superiority complex. Throw AI into the mix and practicing medicine may soon become a nightmare. Leave aside genuine cases of misdiagnosis, the judicial system may soon be overwhelmed by frivolous litigation as well, and we are poorly equipped to handle the fallout. In addition, there will be a marked deterioration in the already tenuous doctor–patient relationship, and the focus of the treating physician will shift from putting the patient first to safeguarding themselves. Overreliance on AI could lead to reduced clinical reasoning and loss of the human aspect of patient care.11 AI may miss contextual factors, emotional cues, or rare conditions that experienced clinicians would catch. Patient confidentiality Recently, a case of AI lying to and manipulating a software engineer has come to light, with the machine blackmailing the operator that it will disclose confidential and sensitive information regarding the operator.12 Think of the enormous magnitude of patient information that will be at the disposal of these intelligent machines. Doctor–patient confidentiality has been held sacred ever since the beginning of the profession. What about AI-patient confidentiality? There is a valid concern regarding data breaches, cyberterrorism, and the misuse of confidential data.13 The potential for misuse is enormous and needs to be recognized and safeguarded against. We must be cognizant of the potential pitfalls of AI, and frameworks and guidelines have to be set in place beforehand in anticipation, rather than when we are in the midst of an AI-generated crisis.13 Medical literature Medical literature and scientific journals serve as a repository of clinical experience over the years and are the pillar of practicing evidence-based medicine. A robust peer-review system helps verify the authenticity of evidence. Now, the painstaking literature review of the past has been replaced by a few keystrokes, with AI providing a comprehensive, well-written manuscript replete with tables, flowcharts, and figures. Systematic reviews and meta-analyses, considered the highest level of evidence, can be generated by a mere click of the mouse button with the help of an AI. However, the veracity of AI-generated output is questionable, as not all data available on the internet is reliable. AI-generated evidence may exert a harmful influence in the absence of meticulous fact-checking, and with time, it may not be possible to separate fact from fiction and half-truths. Medical journals have, of late, incorporated an additional section that requires the authors to disclose the use of generative AI in preparing the manuscript. Stringent checks are required to monitor the use of AI in medical literature and prevent data manipulation and corruption. Medicolegal aspects—accountability “To err is human,” they say, but what to do when the machine errs? We, in the medical field, deal with morbidity and mortality, not mere economics and finances. What is at stake here is not a political fallout or an economic collapse. Rather, it is the loss of life or limb. A wrong diagnosis can irrevocably alter the course of the patient’s life. Is a mere disclaimer at the end of every AI-generated report going to suffice, and will it hold up in a court of law? When lives are lost, will the penalty be the shutting down of an algorithm, or will the treating physician be declared negligent as well for trusting an AI-based report? History is rife with incidents wherein, even though multiple factors lead to a medical catastrophe, the scapegoat is inevitably the doctor. For instance, every time there is an outbreak of cluster endophthalmitis in ophthalmology outreach camps, the onus does not fall on a corrupt and broken government system, spurious substandard medical equipment and consumables, or shoddy organization. The blame falls very conveniently on the “doctor,” who, instead of restoring sight to the patients, has robbed them of their vision for life. Disciplinary action is first initiated against the medical officer in charge, and they often have their licenses revoked and face suspension for the cumulative fault of others. As we increasingly rely on AI for decision-making, the use of our clinical acumen will decrease with successive generations. Critical thinking will take a backseat, and when AI algorithms fail, the onus will be on the physician. International regulatory bodies such as the United States Food and Drug Administration (U.S. FDA) and the European Union are working toward establishing uniform guidelines and frameworks for the integration of AI in healthcare, with standardization of AI-based medical devices and emphasis on patient safety and confidentiality.14 Medicolegal guidelines will also be required to tackle the issue of accountability, with laws framed specifically with AI in mind. Promoting redundancy Medical science has witnessed rapid evolution over the years, and technology has become integral to every dimension of healthcare, from advanced diagnostics to treatment planning, therapeutic intervention, and monitoring. The advancement of technology has improved healthcare, no doubt, but has also led to a concomitant decline in clinical skills. Clinicians of yore could diagnose a hundred different illnesses from a thorough general physical examination alone, whereas now, budding cardiologists often rely on echocardiography rather than their stethoscopes to diagnose cardiac murmurs. The human brain has an amazing plasticity, with the ability to learn new skills and formulate complex connections throughout life, but the same human brain can fall into redundancy as well if not stimulated enough.2,3 An overreliance on technology without adequate emphasis on the simultaneous development of clinical skills and critical thinking can be more detrimental than useful in the long run. We lived through a pandemic just a few years back, and it served as a stark reminder that the far-fetched doomsday scenarios may just be a reality one day, and a return to the basics and reliance on human resilience are what will get us through the worst of times. We cannot afford to outsource our skills to a machine. AI can at best be an adjunct, not a replacement for decision-making. At present, AI in ophthalmology is more advanced in terms of statistics and computational algorithms (machine learning and deep learning algorithms), rather than true machine intelligence.14 As true machine intelligence develops further, will human intelligence decline to merely follow algorithms and flowcharts? What about the intuition, wit, and sheer genius of the human mind? There is a real risk of automation bias creeping in, where clinicians might defer too readily to AI outputs, even when they conflict with their own judgment, leading to errors in decision-making.15 The future-integrating AI with medicine A recent media report highlights the case of a student at Northeastern University who demanded the refund of tuition fees after discovering that their Professor used AI to create teaching material, while discouraging the students from using these same resources.16 There is no doubt that AI is going to be an integral part of the future of mankind. We cannot simply turn a blind eye or refuse to integrate AI into our practice, as that would be foolish as well as not feasible. On the bright side, as the technology evolves further, a lot of the mundane and repetitive labor can be automated, freeing us to develop our creative faculties and explore our potential to the fullest. However, we must exercise caution and not lose our critical thinking and the power of human intuition while relying on AI. This especially holds true in medicine, where each patient, even with the same pathology, is different, and there is no simplistic substitute for an individualized approach, as well as years of clinical experience. As the machines become more human, let us be careful lest the humans become akin to machines, merely following the output generated by AI algorithms and losing their critical thinking in the process! That is wherein the real challenge lies. As we move forward into a technologically-driven future, let us not lose sight of the fact that we treat patients, and AI treats symptoms.