The term Artificial Intelligence (AI) evokes both excitement and anxiety. But what does it mean for healthcare? The hype surrounding AI in healthcare seems like a breakthrough on par with the discovery of antibiotics or vaccines. Yet, the reality may not live up to these expectations — at least not in the near future.
Healthcare has evolved over centuries, contributing to our understanding of health and disease. Today, with the rise of AI, we explore how this technology might shape the future of healthcare. To do this, it is crucial to understand some basic concepts of health and where AI might fit into this larger narrative.
The biomedical concept of health, based on the germ theory of disease, transformed medicine by linking microorganisms to diseases and led to breakthroughs such as vaccines and antibiotics. However, it has limitations: the most significant being its failure to address non-infectious issues such as malnutrition, accidents, and mental illnesses. The ecological concept broadens the health perspective by considering environmental factors, such as food availability and air quality, emphasising the connection between health and the environment. The psychosocial concept includes social and psychological factors, highlighting how societal culture, beliefs, and socioeconomic conditions influence societal issues such as gun violence, addiction, and mental health. The holistic concept integrates all models – physical, mental, emotional and social dimensions for a comprehensive approach to healthcare.
Breakthroughs over the course of history
Over the past 500 years, several breakthroughs have revolutionised how we understand and approach healthcare. James Lind’s historic experiment showed that a citrus-rich diet could prevent scurvy, though the role of Vitamin C was not yet known. This marked evidence-based medicine. The discovery of vaccines by Edward Jenner introduced the concept of disease prevention. Anaesthetics made surgeries painless and made complex procedures possible.
The discovery of antibiotics radically reduced infection mortality, and the microscope opened up a new world of microorganisms, reshaping our understanding of diseases. The birth of epidemiology during the cholera outbreak — before the identification of Vibrio cholerae — demonstrated the power of public health interventions.
A common thread ties these breakthroughs: they did not just improve health services but also fundamentally changed how diseases were viewed, shifting from the reactive to the preventive, curative and scientific approaches. Now, as we face the growing impact of AI, can it, too, transform healthcare, reshaping not just what we do but how we understand and address health challenges?
Where AI fits in
AI, much like healthcare, is a broad concept. To understand AI’s role in healthcare, it is crucial to clarify that AI is not a single, well-defined tool. It is a collective term that encompasses various technologies. These include large language models, diffusion models, deep learning, machine learning, natural language processing and robotics, each offering different tools that could potentially be applied in healthcare.
Generative AI has shown promise in specific applications such as generating medical images, assisting drug design and aiding documentation. Still, its impact remains limited to creative tasks rather than revolutionising treatment or diagnosis. Predictive AI has attracted more attention by aiming to forecast health outcomes through pattern recognition in large datasets. However, its promise is yet to fully materialise, as predictive models require vast training data and can be biased.
When we talk about AI in healthcare, much of the excitement revolves around predictive AI models that are supposed to predict disease outcomes, optimise treatment plans, and even forecast patient deterioration. However, predictive AI is not as advanced or reliable as it is often portrayed to be. Take, for example, the Google Flu Trends experiment, an early application of predictive AI in healthcare. By analysing search terms, Google tried to predict flu outbreaks faster than traditional reporting mechanisms. It failed. Similarly, sepsis prediction models have not performed as anticipated.
The problem of truth erosion
A major concern with AI in healthcare is ‘truth erosion’, where reliance on AI risks diminishing critical thinking in medicine. This is worsened by commercial interests, with proprietary algorithms limiting independent scrutiny and contributing to a reproducibility crisis. AI models often work well in controlled settings but falter in real-world applications. Additionally, many models lack transparency, creating a ‘black box’ issue where predictions are made without clear reasoning. In medicine, explainability is crucial. If doctors can’t understand or trust AI-generated predictions, this undermines confidence in the technology.
Healthcare, by nature, is considered a public good. However, much of the AI development in healthcare is expensive and driven by private companies. In India, the healthcare system is already under considerable strain, and implementing costly AI tools might not be the most efficient use of resources. Investing in AI without first addressing fundamental health system gaps could be an expensive, pointless exercise.
The limitations of AI
AI in healthcare faces several limitations, particularly in predicting human behaviour. Health outcomes are heavily influenced by individual actions, such as medication adherence or lifestyle changes, which are difficult to forecast. Like the unpredictability of financial markets or weather patterns, human behaviour complicates AI’s ability to predict health outcomes accurately. Diseases are complex – for instance, a common condition like diarrhoea has multiple causes — viruses, bacteria, or parasites, all of which require different treatments. Conditions such as diabetes and cancers arise from a multifactorial combination of genetic, environmental, and lifestyle factors, making them harder to predict and treat.
In the future, AI may be able to better predict single-gene disorders – for instance, sickle cell anaemia – with advances in genomics, but most diseases are dynamic and involve an evolving interplay of biological, behavioural, and environmental elements. Predictive AI, which depends on large datasets to recognise patterns, struggles due to healthcare data’s incomplete, biased, and scattered nature. As diseases evolve and human behaviours shift, AI models risk becoming obsolete unless regularly updated with new information.
To effectively integrate AI into healthcare, it is crucial to close this gap. AI developers need a deeper understanding of the healthcare system’s complexity, while healthcare providers must be educated about how AI works, its limitations, and how best to use it. Currently, these two fields often function on parallel, poorly connected islands.
The key to AI’s success
AI in healthcare holds genuine promise for the future but has been overhyped in certain aspects, particularly in predictive healthcare. Unlike the crypto boom, AI is not a technology without substance. AI can help improve healthcare delivery in the future but will not revolutionise it. Its greatest strength may lie in augmenting human decision-making and improving efficiency. The key to AI’s success in healthcare will be a balanced, informed approach that involves better education for healthcare providers, more interdisciplinary collaboration between tech developers and clinicians, and creating policies that ensure AI serves the public good.
(Dr. C.Aravinda is an academic and public health physician. He wishes to acknowledge the book ‘AI Snake Oil: What Artificial Intelligence Can Do, What it Can’t and How to Tell the Difference’ for contributing to his understanding of the subject.)
Published – February 25, 2025 10:30 am IST