Science fiction authors and futurologists often imagine a world where ubiquitous use of artificial intelligence (AI) in medicine means the end of physicians. In such a world, omniscient AI bots effortlessly diagnose and treat diseases, while various robots are busy performing life-saving surgeries. Many physicians, on the other hand, hold the view that medicine is as much an art, as it is science, which makes it unique among other industries. They believe that efforts to incorporate artificial intelligence into healthcare are a fool’s errand. The truth probably lies somewhere in the middle, at least in the present and in the near future.
The current landscape of medicine reveals shortage of physicians and limited access to healthcare in many parts of the world. The physician density is 3.1 per 1000 people in urban USA as compared to 1.3 in rural USA, 0.767 in South Africa and 0.725 in India. World Health Organization sets the norm as one physician for every 1000 people. The desirable physician density is not the only metric, because even urban USA, which has high physician density, sees problems with access to care due to costs, geographic variability and wait times.
A low-tech solution in many countries has been creation of licensed healthcare professionals who are not physicians, such as physician assistants, nurse practitioners and midwives. Most of these professionals are supervised by a physician, depending on local laws, and are often the first or the only point of access to healthcare for patients. These practitioners play in important role in healthcare, especially in the areas with limited access, but their limited training does not provide them with the same breadth of knowledge as physicians.
Artificial Intelligence is perfectly positioned to help address this issue in the immediate to near future. Consider the following examples:
- Point-of-care evaluation: Large number of people in different parts of the world resides prohibitively far away from a medical laboratory or a facility equipped with diagnostic imaging tools. This may lead to delayed diagnosis and increased morbidity and mortality. Artificial intelligence can help in a variety of such scenarios, where a non-physician provider equipped with a smartphone application can be the first step in a diagnostic and screening process of a variety of conditions, and later refer appropriate patients to a physician or request help remotely from a supervising physician. For example, an AI tool and a simple microscope with DIN achromatic lens can be used to diagnose fungal skin infections. The same setup applied to blood analysis can be used to diagnose anemia and parasitic infections like malaria. A fundus camera used in conjunction with a smartphone and an AI algorithm can be used to screen for vision conditions such as diabetic retinopathy.
- Patient monitoring: In any given hospital there is a plethora of patient monitoring devices used for various purposes, such as telemetry for monitoring for aberrant heart rhythms, bed alarms to detect falls, numerous variables tracked during surgery. All of these are designed to increase patient safety, but their alerts and alarms currently have to be interpreted by a human. This frequently results in alarm fatigue, which has recently been getting a lot more attention as a real danger. Using AI to analyze alarms and alerts and to filter out noise can greatly improve patient safety, increase staff productivity and lower costs, by allowing care providers to focus on more important tasks. Similarly, there is now a large number of devices that are increasingly being used by patients themselves to monitor health and wellness. As more information is being gathered about interpretation of these data, AI can be used to learn to analyze this information for appropriate action.
- Medications analysis. All medications come with various side effects and interactions, but unfortunately current databases for cross checking those are cumbersome and still require that medication names be entered manually. An AI tool can be easily trained to automate this task to improve patient safety by increasing awareness of possible interactions.
Applied in appropriate situations, artificial intelligence can help address issues of limited access to care, as well as improve efficiency of care providers by eliminating or reducing menial tasks. By making non-physician medical professionals more productive, it reduces the risk of errors. The healthcare industry can then utilize increased numbers of   physician assistants, nurse practitioners and midwives with higher confidence in more under-served areas. This can result in reduced patient wait times, lower cost of care and increased patient satisfaction.
One area of medicine, where physicians are indispensable is clinical decision making. It is used to arrive at a diagnosis and formulate a treatment plan. There are two factors that limit AI’s applicability in this area. The current AI models lack sophistication due to limited patient data. As more data become available to train the AI models and more deep learning techniques are used, AI sophistication will improve. The regulatory framework is another important and more limiting factor.  The existing laws and regulations in most countries are lagging far behind developments in the science of artificial intelligence, which makes it very difficult to regulate artificial intelligence products and services and thus restricts innovation.
Probably in the next 100 years or more, as some of these issues are sorted out and the science of artificial intelligence continues to develop, AI has the potential to become as disruptive in healthcare as it has been in other industries. Could it replace physicians in this timeframe? Probably yes. Instead of fearing it, however, we can learn to use it to our advantage. If humanity were scared of progress, we would still be riding horse buggies and lighting our homes with candles.
