There are few fields where expert knowledge has such a profound effect on human life as medicine. Doctors train for years to gain the skills required to diagnose diseases and recommend the appropriate treatment. While medical knowledge has developed by leaps and bounds in the past century, the lack of trained medical professionals relative to their patients means that the delivery of potentially lifesaving cures is often delayed.
It is therefore no surprise that experts are turning to artificial intelligence (AI) to increase doctor productivity and thereby improve the efficiency of healthcare. AI builds upon traditional computer automation by introducing algorithms that can learn from data (i.e. machine learning) instead of requiring a programmer to define how a machine should perform a particular action. With recent advances in machine learning, computers are quickly achieving the same level of competence as human doctors in specific tasks, sometimes even exceeding them.
To give a specific example, the Stanford Machine Learning Group has developed an algorithm called CheXNet that can scan chest X-rays and diagnose pneumonia with accuracy exceeding those of practicing radiologists. This has the potential to save countless lives as chest X-rays are a critical tool for screening, diagnosis, and management of pneumonia, a disease which is responsible for more than 1 million hospitalizations and 50,000 deaths per year in the United States alone. It is estimated that it takes at least thirteen years of education and training after high school to become a radiologist with the skills to accurately interpret chest X-rays, leading to a shortage of experts, with an estimated two thirds of the global population lacking access to radiology diagnostics.
With the spread of wearable technology, diagnostic tools are no longer limited to hospitals but can now be found in the health sensors of smartwatches. The Apple Watch uses a combination of optical and electrocardiogram sensors built into the watch’s crown and back to record its wearer’s heart rate and predict atrial fibrillation, a leading cause of stroke, with 97% accuracy.
Wearable devices also extend diagnostic methods to patient monitoring, with patient health data collected 24/7 instead of only during hospital visits, allowing doctors to work with accurate symptom data instead of second-hand information from patient and caregiver interviews. This improvement in both data quality and data quantity makes it more necessary than before for doctors to leverage AI systems to sift through the data for relevant information.
Beyond diagnosis and monitoring, AI can even be used to automate drug discovery. In 2012, Merck sponsored a drug discovery competition where participants were given a dataset describing the chemical structure of thousands of molecules and were asked to predict which were most likely to lead to new drugs. Remarkably, the winning team had only decided to enter the competition at the last minute with no specific knowledge of biochemistry, instead relying on machine learning to discover relationships between molecular structure and drug efficacy.
More recently, the United Kingdom has announced plans to pioneer the world’s first fully-automated drug discovery facility at the Rosalind Franklin Institute in Harwell, Oxfordshire. The facility will harness robotics and AI to investigate hundreds or thousands of candidate molecules simultaneously, thereby increasing productivity five to ten times. This is expected to significantly cut down on the time and cost associated with drug research - it often takes over ten years bring a new drug to market with a development cost of approximately USD $2 billion.
With the high costs associated with healthcare today, AI can make healthcare more accessible by augmenting the expertise of doctors and other medical professionals and scaling services to reach those who need it most.