This occupation has an extremely low probability of automation.
$ hourly wage
What they do
Risk of automation
This occupation is 0% likely to be automated. It ranks of 702 occupations analyzed (higher rank means higher likelihood).
People affected and economic value
In the US, people have this job.
Official growth forecast
This occupation is projected to shrink 0% between 2016 and 2026. At this rate, there would be people with this job at that time.
- Certainly, many aspects of cardiology—such as bedside manner—are less amenable to automation, but AI-driven diagnosis and analysis is already exceeding what the average cardiologist can do.
- Kardia (a portable EKG) by AliveCor catches four times as many atrial fibrillation cases as traditional screening, which has been proven in multiple studies.1
- Researchers have created an algorithm (by training a neural network) that exceeds the performance of board certified cardiologists in detecting heart arrhythmias.2
- This list will likely get long over time, as diagnosing dermatological conditions is something incredibly well-suited to existing machine learning technology. Machines have long had superhuman performance at classifying images.3
- VisualDx is an app that uses Apple’s Core ML to help non-dermatologist doctors diagnose lesions, rashes and other skin conditions.4 (It’s logical that this would cut down on referrals to dermatologists, so it would be great to see some studies of this eventually.)
- At a TechCrunch Disrupt hackathon in September 2017, developers created an app called “Doctor Hazel” that uses artificial intelligence to determine whether a skin condition is skin cancer.5 The speed with which they created the app shows how powerful and accessible AI technology is now. The biggest limiting factor is access to data, the developers say, but such data is already being crowdsourced by users for apps such as SkinVision.
- A number of companies are working to automate aspects of general medicine, especially diagnosis, which relies on the power of machine learning to identify patterns in huge amounts of data—data that’s growing rapidly through growth in adoption of electronic health records.
- A Chinese report claims that researchers there have created an AI doctor that can diagnose diseases 20% more successfully than human doctors.6
- Applying AI isn’t the only way to automate diagnosis. There is inefficiency (and delay) in healthcare with gathering and analyzing biological samples such as blood. But a number of startups are developing at-home tests that bypass the need for doctors and labs. For example, a company called Athelas has created a device some have referred to as the Amazon Echo of blood tests. It can test for diseases such as flu and infections in the comfort of your home.7
- Artificial intelligence isn’t just useful for general medicine, nor only for diagnosis. Researchers from Taiwan have trained neural networks on 340 million health records associating diseases with herbal treatments.8 The networks learned to associate herbal prescriptions with diseases, and diseases with herbal prescriptions. This allows machines to prescribe herbal treatments based on a diagnosis (a diagnosis that could, of course, also be made by a machine).
- AI can pick up patterns that are hard for humans to see or interpret, which is beneficial for diagnosing cancer. For example, researchers have created an app that uses computer vision algorithms and machine learning to diagnose pancreatic cancer from selfies based on identifying a yellowish tinge in the whites of the eyes, which is caused by buildup of the chemical bilirubin.9
- AI is better than humans at identifying the best embryos to choose for in-vitro fertilization.10
- Robots have become more common in surgery since being introduced in the early 200s. Between 2010 and 2017 in the UK, the number of hospitals using robots for prostate surgery more than tripled, from 12 to 42.11
References and notes
This article uses data from "The Future of Employment" (PDF), a 2013 study from University of Oxford researchers, and US Bureau of Labor Statistics files including Standard Occupational Classification 2010 Definitions (XLS), May 2016 Occupational Employment Statistics (zip) and 2016-2026 Employment Projections (XLS).