With so much industry buzz around artificial intelligence’s ability to transform healthcare, it’s easy to lose sight of the fact that not all AI is created equal.
To appreciate AI’s full potential to improve healthcare quality, lower costs and enhance patient access, it’s important to understand the difference between autonomous and assistive AI.
Assistive AI can take measurements and help physicians make decisions, working like a second set of eyes to review medical images, for example. In contrast, Autonomous AI solutions are capable of actually making clinical and treatment decisions that impact patient care and outcomes.
To illustrate the difference, I like to point to the self-driving car industry, which is working to navigate the transition from assistive to autonomous AI. Today, there are cars on the market that can parallel park, drive on the highway, and avoid road blocks – but these vehicles still require a human to sit in the driver’s seat and provide oversight of the system. That’s assistive AI.
The industry is working toward a different use case though – one that functions without any human intervention or oversight whatsoever. In that context, a truly self-driving automobile takes on a lot of responsibility; it has to be capable of responding to a host of different scenarios and challenges – weather, road signs, pedestrians – without human intervention. That’s autonomous AI.
What autonomous AI means for healthcare
From an algorithmic performance perspective, the performance bar is much higher for fully autonomous vehicles versus assistive ones. When cars take over driving responsibility there is no human “safety net” to fall back on.
As it relates to healthcare, the result is that autonomous AI must undergo far more stringent testing before it can be used in the practice of medicine, which enables clinicians to have confidence in the clinical decisions autonomous systems make.
The higher bar associated with autonomous AI is worth pursuing because it yields significant rewards – especially in the realm of enhanced productivity. The shift from assistive to autonomous AI holds the potential to unlock significant economic value via productivity gains realized from physicians being freed from having to perform routine functions. That time could be spent in any number of valuable ways: spending more time counseling patients or analyzing more complex cases and diagnoses. Whatever the scenario, the point is that autonomous AI enables productivity gains that can produce tangible benefits for healthcare.
However, for AI to fully realize its transformative economic potential, it is critical that patients and their clinicians develop trust in autonomous algorithms’ accuracy, reliability and objectivity. That starts with developers offering transparency into the algorithms’ capabilities, data training sources and validation processes. Once physicians begin to develop confidence in specific AI systems, the productivity gains and clinical enhancements enabled by AI will multiply as they spread through the healthcare industry.
A real-world example in the primary care setting
To examine what the application of autonomous AI might look like in healthcare, take the example of a person with diabetes undergoing their recommended annual eye exam. It is essential that all people with diabetes receive a yearly vision test, because all are at risk of developing diabetic retinopathy (DR), a preventable disease that is the leading cause of blindness in working-age adults, according to the International Agency for the Prevention of Blindness.
Typically, the patient would see their primary care physician at a routinely scheduled visit to discuss any complications related to their condition. For the DR exam, though, the primary care physician refers the patient to a specialist, usually an ophthalmologist. From the patient’s perspective, receiving the DR exam usually requires another appointment, an additional co-pay and more time in a waiting room – all for a procedure that results in a negative diagnosis the majority of the time.
In contrast, imagine a scenario in which an AI system that incorporates an easy-to-use camera autonomously identifies DR in medical images in the primary care setting. This technology provides almost-immediate results, help make sure that people with disease are seen by a specialist and unlock new forms of economic value for virtually all stakeholders.
In this scenario, the primary care physician is able to offer more comprehensive care, ensuring people with diabetes are getting tested for DR. The specialists receiving referrals will see a higher density of DR and spend more time treating disease, rather than merely looking for it. Payers may benefit from saving costs on often-unneeded specialist visits. Most significant, though, are the benefits that accrue for people with diabetes and the healthcare system overall: lower costs, greater access and the gift of time more productively spent.
This is not a scenario in the distant future – the technology is available today. The FDA recently authorized the marketing of IDx-DR, which detects DR in primary care. It is the first autonomous AI diagnostic system that does not require a specialist to interpret the images.
It’s important to understand that autonomous AI may not be appropriate for lots of use cases in medicine. However, there are cases where there is a clear set of decision-making criteria related to a given disease state and where patient care access barriers are real – these are areas where autonomous AI can have a profound impact. By focusing today on ways autonomous AI can lead to improved care access without compromising care quality, we can begin to see the outlines of how this promising technology might transform healthcare tomorrow.
Ben Clark has served as chief operating officer of IDx, a privately held AI diagnostics company, since 2012.