Artificial intelligence (AI) has been likened to a new species, “Auto Sapiens.” I know this is wild, but bear with me—it may actually help us “get along,” “collaborate,” and “lead” AI to improve radiology practice.
A recent Harvard Business Review article by Jeremy Heimans and Henry Timms explained Auto Sapiens: AI is able to act autonomously, make decisions, learn from experience, and operate without continuous human supervision, hence “Auto.” Also, AI possesses knowledge and the ability to make judgments in context, hence “Sapiens.” It is hard to think of Auto Sapiens when it’s software running on your data, but it is valid considering this terminology when thinking of AI used in humanoid robots.
As a thought exercise, let’s think of radiology AI as an Auto Sapiens, and a coworker in the role of an “assistant.” Is this assistant going to take our jobs? Will it (they?) help us in our jobs, and could it make radiology practice more profitable? I say, no, to taking our jobs. And, yes, to helping us change our practice for the better.
Here is how: I believe that AI will not displace radiologists. Somebody needs to be liable for mistakes made by AI. Medical malpractice can be established when physicians deviate from the profession’s standard of patient care. If a radiologist uses an AI-enabled medical device for diagnosis or treatment of a patient, and their use deviates from an established standard of care, the physician could be liable for improper use of that AI medical device. As of now, the radiologist must independently review the AI’s recommendations, applying the standard of care in treating the patient regardless of the AI’s output. After all, AI is an assistant needing supervision, right?
Holding AI developers liable is quite difficult. One would have to prove that the AI was defective at the time of product purchase by the user and did not become corrupt as it continued to train itself on user data. There are currently no sufficient industry standards to address this.
I doubt that insurers would take liability; they lack the expertise to minimize liability should an AI application go awry. Radiologists are the ones assuring AI performs consistently in accordance with their intended purpose and scope, as well as at the desired level of precision. Insurers do not have the expertise to check on radiology AI applications’ correctness, relevance, robustness, or interpretability. Radiologists will be the stewards of quality assurance for AI.
I do, however, wonder about a threat to reimbursements. It is conceivable that AI can evolve to perform better than radiologists—faster and with fewer errors. In that event, insurers could cut physician fees. We need to think about reimbursements in the AI era. Will there be a new component to the fee schedule, such as “AI supervision,” which entails auditing and supervising AI? We will need to continuously “invest” in our AI assistants to make sure they are trained up to the latest technological standard.
OK, now that I’ve argued how AI will not replace radiologists, let’s see how our new assistants will help us. First, Auto Sapiens, like a good assistant, will happily do all the stuff many of us like less about our jobs, such as reading endless chest radiographs, scrutinizing CT images for lung nodules, measuring lesions and transcribing measurements into reports, and so many other things. Yes, I want this assistant, like, now!
Additionally, Auto Sapiens will also help us decrease errors of perception and interpretation and delays for reporting incidental critical results, such unexpected intracranial hemorrhage on a nonemergent head CT. What is not to like about this type of assistant? Maybe liability insurance payments will even come down?
And all of this can result in a more profitable business? Sure, as soon as AI enables radiologists, technologists, and imaging equipment to handle larger volumes, there could be a massive increase of imaging orders. Dream on, though, if you think that decision support will help us control imaging utilization. Imaging is already being used in lieu of a thorough clinical exam. In fact, Dr. Joseph Alpert called the physical exam “an ancient ritual” in 2019.
So, the AI assistant can help us grow our business and focus on work we enjoy, like making a diagnosis and providing excellent services to physicians and patients. However, this model relies on us being proper supervisors to our AI assistants. As Curtis Langlotz, MD, PhD, once put it: “AI won’t replace radiologists, but radiologists who use AI will replace those who don’t.”
The time to learn about AI is now, and I am excited about it!
Nadja Kadom, MD
Director for Quality, Department of Radiology, Children’s Healthcare of Atlanta
Interim Director for Quality, Department of Radiology and Imaging Sciences, Emory Healthcare
Professor, Emory University School of Medicine