Learning the Language of AI
Artificial intelligence (AI) has its own language. So does economics. I feel pretty confident with econ-speak and can throw out phrases like intensity of work or acronyms like CPT, HOPPS, and QPP in rapid fire. But I am not as well-versed in the language of AI — a point that became apparent during a recent “Economics of AI” talk I gave to a room full of radiology AI experts.
There I was, espousing the details of radiology payment policy and speculating on how AI might tie into payment policy. With confidence, I explained how payment systems are structured and where AI may fit into that structure.
I illustrated a Current Procedural Terminology (CPT) code set for a head CT without AI and a separate CPT code for a head CT with AI. I can’t recall if I used the term AI or a different term, such as machine learning, algorithm, software, or application. (Remember that I am no expert on the language of AI, so all of these terms sound the same to me.)
After my presentation, I proudly waited for questions and feedback from the AI experts in the room. But to my dismay, silence followed. I quickly realized I had confused, more than enlightened, my audience. The fact that my AI terminology was inconsistent may have worsened the problem.
Finally, a hand went up in the audience. Great, I thought, here comes a question to which I can respond with gleeful stories of the RUC and RVUs and coverage determinations, or some other economic acronyms and phrases.
The audience member said, “But AI is not a thing to simply be added to a head CT. It is much more than that.” He proceeded to describe AI with a handful of terms I didn’t necessarily recognize or understand. But what was clear was that he and I were speaking two different languages.
Herein lies the challenge: economics experts and AI experts must learn each other’s language, at least superficially. And somehow we must foster individuals willing to become fluent in both languages.
Our success in this effort is important to AI’s growth and radiology application. Take the scenario I described above and reverse it. Now imagine an AI expert presenting his or her AI application to a group of payment-policy experts, charged with deciding whether to pay for it and/or how much.
The AI expert, no doubt, could easily overwhelm the payment experts with his or her knowledge. But is that the desired outcome? If the payment experts fail to understand the AI terminology or to consider it within the context of their language, a poor outcome could follow — maybe no payment or, worse, a missed application and resultant decreased payment. For instance, what if the payment experts fail to see where this AI application fits into my earlier head CT example, and what if they lower the payment for all head CTs — or for any CT or radiology study?
The scenarios I describe for AI are not specific to AI when it comes to payment policy. Clinical experts have long found themselves defending emerging technologies to payment experts. And AI experts are very much clinical experts. Imagine the first time a physician explained the “work” of interpreting an MRI or the placement of an abscess drainage catheter to a group of payment policymakers.
So to radiologists in general — and my fellow economics experts in particular — learn the language of AI. And to all of you AI gurus, do the same with economic policy — at least casually. Our ability to advocate for the appropriate recognition of AI’s value in radiology patient care may depend on it.
By Ezequiel Silva III, MD, FACR, FSIR, chair of the ACR Commission on Economics and medical director of radiology at Methodist Texsan Hospital