In Part 1 of his two-part series, radiologist Josh Ewell, DO, discussed developments in AI technology and the potential impact of these advances on radiology.
A self-reinforcing spiral
Initially, AI tools will fulfill their promise to relieve the burden of our worsening radiologist shortage: more studies can be read in less time. Yet economic principles remain unforgiving. As the supply of “interpretive capacity” swells (thanks to AI), reimbursement for each study will predictably decline. This is simple supply-and-demand economics.
Joshua Ewell, DO
Correspondingly, so will the financial incentives that once made radiology attractive. This downward spiral will almost certainly drive a secondary exodus from radiology, not only from those practicing near or past the age of retirement but also from our recruitment pool of medical students, leading top talent to bypass a field whose core tasks are being rapidly automated.
Agentic vs. human-driven care
Andrew Ng, another luminary in machine learning, has described radiology as being “ripe for disruption” by deep learning. If agentic AI can handle large swaths of interpretive tasks, clinicians might come to rely heavily on those automated reads. Remaining radiologists -- fewer in number -- would pivot to roles emphasizing complex case reviews or advanced procedures. Over time, however, the “safe zone” for human radiologists will continue to shrink, particularly as these AI systems enhance their generalization.
Moral and existential dimensions
Technological shifts in healthcare inevitably carry philosophical and even spiritual implications. If radiologists become relegated to niche consultative roles or replaced in routine interpretation, how will that affect the physician-patient relationship and broader notions of healing?
Eric Topol, MD, in his book Deep Medicine, suggests that AI could liberate clinicians to focus on empathy and care. Yet from a more pessimistic vantage point, if economic incentives shape healthcare delivery more strongly than compassion (and they do), the imperative to integrate holistic human elements could be sidelined in favor of cost savings and throughput.
Some might argue that a purely efficiency-based approach underestimates the intangible qualities of human judgment -- qualities that can be neither coded nor easily replicated by algorithms, however sophisticated. Others, such as Demis Hassabis, PhD, of DeepMind, contend that self-improving AI systems will ultimately surpass even those intangible components in ways we cannot yet imagine.
Crossroads
Amid these debates, we as radiologists stand at a crossroads: is the human element in imaging interpretation dispensable or indispensable? As agentic AI systems advance toward artificial general intelligence (AGI), the answer, although frequently dismissed by thought leaders, is a resounding of course, and this epiphany is one that is fraught with far-reaching implications for our field.
Perhaps, this is why so many are quick to dismiss it out of mind. Eliezer Yudkowsky has made a career in AI safety research, having frequently recounted a mantra (paraphrased here): "While we can’t predict the timing or the trajectory, the endpoint is certain." The concern is that, as a specialty, we have bought into the marketing-driven concept that radiology will not be replaced by AI, but rather augmented by it.
This is certainly true within a liminal space, but the endpoint is clear: We interpreters of medical imaging will be replaced. Historically, the resounding voices in our field have been quick to dismiss these concerns. However, on further discussion, it becomes clear that this position is carried by the winds of “not in my lifetime.” Further, these optimistic perspectives are not typically rooted in a depth of knowledge about AI technology, but rather in ego -- much like John Henry, we risk dying with a hammer in our hand, winning a single battle while losing the war.
A glimpse of the singularity
AI development moves with a rapidity not seen in technology before. Amid all these transformations, the discourse around AI has only grown more charged. On X, OpenAI CEO Sam Altman recently penned a striking six-word story that epitomizes the uncertainty of our era:
“I always wanted to write a six-word story. here it is: near the singularity; unclear which side.”18
This brief post raises profound questions. Referring to our current state as a ‘singularity’ underscores our limited ability to predict the true scope of these models’ capabilities. Are we as radiologists merely spectators to an unstoppable progression of AI, or do we still have a role in shaping it? Will the technological leaps of the coming years eclipse the uniquely human capacities of empathy, judgment, and creativity that physicians bring to patient care, or can we harness them to enhance -- and not eliminate -- our profession?
Conclusion
The John Henry metaphor resonates powerfully with this generation of radiologists and serves as a parable for the liminal space between human- and machine-based medical diagnoses. While the steel-driving man defeated his mechanical foe, he lost his life in the effort.
Radiologists today confront a challenge far more complex than a steam hammer. Agentic AI may dramatically relieve their workload, only for the field’s economic engine to sputter out. Then, as AGI and eventually artificial superintelligence (ASI) emerge (with early predictions from Anthropic CEO Dario Amodei raising the alarm for rapid and imminent progress), the very foundation of image interpretation -- once the hallmark of radiology -- could be subsumed by non-human intelligence.
Yet there could be a “Goldilocks zone” in which a reduced radiologist workforce -- thinned by retirements, AI-induced exoduses, decreased recruitment, and declining reimbursements -- still matches patient needs, albeit in a more streamlined manner. In such a scenario, modern AI tools would handle the bulk of “remote” interpretive radiology work, while an attrition-reduced cadre of human radiologists focuses on complex consultations, challenging cases, and both physician and patient consultations. However, achieving this sweet spot would still represent a massive upheaval of today’s practice models, demanding new workflows, reimbursement frameworks, and role definitions.
It is also crucial to note that procedural specialties -- specifically, image-guided procedures -- require an onsite physician presence. These surgical or procedure-based roles will likely remain safer harbors for some time, as AGI’s imposition upon the physical labor aspects of healthcare is expected to lag behind pure interpretive capabilities.
Thus, while Geoffrey Hinton’s warning to “stop training radiologists now” may have been hyperbolic, it embodies a pressing question: does our specialty face unstoppable obsolescence? Or can it evolve into something unassailably human-centered and clinically comprehensive, sheltered from purely automated interpretation?
For those who see a spiritual or existential mission in healing, there may be new meaning to be found in forging deeper patient connections, focusing on procedures and direct patient care, as well as guiding AI interpretation in a manner that preserves dignity amid professional disruption. Yet from a practical and economic standpoint, the forecast is decidedly grim without proactive reinvention. Radiology stands at the threshold of a new era -- one that could swiftly alter its identity in ways even John Henry’s story cannot fully foreshadow.
Joshua Ewell, DO, is the medical director for Synergy Radiology, a teleradiology division of Summit Radiology in Fort Wayne, IN, and Spectrum Radiology in Maine. The opinions expressed in this editorial are his own and do not represent the views of his employer.
The comments and observations expressed herein do not necessarily reflect the opinions of AuntMinnie.com, nor should they be construed as an endorsement or admonishment of any particular vendor, analyst, industry consultant, or consulting group.
References
Whether you are a professional looking for a new job or a representative of an organization who needs workforce solutions - we are here to help.