E-Paper
Next intelligence revolution won’t belong to machines
This article is authored by Kalyan Sivasailam, founder & CEO, 5C Network.
When Deena Mousa wrote in Works in Progress that AI is not replacing radiologists, she captured the line everyone now feels safe repeating. I run India’s largest diagnostic AI network. We deploy the systems that are supposed to do the replacing, and we move more than ten thousand studies a day through them. From where I sit, the headline is different. AI is not failing. It is merging with the people who use it. The next chapter of medicine is hybrid intelligence, where algorithmic speed and human judgment learn from each other in real time.
Radiology shows this merge in the clearest light. For years we all built islands of automation. One model for nodules. Another for pleural effusion. Another for consolidation. Each island performed, none of them reasoned together. In production, stitching islands into a reliable chain is brittle work. Change one operating point and the downstream logic collapses. The way out is not more islands. It is whole-image, multimodal models that take the entire study as context and return a coherent read. That is where the field is finally heading.
Here is what happens once you put this into practice. The machine flags patterns at scale and with a memory that never fades. The radiologist tests that signal against the clinical picture and local disease burden. The correction loops back into the system within hours, so the next time that pattern appears the model catches it first. Humans and machines are not competing in this loop. They are co-evolving inside it.
The economics tell the same story. When you raise productivity in imaging, you do not shrink the market. You expand it. Outpatient imaging volumes are projected to keep climbing over the next decade, with standard outpatient imaging growing by about ten percent and advanced imaging by nearly fourteen percent. That is what elasticity looks like in the real world. Every time you make a unit of work faster and more consistent, you unlock demand that was stuck behind cost, access, or staffing limits. Rural centres that could not justify a radiologist start sending cases. Urban hospitals package imaging into same-day programmes. Employers and insurers add scans to protocols that once felt impractical.
If you want to see why this merge matters for the profession, look at how radiologists actually spend their time. Only about a third of the workday goes to pure image interpretation. The rest is protocol design, communication, quality assurance, board meetings, and a long list of tasks that are essential to care but do not require pixel-level pattern finding. AI does not threaten the core of radiology because pattern recognition was never the whole job. What AI does is strip out the mechanical parts and force a choice. You can supervise high-volume normal studies and manage throughput. Or you can lean into the cases that need clinical correlation, tumor boards, surgical planning, and patient conversations. The second path is where value accumulates as the loop gets tighter.
This is where the Deena Mousa debate needs an update. Saying that AI is not replacing radiologists is true and incomplete. The more interesting development is that radiologists are becoming bionic professionals. Part clinician, part system designer, part reader of the machine that reads the image. That shift is already underway in many departments. You can see it in the day-to-day tools radiologists are adopting for drafting reports, triaging lists, and closing follow-up loops. You can also see it in the regulatory landscape, where the majority of authorised AI devices sit in medical imaging, and where the momentum is toward decision support that fits cleanly inside clinical flow.
When the machine starts missing what the human does not see, you know hybrid intelligence has arrived. That is the moment when the partnership becomes obvious. The model is fast, tireless, and consistent. The clinician is contextual, skeptical, and accountable. Together, they make fewer mistakes, move faster through routine work, and free up the attention that complex cases demand.
So, the question is no longer whether AI will replace anyone. The better question is whether we are ready to practise in systems where learning never stops. In radiology, that means building infrastructure for continuous data, rapid adjudication, site-aware thresholds, and model refresh that is measured in days, not quarters. It means training clinicians who are comfortable interrogating an algorithm and adjusting it, not just approving its output. It means leaders who invest in the plumbing that compounds advantage over time.
This is the real disruption. AI is not coming for radiologists. It is coming for the parts of medicine that were mechanical, siloed, and slow to learn. The sooner we embrace hybrid intelligence as the operating model, the sooner we can focus on the work only humans can do.
This article is authored by Kalyan Sivasailam, founder & CEO, 5C Network.
