AI Is Already Diagnosing Diseases Better Than Some Doctors — And We're Arguing About Paperwork

The conversation around artificial intelligence in healthcare has been stuck in a holding pattern for years. Too many people are debating ethics committees and liability frameworks while actual patients are waiting months for a radiologist to look at a scan. I think that's backwards, and I think we're wasting time.

AI-powered diagnostics should be standard care in every major hospital by the end of this decade. Full stop. Here's why I mean that seriously and not as a tech-enthusiast slogan.

Radiology Results Shouldn't Take Six Weeks

The first reason is speed, and it's not a small thing. A tool called Enlitic, deployed across several hospital networks in Australia and parts of Southeast Asia, cut average radiology turnaround times by roughly 60 percent in pilot programs. That's not a rounding error. That's the difference between catching a tumor at stage two versus stage four.

Human radiologists are excellent. They're also overworked, understaffed, and increasingly concentrated in wealthy urban centers. Rural hospitals in the American Midwest, parts of sub-Saharan Africa, and regional clinics across South Asia simply don't have enough of them. An AI system doesn't need to be flawless to be useful — it needs to be faster and available at 3 a.m. in a clinic in Nebraska that has one doctor on call.

Rare Diseases Are Finally Getting Attention

The second reason is diagnostic accuracy for conditions that most general practitioners will see maybe twice in a career. A platform called Isabel DDx, used in over 7,000 hospitals globally as of early 2024, cross-references patient symptom data against a database of more than 10,000 conditions. That's not something a human brain can reasonably hold.

Consider what that means for rare disease patients, who on average wait between four and seven years for a correct diagnosis. Some of that delay is unavoidable. But a large chunk of it is just pattern recognition that a well-trained model can do in seconds. I find it genuinely hard to argue that prolonging that suffering is acceptable when the tools exist to reduce it.

The "AI Replaces Doctors" Fear Is Real but Misdirected

Here's the counter-argument that deserves respect: AI systems can encode bias, make catastrophically confident errors, and get deployed before they're ready by companies chasing contracts. Those concerns are legitimate. A 2023 internal audit at a mid-sized insurance-linked diagnostics company in Illinois found that a triage algorithm was systematically underscoring symptom severity in female patients over 50 by nearly 12 percent. That's a real problem.

But the answer to a flawed tool isn't to throw out the category. It's to fix the tool, regulate deployment standards, and require transparency in how these systems are trained. The medical profession already runs on fallible human judgment — it always has. The goal isn't perfection. It's whether AI-assisted diagnostics produce better average outcomes than purely human diagnostics under current resource constraints. The evidence says yes.

Doctors should not be replaced. They should have better instruments. That's the actual argument.

What Responsible AI Deployment Actually Looks Like

Getting this right involves a few non-negotiable things:

• Clinical validation trials before deployment, not after — ideally across diverse patient populations and not just data from three academic hospitals in Boston


• Mandatory bias audits every 18 months, especially for models used in triage or diagnostic prioritization


• Clear liability frameworks so that when an AI system contributes to a diagnostic error, there's a legal path forward for patients


• Open-source access to training data sets for any publicly funded healthcare AI, so independent researchers can actually check the work

None of this is radical. It's just what good engineering looks like in a high-stakes environment.

Pathology Labs in 2025 Are Running a 2009 Playbook

Walk into most pathology departments today and you'll find talented scientists doing enormously skilled work with tools that haven't fundamentally changed in 15 years. AI-assisted pathology software, like PathAI's systems currently being piloted at institutions in Texas and North Carolina, can flag anomalous cell structures in tissue samples in minutes rather than hours. Oncology patients waiting on biopsy results to decide treatment paths don't have extra days to spare. The technology exists. The resistance is institutional, not scientific.

I think the honest version of this debate is less about whether AI should be in healthcare and more about who controls it. Health systems should own their diagnostic infrastructure, not rent it from venture-backed startups with opaque pricing. That's a policy fight worth having.