When AI Gets It Right (and Wrong) on Women’s Health
Why inconsistent AI diagnoses aren’t a breakthrough, but a signal that both medicine and machine learning are built on incomplete data and what that means for investors
📕Special note: My book The Billion Dollar Blindspot is available for pre-order on Amazon. It’s currently #3 on Amazon hot new releases in our category. Will you help me make it to #1?
I posted a question on Substack earlier this week which resonated deeply. It was deceptively simple and it read:
“We trust AI because it’s “data-driven” but women’s health remains one of the least-studied areas in medicine. So what data is it actually using?”
The responses came quickly. Thoughtful ones. Frustrated ones. The kind of frustration that accumulates quietly rather than erupts all at once. Then one reply made me stop.
This essay is inspired by that interaction.
She had gone to her doctor with symptoms that didn’t feel right. The response she got was familiar in its tone, if not its consequence: It was nothing urgent, she was told. Nothing requiring escalation. That should have been the end of it but she knew something wasn’t right. She left the doctor without an answer but not without doubt.
At home, she opened ChatGPT and began to type in her symptoms. She didn’t turn to a LLM because she trusted it unconditionally but because she no longer fully trusted what she’d just been told.
That distinction matters.
Her skepticism wasn’t just about the doctor. It was about the system. A system she understood, at least intuitively, to be built on data that had not always included women in full. When she entered her symptoms, the model flagged a potential uterine cancer risk. It didn’t offer certainty. But it changed the direction of her attention.
She went back to the doctor and this time, she pushed. She insisted on being seen and taken seriously. This time the system responded. The diagnosis was confirmed.
The story travels because it feels like resolution. A model surfaces what a doctor missed. Technology corrects for human bias. The problem, it seems, has already been solved or at least the path forward is clear: if we deploy AI, we’ll have better outcomes for women’s health.
But that conclusion doesn’t hold.
Because the same system that surfaced risk in this case can fail to recognise it in others. A recent benchmark evaluating leading AI models on women’s health found failure rates of around 60 percent with particular difficulty identifying when symptoms require urgency.
The issue isn’t whether AI can be right. It’s that it isn’t reliably right. The same system can surface risk in one case and miss it in another while sounding equally certain in both.
And that inconsistency isn’t random. It follows the contours of the data it was trained on.
What the writer encountered isn’t evidence of a system that has solved for bias. It’s a moment where two systems—clinical medicine and artificial intelligence—produce different readings from the same signals. That divergence resists easy interpretation. It doesn’t tell us which system is better but what it tells us is that the foundation beneath both is uneven.
Both systems are downstream of the same source: data. If women have been historically underrepresented in clinical trials (only mandated for inclusion in the U.S. as recently as 1993), if their symptoms have been more frequently dismissed or misread, then those gaps don’t disappear as the system evolves. They are carried forward. The difference is in how they surface.
In a clinical encounter, the gap shows up as dela and as uncertainty. It shows us as a quiet tendency to normalise what hasn’t been fully studied. In an AI system, the output arrives with confidence but that confidence rests on uneven ground. Sometimes it catches what medicine misses. Sometimes it misses what medicine might have caught. Neither outcome is neutral.
This is where the conversation has to move beyond tools.
Much of the current focus on AI in healthcare sits at the interface layer: better diagnostics, faster processing, more accessible platforms. The assumption is that improving the interaction improves the outcome. But the writer’s story points somewhere else. It points to the layer beneath.
Because what isn’t consistently seen isn’t consistently measured. What isn’t measured isn’t systematically funded. And what isn’t funded doesn’t become a market. The gaps that begin in data don’t stay contained within clinical encounters. They propagate outward, shaping what gets researched, what gets built, and where capital flows.
Markets don’t ignore what is visible. They ignore what is inconsistent. And in women’s health, inconsistency has too often been mistaken for absence.
The underdevelopment of women’s health isn’t just a medical oversight. It is a capital allocation story; a reflection of how systems, clinical and technological alike, translate incomplete information into decisions, and how those decisions, compounded over time, shape markets.
Her story isn’t proof that AI is better than medicine. It is evidence that we have built a healthcare system, and now an AI layer on top of it, on an incomplete foundation. And that, occasionally, that foundation reveals itself, not only through failure, but through inconsistency.
The uncomfortable question isn’t which system to trust. It’s what both systems have been built to see, and what they have been built, by design or by default, to miss.
Because in healthcare, as in markets, what remains unseen does not stay unpriced forever. But until it is properly seen, the cost of that gap is not absorbed by the system. It is carried by individuals, one dismissed appointment at a time until something, or someone, forces the system to look again.
Three years ago, I started writing what I thought was a report. It became a book. Today, The Billion Dollar Blindspot is available for pre-order. But this didn’t start as a book. It started in a gynecologist’s office; confused, dismissed, and aware that something wasn’t adding up, not just medically but structurally.
P.S.
By popular request, I am re-starting my Signal Not Noise series on Substack, decoding what is happening in women’s health capital markets. Written by a 25-year veteran of the global financial markets. Look for it in your inbox every other Thursday.
Disclaimer & Disclosure
This content is for informational and educational purposes only. It does not constitute financial, investment, legal, or medical advice, or an offer to buy or sell any securities. Opinions expressed are those of the author and may not reflect the views of affiliated organisations. Readers should seek professional advice tailored to their individual circumstances before making investment decisions. Investing involves risk, including potential loss of principal. Past performance does not guarantee future results.