Where Health Tech Is Headed

AI moving from imaging into more of the record. The models covered in Part 4 started with scans because images are a clean, well-labelled data type. The same pattern-matching approach is already expanding into unstructured clinical notes and lab trends, not just pictures — the same underlying technique, aimed at messier data that used to be too unstructured for it to touch.

Monitoring moving earlier, before a diagnosis exists. Part 3's remote monitoring largely tracks people who already have a diagnosed condition. The next wave targets people who don't yet — flagging early patterns that predate a diagnosis, not just managing one that already happened, which means it inherits the same estimate-versus-diagnosis gap Part 3's wearable chapters covered.

Interoperability actually finishing, not just improving. Part 2 showed why standards took decades to catch on. Regulatory deadlines already in motion are pushing that adoption from “most large systems” toward “essentially everyone,” closing a gap this track spent an entire part on.

Each of those is already happening in some form, not just proposed. For each trend, ask what must remain true for it to keep working.

Every one of those three trends assumes the trust problems covered earlier in this track — bias in Part 4, access gaps in Part 5, security in this part — get managed well enough that adoption keeps expanding rather than stalling out after a visible failure. A single high-profile AI misdiagnosis, breach, or access scandal doesn't just damage one product; it can slow trust in the entire category for years, the way a plane crash affects flying more than the accident statistics justify.

That is not a hypothetical framing device. A hospital system deciding whether to adopt an early-monitoring product is weighing the same bias and access questions this track has already walked through — whether the training data behind it looks like its patients, and whether the product works for someone without a smartphone as well as it does for someone with the latest one. A trend can be technically ready and still stall for reasons that have nothing to do with the underlying technology.

Not every confident health-tech prediction deserves the same weight as the three above. The clearest example is the idea of an AI system reaching a full diagnosis and treatment decision with no clinician reviewing it at all, for anything beyond the narrowest, most tightly bounded case. It is a common prediction, it makes a better headline than any of the three trends above, and the evidence behind it does not currently support it — the bias and error problems in Part 4 are not solved, they are managed by exactly the human review step this prediction proposes removing.

The clinician stays in the loop. Nothing in the current trajectory of medical AI points toward removing a human reviewer from a real medical decision — if anything, the bias and error problems in Part 4 make that reviewer more clearly necessary, not less. Whatever health tech looks like in a decade, the pattern established across this entire track — software assisting a decision, not making it alone — is the one part that isn't up for negotiation.

That is a different kind of prediction from the three trends above. Those are bets on where the technology goes next. This one is closer to a constraint the rest of the track has been building toward the whole time — not a guess about the future, but a boundary nothing so far has given a real reason to expect will move.