When someone you care about needs a treatment that does not exist, the possibility of AI accelerating medical discovery is difficult to regard as just another technology story. You want the researchers to have better tools. You want promising ideas to get a serious test. You also want to know whether the announcement you just read changes anything for the person waiting.
That last question is where the conversation gets difficult.
An AI system might identify something scientists had overlooked, help design an experiment or suggest a molecule worth investigating. Each could be a meaningful contribution. None, by itself, tells a patient that an effective treatment is available.
That distance between possibility and benefit is the subject of our new Techstrong Special Report, AI and the Promise of Better Health: From Discovery to the Patient. The report examines the scientific ambition, the evidence emerging from research and clinical care, and the decisions that determine whether useful advances reach people.
Download the AI and Better Health Report (PDF)
I am optimistic about the opportunity. I would like us to become much more demanding about how we describe it.
Consider Anthropic’s announcement about Claude and a newly characterized enzyme system. The work concerns a biological finding whose function remains unknown, with human scientists performing the laboratory experiments. That is a reason for further investigation. It is not evidence that Claude has produced a new treatment or a replacement for CRISPR.
The distinction does not diminish the discovery. It gives readers a way to understand what has actually happened and what researchers still need to establish.
Our report applies that discipline across a much wider set of possibilities. Could AI help scientists choose better experiments? Could it make the search for medicines more productive? Could it help explain why a treatment works for one group but not another? Could highly individualized medicine become practical for more patients?
Those are substantial ambitions without adding a promise that disease is about to disappear.
I find it useful to think about three clocks. The discovery clock measures the search for something worth pursuing. The evidence clock measures the work needed to establish what it does, for whom and at what risk. The delivery clock measures the distance between an effective intervention and a person actually receiving it.
AI might accelerate parts of all three. But speeding up one does not automatically speed up the others.
A promising candidate still needs an appropriate test. A benefit that appears early may not last. Manufacturing, clinical capacity, payment and access remain consequential even after the scientific questions have encouraging answers. The report follows those distinctions rather than treating every stage as another version of the same breakthrough.
It also examines evidence from patient care. The MASAI breast-screening trial, for example, offers a more useful conversation than a general claim that AI can detect cancer. Its interval-cancer analysis met a specified noninferiority standard. That is not the same as proving superiority or showing that fewer people died of breast cancer.
We should be able to value a result without asking it to establish something the study did not measure.
For the people building AI products, this changes what a convincing demonstration looks like. Show how the tool affects the complete workflow, including the work people do to review, correct and act on its output. A faster answer is not necessarily a faster route to appropriate care.
The economics deserve the same scrutiny. If AI makes a task less expensive, who receives the saving? Does a practice offer appointments sooner, give clinicians more time with patients or simply increase the workload? Does a lower research cost make treatment more affordable, or does the benefit remain elsewhere?
Those are management and policy choices, not outcomes a model can guarantee.
The report also asks who gets access to the healthier future being described. A useful service can help someone who previously had little assistance. That does not mean an automated answer should become the only option for people who cannot afford a clinician. We need to distinguish expanding access from offering a cheaper substitute for care that remains out of reach.
I do not want the limitations to become an excuse for dismissing the work. Better scientific tools and carefully tested clinical applications are worth pursuing. So are the less glamorous improvements that help an effective treatment reach a patient reliably and affordably.
But the measure cannot be how many AI systems healthcare adopts. It has to include what people become able to do because their health is better.
Read the full Techstrong Special Report for the research, clinical examples and access questions behind that promise.
