Healthcare has traditionally relied heavily on snapshots.
A blood test captures what is happening at one moment. Blood pressure provides another measurement. Imaging gives us a structural view at a particular point in time. Even many wearable devices currently focus on a relatively small number of signals such as heart rate, activity, sleep, or glucose.
A new generation of health technologies is beginning to explore something different: continuous, multidimensional monitoring of human biology.
Biosensors: Beyond Heart Rate and Glucose
The U.S. Advanced Research Projects Agency for Health (ARPA-H) recently announced research teams for its Delphi program, an initiative designed to advance next-generation personalized biosensors.
The program is exploring modular, miniaturized sensor technologies that could eventually allow multiple biological signals to be monitored through adaptable platforms.
Research directions include biomarkers related to inflammation, hormones, drug concentrations, and other physiological processes.
This represents an important conceptual shift.
Instead of asking only:
“What is this biomarker today?”
future monitoring systems may increasingly ask:
“How is this biological signal changing over time, and what was happening when it changed?”
However, Delphi remains a research and development program. These technologies should not be interpreted as established clinical tests or currently validated consumer health devices.
Medical AI: Knowing When to Hand the Decision Back
More biological data also creates another challenge: who—or what—interprets it?
Recent discussion in Nature Medicine has highlighted the concept of selective autonomy for agentic medical AI.
The important question is no longer simply whether an AI system can produce an accurate answer.
A clinically useful system may also need to recognize uncertainty.
It must potentially determine:
When can AI continue?
When should it stop?
When should the decision be handed back to a clinician?
This moves medical AI beyond the question of model accuracy and into clinical workflow, human oversight, accountability, and governance.
The future of medical AI may therefore depend not only on how much autonomy machines receive, but also on how intelligently that autonomy is limited.
Multi-Cancer Blood Testing: Detection Is Only the Beginning
Multi-cancer early detection (MCED) blood tests represent another rapidly developing area.
GRAIL’s Galleri test has recently advanced further through the U.S. regulatory process, including preliminary support from an FDA advisory committee. Advisory committee support, however, should not be confused with final FDA approval.
The larger scientific and clinical questions extend beyond whether cancer-related signals can be detected in blood.
If a test produces a positive result:
What diagnostic procedure comes next?
How many patients will require additional imaging, endoscopy, or biopsy?
What are the consequences of false-positive findings?
Can earlier detection ultimately improve meaningful clinical outcomes?
And how will these technologies be integrated into healthcare systems and insurance coverage?
These questions illustrate a fundamental principle of modern diagnostics:
Detection is valuable only when we understand what to do with the information.
More Data Does Not Automatically Mean Better Healthcare
Biosensors, medical AI, and multi-cancer blood testing may appear to be very different technologies.
But they are part of the same larger transformation.
Healthcare is gradually moving from isolated measurements toward continuous, contextual, and multidimensional information.
This creates extraordinary opportunities—but also a new responsibility.
Every new health measurement should ultimately be evaluated through three questions:
What exactly are we measuring?
How reliable and clinically meaningful is the signal?
Will knowing the result change what we do next?
The best diagnostic technology may not be the one that measures the greatest number of variables.
It may be the one that provides the greatest amount of actionable information at a reasonable cost.
That distinction will become increasingly important as healthcare enters an era in which collecting data may become easier than deciding what the data actually means.
Science & Education:
BI 身体智慧（Body Intelligence）
AI-assisted Research & Illustration:
BI × GPT
Professional Review:
林存默（Thomas Lin）
Professional Community:
ACPN — The Association of Certified Professional Nutritionists
For educational purposes only. This article does not constitute medical advice, diagnosis, or treatment.