Sign in

When patients can't speak, AI must learn to listen

Edge AI wearables could detect changes in a patient’s personal physiological baseline, helping caregivers identify distress sooner. 

Published on: Nov 18, 2026, 10:34:19 IST
Share
Share via
  • facebook
  • twitter
  • linkedin
Copy link
  • copy link

Imagine living with severe pain but having no way to tell anyone about it. For millions of people affected by critical illness that cause cognitive and speech impairment, this isn't something to imagine--it's their reality.

Representational Image (Pixabay)
Representational Image (Pixabay)

As AI reshapes the health care industry, its responsibility shouldn’t be in replacing clinicians. Instead, it must be in helping them understand non-verbal cues of these patients. And with intelligent wearables that infer physiological signals and convert them into actionable insights, that’s very much possible, without having to compromise on privacy.

A study by the INCLEN Trust found that roughly 1 in 100 children in India under age 10 may have autism, with nearly one in eight showing at least one neurodevelopmental condition. Among older adults, dementia is just as widespread: A nationwide analysis in India estimated that about 8.8 million Indians over 60 already live with dementia, at a national prevalence of roughly 7.4%. In addition to this, is the differently abled community, who like the previous cases, can hardly communicate their physical ailments.

The default fallback for medical care in all these situations is human observation: a nurse noticing a frown, a caregiver sensing something is off. While this vigilance is commendable, it is never scalable.

Consumer wearables have made physiological data mainstream—step counts, heart rate, sleep scores. But there's a wide gap between fitness tracking and clinical-grade monitoring. A device built to nudge someone toward more movement isn't built to detect a medical emergency.

Heart rate alone is a poor proxy for pain or distress; it rises for countless benign reasons, from a warm room to a stressful call. Systems tuned to averages tend to generate high false-positive rates, eroding trust with every unnecessary alert. What's often missing is a sense of the individual: each person has their own baseline, and meaningful change is relative to that, not a population norm.

This gap matters more in India than the raw numbers suggest, because the wearables market here is scaling fast in the fitness-and-lifestyle direction. India's wearable medical device segment alone is projected to grow to $ 4.20 billion by 2033, and the broader remote patient monitoring market is expected to expand even faster, to over $ 34 billion by 2035. Most of that growth is still being built on consumer-grade sensors optimised for step counts and sleep scores, not for detecting a non-verbal patient's escalating distress.

Edge AI has the tremendous potential of addressing these challenges. Instead of measuring each patient against a standard, population-generic threshold, on-device models can perceive and individual’s particular baseline. It is akin to having a personal doctor at the individual’s call, who can diagnose based on their heart rate variability, skin, and movement.

This shift is necessary as medical distress is rarely recognised from a single signal. Accurate diagnosis requires a collective data set across different conditions. Reading these signals on-device in the wearables, is advantageous in two ways: low latency, so that the insights arrive in real time, and better privacy, since sensitive data never leaves the device.

The motive behind this shift isn’t to handover medical care to machine. It is to equip clinicians and caregivers with the technology required to identify the discrepancies and mitigate them through human judgement.

It's tempting to think of this as primarily a hardware and sensor problem. It isn't. The greatest challenge isn't collecting bio signals—it's ensuring that every prediction is accurate enough for health care.

What’s required is clean, well-labelled data collected under realistic conditions, not just lab simulations. Models could use these real-world data to refine themselves and present their insights, so that they can be trusted with patient safety and reading credibility. Privacy must be plugged in, not layered on as an afterthought.

None of this is optional. In consumer tech, a wrong prediction is an inconvenience. In healthcare, it can mean a missed emergency—or a caregiver who stops trusting the alerts altogether.

The future of health care won't be defined solely by more powerful models or sensors. It will be defined by technologies that make care more humane. In the current market, intelligent wearables can identify the distressful signals of patients, which eases caregiver burden, enables earlier intervention, and improves quality of life for. However, unlocking its fullest potential will take real advances in AI engineering, edge computing, and responsible product design. But the outcome is worth pursuing.

Because sometimes, the most meaningful innovation isn't teaching machines to speak. It's teaching them to listen.

(The views expressed are personal)

This article is authored by Rajesh Subramanian, CEO and founder, embedUR.