Most of the conversation around artificial intelligence right now revolves around chatbots. Tools that write your emails, summarize long PDFs, or spit out blocks of code in seconds. That is the AI most people interact with daily.

Behind the scenes in health research labs, however, a profound shift is taking place that has nothing to do with words.

Scientists are training AI models to read raw human biology instead: optical heart rate signals, ECG readings, movement patterns, and sleep data. The goal is teaching AI to understand what is happening inside your body long before you ever feel sick.

During Samsung’s panel discussion on the future of Connected Health at the Pan Pacific London, Otavio Penatti, Samsung’s Head of Health AI R&D, captured this vision simply: “Your body speaks, Samsung Health translates.”

So what makes this different from your average fitness app?

Most health apps you already use run on basic rules. Your heart rate crosses a certain number while you’re resting, so the app sends an alert. That’s about as sophisticated as it gets.

These new models work on a different level entirely. Researchers feed them massive amounts of raw biometric data without labeling any of it first.

No doctor has to sit there and tag every second of heart rate or movement data by hand. The AI figures out the patterns on its own, essentially learning the basic language of how a human body functions.

Once a model has been trained on millions of hours of this kind of data, it can then be fine-tuned for a long list of specific medical uses.

And it’s not just watching for one odd spike in your heart rate. It’s cross-referencing your heart rate variability, skin temperature, movement, and sleep patterns all at once to build a detailed picture of what’s normal for you specifically.

One night of sleep data, over 130 diseases

Here’s where it gets interesting. Researchers at Stanford built a foundation model trained specifically on sleep data, using both raw polysomnography (the kind of detailed sleep study done in a lab) and data from wearables.

When they compared just 24 hours of sleep data against years of longitudinal health records, the model was able to flag onset risk for more than 130 different diseases. From one night of sleep.

Dr. Hon Pak, who leads Digital Health globally at Samsung, described where this is headed: “We are collecting data right now in which we have clear signals of things to come. We just don’t know what it is right now. But through the work that we’re doing, we’ll be able to share these insights early and in the right, meaningful way.”

The missing link: your wearable talking to your medical chart

Right now, your Apple Watch or Galaxy Watch has no idea what your last blood test looked like. Your wearable data and your actual medical records live in two completely separate worlds.

That’s the real opportunity here. If continuous biosignal data from your wearable gets linked with your actual health records, prescription history, and lab results, AI could start catching subtle shifts in your body that even your doctor might miss during a routine checkup.

Not because your doctor isn’t paying attention, but because a once-a-year appointment simply can’t compete with data collected every single day.

Instead of catching a heart attack or a metabolic crisis after it’s already happened, the idea is to catch the early warning signs years before that. It’s a shift from reacting to illness to catching it early enough that it might never become a crisis at all.

It’s worth noting that most of this is still in research stages. How it eventually shows up in the health apps on your phone will depend a lot on how these companies handle the data, and how accurate these predictions actually turn out to be in practice.

But the direction is clear enough: the next major leap in AI might not be about how well it writes. It might be about how well it listens to you.

Here’s a paradox worth sitting with: Wearable adoption keeps climbing. A lot of people now walk around with a sensor on their wrist, tracking heart rate, sleep, steps, sometimes ECG. And yet chronic disease rates haven’t moved.

That’s the problem Dr. Hon Pak, Samsung’s Head of Digital Health globally, opened with at a recent panel on the future of connected care. More data hasn’t meant better healthcare. His diagnosis: a first-mile, last-mile issue.

The first mile is that all this wearable data rarely makes it to a doctor in any usable form. The last mile is that even when a doctor says “lose weight, eat better, move more,” that advice tends to dissolve the moment real life takes over.

Picking up kids, caring for parents, getting through a Tuesday, it’s not that people don’t know what to do. It’s that knowing and doing are two different problems.

Samsung’s answer involves three things it says connected care needs to mature: meeting people where they actually are, building enough trust that people rely on the data, and using AI to turn that data into something a person can act on rather than just look at.

The Tuesday evening problem

Karthik Poriya, Head of Product at Samsung Food, framed the everyday failure point clearly. Picture it: long day, you’re tired, staring into the fridge. You’re not reaching for the healthiest option in that moment, you’re reaching for whatever’s fastest and most satisfying.

For years, nutrition tracking has focused on logging what already happened. Poriya’s team wants to intervene right at that decision point instead, pairing Samsung Health’s underlying data (BMI, antioxidant index, glycation markers) with Samsung Food’s recipe index of roughly 40,000 dishes mapped across 34 nutrients, so a health number turns into an actual dinner suggestion rather than another stat to check later.

Dr. Pak backed this with a small, concrete example from his own family: moving leftover cheesecake out of eye-level in the fridge and putting washed, sliced carrots there instead.

Kevin Duffy, CEO of connected fitness company iFit, made a similar point about exercise. The most effective tool for sticking to a fitness plan is a personal trainer, he said, citing roughly 80% higher adherence compared to going it alone.

The problem is cost: US$100 an hour or more in a major city puts that out of reach for most people. iFit’s bet is that AI can deliver something closer to a personal trainer’s precision and motivation at a price that isn’t elitist.

Duffy pointed to three reasons his company partners with Samsung specifically: reach (Samsung hardware is already in people’s homes, on their wrists and TVs), access to the biomarker data needed to build an accurate plan, and the fact that fitness goals don’t exist in isolation from sleep and nutrition data.

Trust has to come before any of this works

None of the behavior-change ambitions matter if people don’t trust the data or the company holding it. Rohit R. L., who heads Samsung’s Technology Innovation Lab in the UK, laid out how his team approaches that.

Privacy comes first: Samsung, he said, treats itself as a custodian of user data rather than an owner of it. On top of that sits clinical validation, meaning published, peer-reviewed evidence that a feature actually works in the real world, not just in a lab.

He gave a specific example: Samsung’s ECG and irregular heart rhythm notifications have been clinically validated and cleared by regulators to detect signs of atrial fibrillation. He was careful with the wording there, detect signs of, not diagnose, since that distinction matters both medically and legally.

A separate study on fall detection for elderly users turned up a more human finding. The detection algorithm worked well in testing, but in real life people don’t wear their watches around the clock, and falls don’t happen on a schedule.

Dr. Pak drew a useful comparison: nobody tells their MRI technician they’re going to fidget through the scan, but with a wearable, you don’t get to control how or when someone wears it. Real-world validation has to account for that.

On the infrastructure side, Rohit described a three-layer privacy approach: on-device protection through Samsung’s Knox security framework, GDPR-compliant data handling, and alignment with the European Health Data Space, including a decentralized clinical trial system where patient identity stays with the hospital rather than moving to Samsung’s side.

Where AI actually comes in

The most technically dense part of the discussion came from Otavio Penatti, who leads Samsung’s Health AI R&D team in Brazil.

His team works on foundation models, the same category of large-scale AI models trained on unlabeled data that underpin most of today’s popular AI systems. Trained once on a huge amount of sensor data, these models can then be fine-tuned for many specific health applications, which tends to make them more accurate than models built for a single narrow task.

Penatti’s team is training these models on wearable sensor data (PPG, ECG, accelerometer) with the goal of getting the model to, as he put it, understand the body’s own signals well enough to translate them into something useful. Longer term, he sees potential in correlating that sensor data with clinical records to flag disease risk before symptoms show up.

Dr. Pak added a striking data point from Stanford research: a foundation model trained on sleep study data was able to look at just a 24-hour window of sleep and predict risk across more than 130 different diseases.

His takeaway was that the signals for a lot of future health problems are already sitting in the data being collected today, we just don’t yet know how to read all of them.

He closed this part with a number worth sitting with. A recent American Medical Association survey found that 97% of physicians have looked at wearable data at some point. Only 15 to 16% actually use it in day-to-day practice.

The top reason cited wasn’t distrust of the data, it was that the data doesn’t plug into existing clinical workflows. That’s part of why Samsung acquired health data platform Xealth, to build a pipeline that gets wellness data into the systems doctors already use.

Five years out

Asked to look five years ahead, each panelist’s answer tracked closely to their own corner of the problem. Poriya described a proactive dietitian in your pocket, one that doesn’t wait to be asked.

Rohit pointed to connected care that clinicians actually trust and patients fully control. Penatti envisioned AI that continuously links everyday behavior to clinical outcomes, catching problems before they start. Duffy’s version was a constant personal wellness coach, sifting through the data noise to tell you exactly what to do next.

The throughline across all four answers is the same: less raw data, more action. Samsung’s panel made the case that the wearable industry has spent the last decade solving for collection, more sensors, more metrics, more dashboards, while the harder problem, turning that information into something a tired person actually does on a Tuesday night, has barely been touched.

Whether foundation models and better clinical integration actually close that gap is still an open question. But it’s clearly the one Samsung is choosing to spend its next five years on.

Smart glasses are in a weird place right now. While early adopters are celebrating the ability to record everything, a growing group of critics are decrying the loss of privacy. Formerly at full throttle, Apple, in response to these controversies, will reportedly delay its version of smart glasses to next year.

According to Bloomberg, Apple is heading back to the drawing board to further emphasize the unreleased wearable’s privacy protections, which includes “hardware and software-related privacy features not available on current products.” The report, however, doesn’t enumerate what these features are.

If anything, these features will likely prevent malicious recording. Most smart glasses today have a light that tells passersby that the device is actively recording video. However, users have discovered ways to tamper and disable this light, so their devices can start recording without alerting others. Apple will likely introduce better measure to prevent such tampering.

Additionally, it’s also possible that Apple introduces a version of the smart glasses without a camera attached.

Because of these new initiatives, the brand will delay the launch of the wearable to sometime next year. Previously, there was heavy speculation that the smart glasses will launch later this year.

On the bright side, there’s still substantial reason to believe that Apple is launching its first ever foldable device later this year.

SEE ALSO: New York becomes first state to ban smart glasses