As learning language models (LLMs) continue to advance, so do questions about how they can benefit society in areas such as the medical field. A recent study from the University of Cambridge's School of Clinical Medicine found that OpenAI's GPT-4 performed nearly as well in an ophthalmology assessment as experts in the field, the Financial Times first reported.

In the study, published in PLOS Digital Health, researchers tested the LLM, its predecessor GPT-3.5, Google's PaLM 2 and Meta's LLaMA with 87 multiple choice questions. Five expert ophthalmologists, three trainee ophthalmologists and two unspecialized junior doctors received the same mock exam. The questions came from a textbook for trialing trainees on everything from light sensitivity to lesions. The contents aren't publicly available, so the researchers believe LLMs couldn't have been trained on them previously. ChatGPT, equipped with GPT-4 or GPT-3.5, was given three chances to answer definitively or its response was marked as null. 

GPT-4 scored higher than the trainees and junior doctors, getting 60 of the 87 questions right. While this was significantly higher than the junior doctors' average of 37 correct answers, it just beat out the three trainees' average of 59.7. While one expert ophthalmologist only answered 56 questions accurately, the five had an average score of 66.4 right answers, beating the machine. PaLM 2 scored a 49, and GPT-3.5 scored a 42. LLaMa scored the lowest at 28, falling below the junior doctors. Notably, these trials occurred in mid-2023. 

While these results have potential benefits, there are also quite a few risks and concerns. Researchers noted that the study offered a limited number of questions, especially in certain categories, meaning the actual results might be varied. LLMs also have a tendency to "hallucinate" or make things up. That's one thing if its an irrelevant fact but claiming there's a cataract or cancer is another story. As is the case in many instances of LLM use, the systems also lack nuance, creating further opportunities for inaccuracy.

It’s tough to give a blanket recommendation for ergonomics, as the term most accurately translates to “what works best for your body” — and all bodies are different. But if you’ve landed here, chances are you’re looking for a keyboard that’s easier on your shoulders, neck and wrists. Traditional keyboards keep your arms close together and force you to splay your hands outward. After a while, that can feel straining. By shifting the orientation of the keys, ergonomic keyboards can keep your upper body in a more neutral position, preventing you from twisting or over-extending your hands and arms. We’ve broken down the various features that make some boards more body-friendly than others and tested a handful to come up with the best ergonomic keyboards to suit different needs and preferences.

What to look for in an ergonomic keyboard

Alice vs split

Most ergonomic keyboard layouts fall into two categories: Alice and split. The former is a single board with the two halves of the keys rotated about 30 degrees apart at the bottom. The separation forms an A-shaped space between the keys — which has nothing to do with why it’s called an Alice layout, it’s just a happy coincidence. This subtle tweak pushes your elbows away from your ribs while keeping a straight line from your forearm to your middle knuckle. Using one, I pretty instantly felt more open along the front side of my body. This layout more closely resembles a traditional keyboard, so it should be easier for most folks to get used to than a fully split option.

Speaking of, split boards break the keys into two separate parts you can position individually. You can put them shoulder distance apart, bring them closer together or angle them as much as feels comfortable. You can also put your mouse between the halves, which may feel like an easier trip for your cursor hand. Personally, I like being able to put my current snack between the two parts.

Tenkeyless

You can find ergonomic keyboards with and without number pads. Not having those number keys on the right side lets you keep your mouse closer in, minimizing overall reach. But if you work with numbers a lot, you’ll likely want that pad included. Some programmable boards allow for the use of layers, which temporarily repurpose keys and can provide you with a ten-key option through clever remapping of letter keys.

Tenting and negative tilt

Tenting raises the middle of the keyboard up, so your hands move closer to a “handshake” position. Alice keyboards usually angle up towards the middle and always to a fixed degree, since the two sides are connected. Split boards often let you adjust the degree of tenting, going from flat to subtle to extreme lift.

You may have encountered keyboards with an optional lift at the back of the board, raising the top keys higher than the space bar. Every set of hands is different, but for most people, pulling the backs of the hands towards the forearms increases strain. Negative tilt has the opposite effect by sloping in the other direction, lowering the top number keys while raising the edge with the spacebar. Many Alice and some split keyboards offer an optional negative tilt. I found it was more comfortable to enable that feature when I’m standing, and I preferred to have the keys flat when sat at my desk.

Staggered vs columnar

This decision seems to be one of the more hotly-contested among ergo enthusiasts. A standard keyboard has staggered keys, with each row slightly offset to the rows above and below it — so the A key is about halfway between the Q and W above it. This is a holdover from vintage mechanical typewriters, in which each press activated a hammer that smashed ink onto paper in the shape of a letter. To fit the hammers as close together as possible, while still allowing for finger pads, the keys were staggered.

Columnar or ortholinear keyboards stack the keys in orderly columns, often with rows that are not linear. Proponents claim this makes the keys easier to reach. Whether that’s true will be up to your fingers to decide, but I can say for certain that if you learned to type on a staggered keyboard, switching to a columnar layout is tough. It will take days, possibly weeks before you instinctively hit the C key. The N, M and B keys don’t fare much better.

Programmable keys

With a few exceptions, most ergonomic keyboards will work with PCs or Macs as a standard typing input, but the use of function and hot keys may require some remapping. It can be as easy as an onboard switch to toggle between Mac and PC layouts, or as involved as downloading software to change up the keys. Some boards even include (or let you buy) extra keycaps to change, say, the Mac’s Command and Option keys to PC’s Start and Alt buttons.

For some boards, remapping or programming keys is a crucial feature. Gaming peripherals have extra keys that you can set to execute a series of keystrokes with the push of a single button. Keyboards that work with layers, in which a single button can perform several functions, typically allow you to change what those are. Some ergonomic keyboards have non-standard layouts, like thumb clusters with multiple keys near the space bar that you operate with your thumb. You’ll also be able to program those.

Other considerations

Ergonomic keyboards come in mechanical, membrane, and scissor switch versions. Which works best for you is, again, up to your preference. I won’t get too deep into the particulars here, but the short of it is that membrane and scissor switches are less customizable than mechanical and typically cheaper. Typing on them tends to be quieter and softer. Mechanical switches are more customizable, offer a more responsive typing experience and are usually pricier.

You’ll also have the option of wired or wireless ergonomic boards. All other things being equal, wired models are less expensive. Competitive gamers who rely on split-second responses may prefer the zero-lag of wired keyboards. Wired models also never run out of battery life and have fewer connectivity issues. But wireless keyboards keep your desk less cluttered.

Some ergonomic keyboards come with permanent or removable wrist or palm rests, which can be cushioned or hard. This is another area where opinions diverge: proponents claim they help you maintain a neutral hand position, while detractors say they put pressure on the tendons in your wrist. Ideally, your palms should be resting, not your wrists, and you might find you like having that support or you may find the pressure uncomfortable. 

Photo by Amy Skorheim / Engadget

How we tested

All our guides begin with extensive research to figure out what’s out there and what’s worth testing. We consider brands with good reputations that we’ve heard good things about from colleagues and other trusted publications. For this guide, I looked for keyboards with ergonomic features like tenting, split keys, palm support and so on. I also zeroed in on boards that didn’t require a deep amount of familiarity with the vast and exhaustive world of custom keyboards.

Once I settled on ten boards, I acquired them and used each one for anywhere from a few days to a few weeks. I tried out the remapping and macros software and considered the comfort, design, price and durability of each model before arriving at picks I think will work best for the most people out there.

England’s National Health Service (NHS) said on Tuesday that “tens of thousands of children and adults” with type 1 diabetes will receive an “artificial pancreas” to help manage their insulin levels. The hybrid closed loop system — a sensor under the skin that sends wireless readings to an externally worn pump, which delivers insulin as needed — can help patients avoid the risks of type 1 diabetes without worrying about finger sticks or injections.

This isn’t the first device of its kind. Tandem makes similar insulin pumps in the US after it received FDA authorization in 2019. Gizmodo notes that another company called iLet got FDA approval for a similar device last year. Although the NHS hasn’t said which specific device(s) its program will use, what’s different here is the nation’s publicly funded health care system providing them for free rather than as an exclusive privilege for the well-to-do. (Sigh.)

The hybrid closed loop system starts with a sensor implanted beneath the skin, which continually monitors glucose levels at regular intervals. The sensor sends that data wirelessly to a pump, worn externally, which delivers the proper insulin dosage. The “hybrid” part of its name comes from the fact that some user input, including entering carb intake, is still required in the otherwise self-regulating system.

The government agency gave an ultra-precise figure of 269,095 people in England living with type 1 diabetes, highlighting how many folks could potentially benefit from the rollout. The NHS says local branches will begin identifying patients for the program starting on Tuesday.

“Diabetes is a tough and relentless condition, but these systems make a significant, life-changing difference — improving both the overall health and quality of life for people with diabetes,” Colette Marshall, chief executive of Diabetes UK, wrote in the NHS’s press release announcing the rollout. “This really is a landmark moment and we’ll be working with the NHS and others to ensure a fair rollout that reaches people as quickly as possible.”

Meditation is often touted as a mental cure-all, purported to help with stress, sleep, mood, focus and even certain medical conditions. I’ve been meditating most of my adult life. I’ve done silent retreats. I’ve been formally trained in various techniques. I’ve had someone in my contacts list who I referred to as a “guru.” So I feel I’m relatively qualified to give some bad news: Meditation won’t fix your life, despite what David Lynch says. However, there’s also some good news: Despite not actually being a cure-all for everything bad in the universe, meditation can certainly take the edge off.

This is where meditation apps can come into play. Of course, practicing mindfulness doesn’t require an app; people have been doing it for thousands of years, with nary a smartphone in sight. But mindfulness apps can be useful in a number of ways. They provide access to all kinds of guided meditations to suit different styles. Some even offer social connections, which can motivate you to keep up your practice via the magic of peer pressure. They are also particularly well-suited to beginners, with many of them offering a free trial. With all of this in mind, I downloaded some of the most popular meditation apps and set about sitting calmly on a comfortable chair to test them out. What follows is a comparison aimed at real people just looking to squeeze a bit more joy and relaxation out of daily life.

How we tested meditation apps

Every brain is different, so I did not rate these apps based on if they sync up with my preferred meditation style. First and foremost, I looked for apps that cater to various methods and those that offer guided meditations that go beyond what’s free on YouTube. All of the items on this list are available on both Android and iOS, so you won’t have to worry about something being only for iPhone owners.

Of course, there’s lots of free stuff out there, from podcasts and videos on YouTube to audio tracks on streaming services. You can even find guided breathing sessions on an Apple Watch or Fitbit, as well as meditations in Fitness+, Samsung Health or any number of workout video providers. For this guide, I focused on apps that stood out in some way. I liked apps with huge libraries of guided meditations and those that offer additional mindfulness activities, like yoga routines. I also looked for easy-to-use apps with well-designed layouts. You don’t want to start your meditation journey with a clunky app that actually increases anxiety.

The most important thing with meditation is to keep doing it, so I awarded points for clever gamification elements, simple social network integration and anything else that encourages repeat visits. Finally, I considered extra features that set an app apart from the glut of competitors out there. For example, some meditation apps offer novel ways to track your progress, access to yoga routines and a whole lot more.

At the end of the day, each of these apps has its strengths. But if installing an app or using a device is not how you prefer to meditate, you can always turn off your phone and find a quiet room or environment. For those of us who need a little help from a digital guru, though, here are our favorite apps for meditation.

Other meditation gear we tested

Brain-tracking wearables have been around for years, but there are some newer devices that have been tailor-made for meditators. These gadgets track the brain during meditations and offer real-time feedback. It’s a real boon for the data-obsessed, but also a real bank account drainer, with some gadgets costing thousands of dollars. I took two of the more-popular options for a spin to see what they’d make of my brain.

Sens.ai Neurofeedback System

Sens.ai is a weird contraption that not only claims to track brainwaves, but gives real-time feedback to “teach” people how to meditate and enter a flow state. The device involves a giant headset that’s stuffed with brainwave sensors that detect beta, alpha, theta and gamma waves, in addition to heart-rate sensors. It also comes with a truly bizarre companion gadget that uses light stimulation (transcranial photobiomodulation) to keep an eye on focus and attention levels. The whole thing is combined with an app that keeps track of dozens of data metrics and allows access to various guided meditations.

I’m as surprised as you to say that this thing appears to work, with some caveats. It’s uncanny how well it monitors the brain during meditations. If I got lost in a thought spiral about lasagna at six minutes in, sure enough, there would be a dip in analytics at the six-minute mark. It’s also fairly easy to use, despite a process that involves wetting a number of electrodes. As magical as the accurate brain-tracking seems to be, however, I wasn’t as keen on the actual training portion, which often involves staring at a screen throughout the entirety of the practice. It’s also not for the financial faint of heart, as the Sens.ai device costs $1,500.

NeoRhythm Omnipemf

NeoRhythm’s Omnipemf is another wearable to help people get into that ever-elusive flow state. It doesn’t track your brain, but rather floods it with electromagnetic fields at specific frequencies to make it more susceptible to meditation and focus. This is supposed to prime your brain for the meditative state and, in theory, make it easier to capture that zen. However, I didn’t get much from it, other than a placebo-esque buzzing in my head.

To use it, you simply pop on the wearable and go about your day. You aren’t tied to an app, so you can meditate in whatever way you like. There are multiple modes that go beyond meditation, as this thing is supposed to help with focus, pain relief and sleep. I’d wait for some peer-reviewed studies, however, before buying this.

The US Food and Drug Administration has approved the first continuous glucose monitor (CGM) people can buy without a prescription. Dexcom's Stelo Glucose Biosensor System has a sensor users are meant to insert into their upper arm, similar to the company's other CGMs that need a doctor's prescription for purchase. It pairs with a smartphone application that can show the user's blood glucose measurements and trends every 15 minutes. 

The company designed the device specifically for adults 18 and up who are not using insulin, such as those managing their diabetes with oral medications and non-diabetics making a conscious effort to control their sugar intake. It could be a great tool for people with insulin resistance, including individuals with PCOS and other metabolic issues that heighten their probability of developing diabetes in the future. In general, it could give users the insight to be able to better understand how the food they eat and the movements they make impact their overall health. 

While CGMs aren't anything new, they've become a wellness trend on social media last year, and even non-diabetics started using them. By clearing Stelo, the FDA is making the monitors more accessible than before. "CGMs can be a powerful tool to help monitor blood glucose," said Jeff Shuren, MD, director of the FDA's Center for Devices and Radiological Health. "Today's clearance expands access to these devices by allowing individuals to purchase a CGM without the involvement of a health care provide. Giving more individuals valuable information about their health, regardless of their access to a doctor or health insurance, is an important step forward in advancing health equity for U.S. patients."

Stelo will be available starting this summer. Each patch is meant to last for 15 days before users will need to replace it. Dexcom has yet to reveal how much it would cost, but it said Stelo will "provide an option for those who do not have insurance coverage for CGM."

Dexcom

There’s a reason smartwatches haven’t replaced clinically-validated gear when you visit the hospital — accuracy and reliability are paramount when the data informs medical procedures. Even so, researchers are looking for ways in which these devices can be meaningfully used in a clinical setting. One project in the UK has explored if a Garmin Venu 2 and dedicated companion app could be used to free up doctors and nurses, six minutes at a time.

The Six Minute Walk Test (6MWT) is used to diagnose and monitor a number of cardiovascular maladies. This includes conditions like Pulmonary Hypertension that, if left untreated, are eventually fatal. “[The test has been] a cornerstone of hospital practice and clinical trials for decades all around the world as [...] a marker of how well the heart and lungs are working,” explained project leader Dr. Joseph Newman. While a change in a blood test marker might be clinically relevant, Newman said “it’s probably more important to someone that they can walk to the shop and back.” The test requires a patient walk on a flat, hard surface for six minutes straight, which stresses the heart enough to measure its capacity. A professional tests the patient’s heart rate and blood oxygen levels at the start and end, and while it’s simple and reliable, "it’s not perfect,” according to Newman. “This is why we’ve looked to change it in two important ways," he said, "can we make it shorter [...] and digitize it for remote use?"

After all, six minutes is a lifetime in a clinical setting, and patients dislike having to schlep all the way to their hospital just to walk up and down a corridor. It’s why Newman and Lucy Robertson — both researchers at the Royal Papworth Hospital in Cambridge — began looking for ways to revolutionize the test. They wanted to see if the test could be shortened to a single minute, and also if it could be carried out by a patient at home using a Venu 2. The watch was connected to a secure and dedicated clinical trial platform built by Aparito – a Wrexham-based developer – for testing. This was then sent out to patients who were instructed to wear the watch and walk outdoors to complete their own tests. “They’re asked to walk on flat, even, dry, relatively straight roads rather than in laps or circuits,” said Dr. Newman, with patients walking at their own natural pace.

“We carried out a product appraisal early on in the research process and were open-minded as to the brand or model,” said Newman. “Garmin came out on top for a few reasons; we can access raw data as well as Garmin’s algorithmically-derived variables,” he said. Because the research was being funded by a charity, the British Heart Foundation, the watch had to offer good value for money. It helped that Garmin, because of its existing health research division, gave the team “confidence in the accuracy of the sensors,” not to mention the fact that Aparito feels that “the Garmin SDK is relatively easy to work with,” he said. But while Garmin is in use right now, there’s no reason this setup couldn’t eventually work with a number of other brands. “As long as the technology works, it’s accurate, reliable and patients accept it, then we’re not tied to any brand.”

There are several benefits in giving patients the ability to run the tests at home: it’s more representative of the demands of their actual life, and patients can retake the test at regular intervals, making it easier to track that person’s health over time. “We can see real value in providing patients with pulmonary hypertension with an app and smartwatch to monitor their progress,”said Dr. Newman. “It’s unlikely to ever fully replace the need for in-person hospital reviews, but it will likely reduce their frequency.”

The results of the study right now suggest cutting the test to one minute has no detrimental effect on its outcome or accuracy,and that patients are far more likely to run the test regularly if they’re able to do so at home. “It’s likely that the upfront costs of wearables [to a hospital] may be offset by the longer term reduction in hospital visits,” said Newman. If that turns out to be right, then it means clinicians can better focus their time and efforts where their expertise is more valuable.

After designating social media as a "public health hazard" in late January, New York City is now suing Meta, Google, Snap and TikTok for "fueling nationwide youth mental health crisis." Specifically, these companies face three counts in the lawsuit: public nuisance, negligence and gross negligence. The Mayor Eric Adams administration accuses TikTok, Instagram, Facebook, Snapchat and YouTube of "endangering our children's mental health, promoting addiction, and encouraging unsafe behavior."

These are allegedly achieved by way of harmful algorithms, gambling-like mechanisms and manipulation through reciprocity — making the user "feel compelled to respond to one positive action with another positive action." The city believes that there is a correlation between the increase in social media usage and the decline in local youth mental health over "more than a decade."

In response, Google and Meta told CNBC that they have always worked with youth safety experts and provided parental control tools. ByteDance's TikTok also highlighted some of its specific tools to Axios, namely age-restricted features, parental controls and an automatic 60-minute time limit for users under 18. However, none of the tech companies acknowledged the problematic features listed by the Adams administration.

This lawsuit follows a recent Senate hearing on online child safety, in which the CEOs of all the aforementioned tech companies (except Google) were present. In his opening remarks, Senator Lindsey Graham told the tech execs that "you have blood on your hands" — a reference to online child exploitations and cyberbullying that unfortunately led to deaths. 

Through this case, the Adams administration wants these tech companies to pay up for the city's youth mental health services, which apparently cost more than $100 million each year. But ultimately, it's about forcing these tech giants to stop manipulating young users into addictive behavior, as well as getting policymakers to place new federal laws that safeguard youth mental health on social platforms.

Before this New York City lawsuit, Meta already faces a similar case from 41 states back in October 2023, in which it was accused of misleading the public about the safety of its platform's "addictive" features. Meta, Snap, TikTok and Google were also sued in a multi-district litigation in 2022 for their addictive features that allegedly cause "emotional and physical harms, including death" to adolescents.

The “P” in HIPAA doesn’t stand for privacy. It’s one of the first things a lot of experts will say when asked to clear up any misconceptions about the health data law. Instead, it stands for portability — it’s called the Health Insurance Portability and Accountability Act —and describes how information can be transferred between providers. With misinterpretations of HIPAA starting with just its name, misunderstandings of what the law actually does greatly impact our ability to recognize how the kinds of data do and don't fall under its scope. That’s especially true as a growing number of consumer tech devices and services gather troves of information related to our health.

We often consider HIPAA a piece of consumer data privacy legislation because it did direct the Department of Health and Human Services to come up with certain security provisions, like breach notification regulations and a health privacy rule for protecting individually identifiable information. But when HIPAA went into effect in the 1990s, its primary aim was improving how providers worked with insurance companies. Put simply, “people think HIPAA covers more than it actually does,” said Daniel Solove, professor at George Washington University and CEO of privacy training firm TeachPrivacy.

HIPAA has two big restrictions in scope: a limited set of covered entities, and limited set of covered data, according to Cobun Zweifel-Keegan, DC managing director of the International Association of Privacy Professionals. Covered entities include healthcare providers like doctors and health plans like health insurance companies. The covered data refers to medical records and other individually identifiable health information used by those covered entities. Under HIPAA, your general practitioner can't sell data related to your vaccination status to an ad firm, but a fitness app (which wouldn't be a covered entity) that tracks your steps and heart rate (which aren't considered covered data) absolutely can.

“What HIPAA covers, is information that relates to health care or payment for health care, and sort of any piece of identifiable information that’s in that file,” Solove said. It doesn’t cover any health information shared with your employer or school, like if you turn in a sick note, but it does protect your doctor from sharing more details about your diagnosis if they call to verify.

A lot has changed in the nearly 30 years since HIPAA went into effect, though. The legislators behind HIPAA didn’t anticipate how much data we would be sharing about ourselves today, much of which can be considered personally identifiable. So, that information doesn’t fall under its scope. “When HIPAA was designed, nobody really anticipated what the world was going to look like,” Lee Tien, senior staff attorney at the Electronic Frontier Foundation said. It’s not badly designed, HIPAA just can’t keep up with the state we’re in today. “You're sharing data all the time with other people who are not doctors or who are not the insurance company,” said Tien.

Think of all the data collected about us on the daily that could provide insight into our health. Noom tracks your diet. Peloton knows your activity levels. Calm sees you when you’re sleeping. Medisafe knows your pill schedule. Betterhelp knows what mental health conditions you might have, and less than a year ago was banned by the FTC from disclosing that information to advertisers. The list goes on, and much of it can be used to sell dietary supplements or sleep aids or whatever else. “Health data could be almost limitless,” so if HIPAA didn’t have a limited scope of covered entities, the law would be limitless, too, Solove said.

Not to mention the amount of inferences that firms can make about our health based on other data. An infamous 2012 New York Times investigation detailed how just by someone’s online searches and purchases, Target can figure out that they’re pregnant. HIPAA may not protect your medical information from being viewed by law enforcement officers. Even without a warrant, cops can get your records just by saying that you’re a suspect (or victim) of a crime. Police have used pharmacies to gather medical data about suspects, but other types of data like location information can provide sensitive details, too. For example, it can show that you went to a specific clinic to receive care. Because of these inferences, laws like HIPAA won’t necessarily stop law enforcement from prosecuting someone based on their healthcare decision.

Today, state-specific laws crop up across the US to help target some of the health data privacy gaps that HIPAA doesn’t cover. This means going beyond just medical files and healthcare providers to encompass more of people’s health data footprint. It varies between states, like in California which provides options to charge anyone who negligently discloses medical information or some additional breach protections for consumers based in Pennsylvania, but Washington state recently passed a law specifically targeting HIPAA’s gaps.

Washington State’s My Health My Data Act, passed last year, aims to “protect personal health data that falls outside the ambit of the Health Insurance Portability and Accountability Act,” according to a press release from Washington’s Office of the Attorney General. Any entity that conducts business in the state of Washington and deals with personal information that identifies a consumer’s past, present or future physical or mental health status must comply with the act’s privacy protections. Those provisions include the right not to have your health data sold without your permission and having health data deleted via written request. Under this law, unlike HIPAA, an app tracking someone’s drug dosage and schedule or the inferences made by Target about pregnancy would be covered.

My Health My Data is still rolling out, so we’ll have to wait and see how the law impacts national health data privacy protections. Still, it’s already sparking copycat laws in states like Vermont.

The Food and Drug Administration has given the green light to a sleep apnea detection feature on Galaxy Watch devices in the US, Samsung has revealed. The company notes this is the first approval of its kind in the US — South Korean officials previously rubberstamped the feature for use in that country.

Samsung plans to add the sleep apnea monitoring tool to compatible Galaxy Watch wearables in the third quarter of this year. It will be available via the Samsung Health Monitor app.

The feature allows those aged 22 and older who have not been diagnosed with the condition to check for signs of sleep apnea using their smartwatch and phone. It looks for signs of moderate to severe obstructive sleep apnea (OSA) over a two-night monitoring period. Users will need to track their sleep for more than four hours twice over a ten-day period to use the feature.

OSA is a common, chronic condition that affects around 25 percent of men and a tenth of women in the US, according to the National Sleep Foundation. Those with the condition tend to stop breathing while they sleep, which can reduce their sleep quality, disrupt oxygen supply and lead to more daytime tiredness. Left untreated, "sleep apnea can compound the risk of cardiovascular diseases such as hypertension, coronary artery disease, heart failure, cardiac arrhythmia and stroke," Samsung notes. The company added that the feature should help more people detect moderate and severe forms of the condition, and for them to seek medical advice when they do.

Other versions of the feature, which ties into Samsung's other efforts to help its customers have a good night's sleep, have popped up in devices elsewhere. In 2020, Withings added a sleep apnea detection feature to its sleep tracking mat.

If there’s one thing we can all agree upon, it’s that the 21st century’s captains of industry are trying to shoehorn AI into every corner of our world. But for all of the ways in which AI will be shoved into our faces and not prove very successful, it might actually have at least one useful purpose. For instance, by dramatically speeding up the often decades-long process of designing, finding and testing new drugs.

Risk mitigation isn’t a sexy notion but it’s worth understanding how common it is for a new drug project to fail. To set the scene, consider that each drug project takes between three and five years to form a hypothesis strong enough to start tests in a laboratory. A 2022 study from Professor Duxin Sun found that 90 percent of clinical drug development fails, with each project costing more than $2 billion. And that number doesn’t even include compounds found to be unworkable at the preclinical stage. Put simply, every successful drug has to prop up at least $18 billion waste generated by its unsuccessful siblings, which all but guarantees that less lucrative cures for rarer conditions aren’t given as much focus as they may need.

Dr. Nicola Richmond is VP of AI at Benevolent, a biotech company using AI in its drug discovery process. She explained the classical system tasks researchers to find, for example, a misbehaving protein – the cause of disease – and then find a molecule that could make it behave. Once they've found one, they need to get that molecule into a form a patient can take, and then test if it’s both safe and effective. The journey to clinical trials on a living human patient takes years, and it’s often only then researchers find out that what worked in theory does not work in practice.

The current process takes “more than a decade and multiple billions of dollars of research investment for every drug approved,” said Dr. Chris Gibson, co-founder of Recursion, another company in the AI drug discovery space. He says AI’s great skill may be to dodge the misses and help avoid researchers spending too long running down blind alleys. A software platform that can churn through hundreds of options at a time can, in Gibson’s words, “fail faster and earlier so you can move on to other targets.”

CellProfiler / Carpenter-Singh laboratory at the Broad Institute

Dr. Anne E. Carpenter is the founder of the Carpenter-Singh laboratory at the Broad Institute of MIT and Harvard. She has spent more than a decade developing techniques in Cell Painting, a way to highlight elements in cells, with dyes, to make them readable by a computer. She is also the co-developer of Cell Profiler, a platform enabling researchers to use AI to scrub through vast troves of images of those dyed cells. Combined, this work makes it easy for a machine to see how cells change when they are impacted by the presence of disease or a treatment. And by looking at every part of the cell holistically – a discipline known as “omics” – there are greater opportunities for making the sort of connections that AI systems excel at.

Using pictures as a way of identifying potential cures seems a little left-field, since how things look don’t always represent how things actually are, right? Carpenter said humans have always made subconscious assumptions about medical status from sight alone. She explained most people may conclude someone may have a chromosomal issue just by looking at their face. And professional clinicians can identify a number of disorders by sight alone purely as a consequence of their experience. She added that if you took a picture of everyone’s face in a given population, a computer would be able to identify patterns and sort them based on common features.

This logic applies to the pictures of cells, where it’s possible for a digital pathologist to compare images from healthy and diseased samples. If a human can do it, then it should be faster and easier to employ a computer to spot these differences in scale so long as it’s accurate. “You allow this data to self-assemble into groups and now [you’re] starting to see patterns,” she explained, “when we treat [cells] with 100,000 different compounds, one by one, we can say ‘here’s two chemicals that look really similar to each other.’” And this looking really similar to each other isn’t just coincidence, but seems to be indicative of how they behave.

In one example, Carpenter cited that two different compounds could produce similar effects in a cell, and by extension could be used to treat the same condition. If so, then it may be that one of the two – which may not have been intended for this purpose – has fewer harmful side effects. Then there’s the potential benefit of being able to identify something that we didn’t know was affected by disease. “It allows us to say, ‘hey, there’s this cluster of six genes, five of which are really well known to be part of this pathway, but the sixth one, we didn’t know what it did, but now we have a strong clue it’s involved in the same biological process.” “Maybe those other five genes, for whatever reason, aren’t great direct targets themselves, maybe the chemicals don’t bind,” she said, “but the sixth one [could be] really great for that.”

FatCamera via Getty Images

In this context, the startups using AI in their drug discovery processes are hoping that they can find the diamonds hiding in plain sight. Dr. Richmond said that Benevolent’s approach is for the team to pick a disease of interest and then formulate a biological question around it. So, at the start of one project, the team might wonder if there are ways to treat ALS by enhancing, or fixing, the way a cell’s own housekeeping system works. (To be clear, this is a purely hypothetical example supplied by Dr. Richmond.)

That question is then run through Benevolent’s AI models, which pull together data from a wide variety of sources. They then produce a ranked list of potential answers to the question, which can include novel compounds, or existing drugs that could be adapted to suit. The data then goes to a researcher, who can examine what, if any, weight to give to its findings. Dr. Richmond added that the model has to provide evidence from existing literature or sources to support its findings even if its picks are out of left-field. And that, at all times, a human has the final say on what of its results should be pursued and how vigorously.

It’s a similar situation at Recursion, with Dr. Gibson claiming that its model is now capable of predicting “how any drug will interact with any disease without having to physically test it.” The model has now formed around three trillion predictions connecting potential problems to their potential solutions based on the data it has already absorbed and simulated. Gibson said that the process at the company now resembles a web search: Researchers sit down at a terminal, “type in a gene associated with breast cancer and [the system] populates all the other genes and compounds that [it believes are] related.”

“What gets exciting,” said Dr. Gibson, “is when [we] see a gene nobody has ever heard of in the list, which feels like novel biology because the world has no idea it exists.” Once a target has been identified and the findings checked by a human, the data will be passed to Recursion’s in-house scientific laboratory. Here, researchers will run initial experiments to see if what was found in the simulation can be replicated in the real world. Dr. Gibson said that Recursion’s wet lab, which uses large-scale automation, is capable of running more than two million experiments in a working week.

“About six weeks later, with very little human intervention, we’ll get the results,” said Dr. Gibson and, if successful, it’s then the team will “really start investing.” Because, until this point, the short period of validation work has cost the company “very little money and time to get.” The promise is that, rather than a three-year preclinical phase, that whole process can be crunched down to a few database searches, some oversight and then a few weeks of ex vivo testing to confirm if the system’s hunches are worth making a real effort to interrogate. Dr. Gibson said that it believes it has taken a “year’s worth of animal model work and [compressed] it, in many cases, to two months.”

Of course, there is not yet a concrete success story, no wonder cure that any company in this space can point to as a validation of the approach. But Recursion can cite one real-world example of how close its platform came to matching the success of a critical study. In April 2020, Recursion ran the COVID-19 sequence through its system to look at potential treatments. It examined both FDA-approved drugs and candidates in late-stage clinical trials. The system produced a list of nine potential candidates which would need further analysis, eight of which it would later be proved to be correct. It also said that Hydroxychloroquine and Ivermectin, both much-ballyhooed in the earliest days of the pandemic, would flop.

And there are AI-informed drugs that are currently undergoing real-world clinical trials right now. Recursion is pointing to five projects currently finishing their stage one (tests in healthy patients), or entering stage two (trials in people with the rare diseases in question) clinical testing right now. Benevolent has started a stage one trial of BEN-8744, a treatment for ulcerative colitis that may help with other inflammatory bowel disorders. And BEN-8744 is targeting an inhibitor that has no prior associations in the existing research which, if successful, will add weight to the idea that AIs can spot the connections humans have missed. Of course, we can’t make any conclusions until at least early next year when the results of those initial tests will be released.

Yuichiro Chino via Getty Images

There are plenty of unanswered questions, including how much we should rely upon AI as the sole arbiter of the drug discovery pipeline. There are also questions around the quality of the training data and the biases in the wider sources more generally. Dr. Richmond highlighted the issues around biases in genetic data sources both in terms of the homogeneity of cell cultures and how those tests are carried out. Similarly, Dr. Carpenter said the results of her most recent project, the publicly available JUMP-Cell Painting project, were based on cells from a single participant. “We picked it with good reason, but it’s still one human and one cell type from that one human.” In an ideal world, she’d have a far broader range of participants and cell types, but the issues right now center on funding and time, or more appropriately, their absence.

But, for now, all we can do is await the results of these early trials and hope that they bear fruit. Like every other potential application of AI, its value will rest largely in its ability to improve the quality of the work – or, more likely, improve the bottom line for the business in question. If AI can make the savings attractive enough, however, then maybe those diseases which are not likely to make back the investment demands under the current system may stand a chance. It could all collapse in a puff of hype, or it may offer real hope to families struggling for help while dealing with a rare disorder.