MOLLY WOOD: In the present day I’m speaking to Peter Lee, President of Microsoft Analysis, about what enterprise leaders throughout industries can study from the way in which that AI is remodeling medication and life sciences. He delivers a report from the entrance traces on the technological improvements which might be remodeling each side of drugs, from analysis to prognosis to safety and privateness, and even the basic method that medical doctors and sufferers talk with one another. AI improvements are serving to to evolve a healthcare system that’s much less siloed, much less complicated, extra thorough, extra environment friendly, safer, and much more empathetic. And if comparable transformations aren’t occurring in your trade but, relaxation assured, they are going to be quickly. Right here’s my dialog with Peter.
[Music]
MOLLY WOOD: Let’s begin along with your shift three and a half years in the past, when Microsoft CEO Satya Nadella requested you to rethink the corporate’s healthcare technique. I need to ask you when AI type of entered and have become a significant focus of what was already a fairly large technique shift into healthcare, proper?
PETER LEE: Proper. Satya first requested me to take a brand new take a look at healthcare method again in 2016, and I used to be truly fairly confused by that. I used to be questioning, why is he punishing me? [Laughter]
MOLLY WOOD: It’s not thought-about like a enjoyable subject to attempt to remodel.
PETER LEE: It isn’t, however I feel Satya actually noticed the longer term and was understanding, you already know, Microsoft is in actually each single healthcare group on the planet. Every little thing from Kaiser Permanente and the UnitedHealth Group, all the way in which to a one-nurse clinic in Nairobi, Kenya, and the whole lot in between. You realize, his level was, the longer term goes to be quite a bit about AI and concerning the cloud and about well being information, and are we doing sufficient there? And in order that was the task. I joked that it was somewhat bit like him dropping me and a few of my group into the center of the Pacific Ocean and asking us to search out land, since you simply don’t know which option to swim. It took somewhat little bit of time to type of perceive, what’s it about Microsoft that provides us a proper to take part right here? What are the differentiated new issues that we might supply? And the way in which that we ask that query is, If Microsoft had been to vanish right now, in what methods would the world of healthcare be harmed or held again? When ChatGPT was launched in November of 2022, three days after the discharge I bought emails from some clinician buddies of mine all over the world saying, wow, Peter, that is nice stuff. And I’m utilizing it in my clinic to do such and such a factor.
MOLLY WOOD: Instantly.
PETER LEE: Instantly. And so that actually motivated us to attempt to research and likewise educate the world of drugs as shortly as attainable, what this new know-how is.
MOLLY WOOD: I imply, healthcare is common. We’ve all interacted in a method or one other, and it may be actually private and emotional, but it surely may also be tremendous bureaucratic and complex. What’s the potential you see for AI to enhance the entire expertise?
PETER LEE: Nicely, I feel everybody who has contact with the healthcare system has moments of confusion and frustration. Should you dwell and work in the US, for instance, and you’ve got medical insurance out of your employer, let’s say, and also you get some therapy of some sort, just a few weeks later you’ll get one thing within the mail referred to as an Rationalization of Advantages kind, an EOB, and that’s completely mysterious. At the least for me, you already know, I take a look at these issues. I don’t know. Is that this a invoice? Um, you already know, what’s being defined right here? You’ve these bizarre codes, they’re referred to as CPT codes. You shouldn’t really feel dangerous about not with the ability to decode these issues as a result of I’ve truly interacted with fairly just a few C-suite executives in main American medical insurance corporations. And I’ve realized that they’ll’t even parse this stuff. And so a easy factor is once you get one thing like that, or perhaps you get lab take a look at outcomes from a bodily examination, you may present these issues to GPT-4 or to Microsoft Copilot, and simply say, check out this, clarify this to me. In order that’s actually empowering. Final yr, my father handed away after a protracted sickness. And it was a wrestle for me and my two sisters to take care of his care as a result of all of us lived a number of hundred miles away from my father. And there have been moments when the stresses of that might trigger the relationships between me and my two sisters to fray. And what I’ve realized over the previous few years is that so many individuals in our world undergo this. And so the power to present all of the lab take a look at outcomes, all of the notes, to GPT-4, clarify the scenario and clarify that we’re going to have a 15-minute cellphone dialog with Dr. Ok, after which simply ask the query, What could be the most effective two or three issues to ask? What’s the most effective use of this time? The power of that interplay to type of convey the temperature down and actually protect household concord and provides us a option to really feel empowered in interacting with a posh healthcare system is one thing that was very significant.
MOLLY WOOD: First, I’m so sorry to listen to about your father.
PETER LEE: Oh, thanks. It was actually his time and likewise, you already know, he handed peacefully and with household round, so all of that was nice.
MOLLY WOOD: I imply, these conditions are so attempting for households, and it’s actually profound to consider know-how serving to to make experiences like that somewhat bit simpler. It’s attention-grabbing how significant a rise in empathy might be in these conditions, and also you discovered that introducing AI into medication truly can introduce extra empathy. Was that shocking to you?
PETER LEE: You realize, as a techie, I used to be responsible of pondering, when you consider medication and healthcare, of instantly zooming in on AI. Prognosis. So a technologist, historically, when they consider healthcare, will assume, Oh, can we make an AI system take a look at radiological photographs? Can we get an AI system to cross the US medical licensing examination? All these issues are good and necessary, however there’s a lot extra to healthcare. An enormous a part of healthcare is the connection between the physician or nurse and the affected person. Simply a physician with the ability to preserve eye contact and be current with the affected person throughout an encounter as a substitute of typing at a laptop computer, it issues a complete lot. A health care provider being reminded by an AI system, oh, your affected person is about to make her very first journey ever to France subsequent month. Possibly it’s good to place an additional line in your e-mail to her to want her the most effective. These additional little human touches. And so there are two issues concerned in making that attainable. One is doing what I name reverse prompting. We all the time take into consideration the human being prompting the AI system after which the AI system reacting, however the AI system can oftentimes immediate the human. However the different is simply giving extra time to medical doctors, to nurses, making them extra productive. And so simply the help of an AI system that may say, take heed to the doctor-patient dialog and unload more often than not and labor concerned in, say, writing the medical encounter word. These items, they add up and so they actually matter quite a bit for that human connection between physician and affected person.
MOLLY WOOD: You stated one thing a bit counterintuitive, in a method, at a convention not too long ago about how that point that’s freed up that ought to enable medical doctors and nurses to do the work, you already know, not offload the technical work to AI, and that AI can, as you simply identified, truly be the extra empathetic communicator.
PETER LEE: Yeah, I’ve a colleague, he’s a neuroradiologist, Greg Moore, and he had a buddy, a vibrant, very profitable buddy, and she or he sadly bought identified with pancreatic most cancers. And utilizing Greg’s connections, he bought her into the specialist clinic, at Mayo Clinic, actually one of many high locations for that individual type of most cancers. And being the go-getter that she was, she was insisting on a cutting-edge immunotherapy. However these specialists, these are the easiest individuals on the planet in treating this sort of most cancers, had been lifeless sure that that was the unsuitable method, that they wanted to start out with a specific chemotherapy. The affected person was insistent in disagreeing, and so there was a battle that finally led the specialist to come back again to Greg and say, We’re having an issue interacting with this affected person, are you able to speak to her? Greg, not figuring out what to say to this positively determined affected person, consulted with GPT-4. GPT-4, curiously, got here to the identical conclusion because the specialist. They usually had this dialog, GPT-4 and Greg, on the right way to speak to the affected person. On the finish of that interplay, Greg, in a weirdness about AI right now, thanked GPT-4. And GPT-4 stated, you’re welcome, Greg, however let me ask, how are you doing? Are you holding up okay? And are you getting all of the assist that you simply want?
MOLLY WOOD: Whoa.
PETER LEE: Once more, it’s on this concept of reverse prompting that simply bought Greg to simply take a step again and replicate on his personal psychological state and on his personal psyche and skill to deal with the scenario of such a detailed buddy in such a determined scenario. That’s very excessive, however there are many smaller issues as properly. The most important producer of digital well being report methods is Epic, and Epic has been quickly integrating GPT-4 and GPT-3.5 into numerous purposes of their EHR system. They usually’ve been then working with educational medical facilities to do managed research to see if it really works properly, if it’s not making a lot of errors, affected person satisfaction, physician satisfaction, and so forth.
One of many issues that they’re discovering is that when GPT-4 writes the after-visit abstract e-mail to a affected person, the sufferers are persistently ranking these notes as extra human than the notes written by the medical doctors themselves.
MOLLY WOOD: Wow.
PETER LEE: And naturally, it’s not the case that they’re extra human. They’re written by a machine. However once you’re a busy physician, you won’t simply take the time to, say, congratulate your affected person on changing into a grandparent. These additional little touches, it simply exhibits that someone remembers and cares. It will possibly simply make a lot of a distinction within the connection between physician and affected person.
MOLLY WOOD: I imply, that’s fascinating and type of heartbreaking that AI clearly realized from the info it was educated on that empathy is a key a part of medication, however our medical professionals are so overtaxed that they’ll’t take the time to do it. I additionally love this type of reverse immediate concept, like AI as an assistant taking a number of the load off so medical professionals can get again to fundamentals, that are about care.
PETER LEE: Nicely, it’s such an necessary level as a result of proper now there may be this disaster within the US, however there have been quite a few research that present over 40 % of a clinician’s day, on common, is spent on clerical work, documentation, and note-taking. I actually love my major care doctor, however each time I see her, her again is turned to me. She’s sitting there at a pc, typing whereas she’s speaking to me. And the rationale she’s doing that’s she has a life. What I imply by that’s if she didn’t take the time to write down these notes throughout the encounter with me, she’d should take that work residence along with her. That’s referred to as, within the occupation, pajama time. Some medical doctors don’t need to do this whereas they’re with their sufferers and so they take that work residence and leap in mattress with a laptop computer and spend two hours doing that documentation and clerical work. And so what if AI might scale back that by half or by 80 %? A lot extra could be attainable.
MOLLY WOOD: You talk about this matter globally, and I’m interested by how your findings apply to medical doctors and nurses the world over. Is it simply within the US that we’ve, you already know, burnout and clerical hundreds which might be untenable? How do you discover that this know-how is translating to medical doctors in different elements of the world?
PETER LEE: It’s a world difficulty. Nonetheless, it’s price emphasizing simply how excessive the issue is in the US. Over the subsequent 5 years, there may be projected to be several-hundred-thousand-nurse scarcity within the US healthcare system. After which in the event you go to the UK, the Nationwide Well being Service, it’s not uncommon exterior of London to have a multi-month wait if it is advisable to see somebody for major care. There are enormous elements of Africa the place individuals nonetheless would possibly dwell a complete lifetime by no means seeing a physician. After which in China, the caseloads on major care physicians in China is now approaching 80 sufferers per day.
MOLLY WOOD: Whoa.
PETER LEE: For a single major care doctor. And the type of burnout and, in some instances, violence fueled by simply frustration that individuals have. It actually makes headline information in that nation. We even have one thing referred to as the “silver tsunami” that’s coming. There are demographic modifications the place the getting old inhabitants is reaching some extent the place there won’t be sufficient younger healthcare employees to take care of an getting old inhabitants. And so all of this stuff are about to actually grow to be excessive points. And all of that results in fewer and fewer vivid younger individuals desirous to enter into the occupation. Now, the US healthcare system is reacting—for instance, there’s a complete slew of recent medical colleges which have sprung up. Actually, I’m on the board of administrators of a brand new medical faculty, Kaiser Permanente College of Medication. However that’s simply considered one of a dozen new medical colleges which have sprung up within the US simply previously three years, in an try to provide extra medical doctors and nurses. The basic root trigger is, can we make being a physician, being a nurse, the type of satisfying occupation that permits individuals to attach with their private wishes to assist individuals versus do paperwork? Can we create that scenario that may inspire individuals? And that’s the most necessary downside for us as technologists to work on. Sure, it’ll be nice for us to unravel genomics with AI, to unravel most cancers with AI, to have higher radiological imaging methods with AI. All of that’s nice. However on the finish of the day, if the one factor that we are able to accomplish is to have AI make a dent in this type of workforce scarcity after which day-to-day employee satisfaction in healthcare, we’ll have actually accomplished the world an incredible service.
MOLLY WOOD: Healthcare is clearly such a novel trade and it presents its personal set of challenges. However you may think about that these are additionally classes that stretch into different industries. I’m wondering, in your learnings, what’s your message about the way in which that leaders throughout industries ought to implement AI on this option to convey extra time and doubtlessly extra empathy?
PETER LEE: That is going to sound humorous, however the way in which I clarify it’s that generative AI, that a big language mannequin, isn’t a pc. You possibly can substitute any kind of knowledge employee for this, however let’s think about you’re a nurse. Your psychological mannequin of a pc is a pc is a machine that does good calculation and has good reminiscence recall. So, in the event you ask a pc to come back up—let’s say you do an internet search, it’s going to provide you with exact solutions. Should you ask a pc to do some calculations, it’s going to provide you with a exact reply. The factor that’s odd about a big language mannequin is it’s much like the human mind in being very defective with reminiscence and really defective with calculation. And so, it’ll make errors. Should you ask it to do an enormous pile of arithmetic, it’ll get it unsuitable in methods similar to the way in which a human being would get it unsuitable. The factor that’s so necessary for individuals to appreciate is that that is now a brand new kind of machine, a brand new kind of software, that doesn’t have that good calculation or good reminiscence functionality. There’s a professor on the Wharton College at College of Pennsylvania, Ethan Mollick, who actually places it properly. He says it’s higher to think about a big language mannequin as an keen and tireless intern, and so in case you are a physician, it may be harmful to make use of the big language mannequin as if it’s a pc. It’s a lot better to deal with it like an intern. And the solutions you get from it, you need to assess and you need to take into consideration in the identical method as you’d out of your intern. And it’s excessive stakes, significantly on this planet of drugs. Should you don’t perceive this, you may find yourself hurting somebody. And so, as I’ve gone round to healthcare organizations all over the world over the previous yr, I all the time begin with that lesson.
MOLLY WOOD: Yeah, that may be a very completely different mindset. And really looks as if an necessary one for utilizing these instruments in any trade. So what’s your basic recommendation to leaders for the right way to use AI in a method that actually faucets into these strengths?
PETER LEE: The way in which to start out, after all, is to be very hands-on with these methods. And the best method for a human being to be hands-on is to do it by way of a chat interface. And you’ll simply speak to it. There’s one other stage the place, if in case you have a complete bunch of information, you may ask the system, Can you determine how finest to construction this information and put together it for evaluation and machine studying? That’s one other factor that’s rising in super significance. An ideal venture in Microsoft Analysis includes medical trials matching. So, proper now, when there are potential new therapies and new medicine, new diagnostic methods which might be proposed by medical researchers, they should undergo a validation course of. A part of the validation course of includes standing up what’s referred to as a medical trial to type of take a look at underneath circumstances, whether or not let’s say some new remedy is each secure and works properly. A tragic factor is that over half of medical trials which might be stood up fail to recruit sufficient contributors. And this holds again the development of medical science by enormous quantities. It’s actually a tragic factor. And a part of the issue is that once you take a look at medical trials paperwork, they’re extremely difficult issues to learn. They usually’re extremely unstructured textual content paperwork. What we’re studying is that a big language mannequin like GPT-4 can learn all these medical trials paperwork and put them in a structured database that permits instruments to raised match up sufferers with these trials. It simply opens up the chances that we’ll be capable to speed up the development of medical science by doing that. And so every considered one of these phases, you already know, the place you simply begin with the uncooked massive language mannequin, then you definitely give the big language mannequin entry to instruments, and then you definitely use the big language mannequin to make sense of all that information out on this planet. These three phases, I feel, is a pure development.
MOLLY WOOD: And once more, we must always say these phases are relevant to nearly any trade. It’s actually type of that mindset of fascinated by it and type of understanding what you need to undertake for and what you shouldn’t.
PETER LEE: Oh, yeah, completely. I imply, transportation, retail, manufacturing, legislation, finance, you title it. These identical concepts apply throughout the board.
MOLLY WOOD: While you hear reluctance to have interaction with a few of these instruments, what’s your type of go-to response?
PETER LEE: I simply attempt to present empathy. You realize, when people first confirmed what we now referred to as GPT-4 to me and defined to me what it might do, I used to be tremendous skeptical. Like, give me a break. After which I handed from skepticism to annoyance as a result of I noticed a few of my Microsoft Analysis colleagues getting what I felt was duped by these things. After which I bought type of upset as a result of it turned clear that my boss, Kevin Scott, and his boss, Satya Nadella, had been going to make an enormous wager on this know-how. So I believed, what? That is loopy. After which, with my very own private investigations, I bought into the section of amazement. As a result of it was true. These items that OpenAI was claiming about this factor had been truly true. They had been occurring. That led to a interval of depth the place you attempt to determine, okay, so what is that this going to imply? How can we use it? Then you definitely get right into a interval of concern since you begin to encounter issues like hallucination, points with bias, transparency, and so forth. And then you definitely notice this can be a actual know-how that’s going to alter the whole lot. And so I share my very own journey as a result of I’ve seen so many different individuals undergo the identical journey. And I’ve seen complete organizations and companies step by way of this stuff. And so what I inform individuals is, it is advisable to have endurance. Everybody must undergo this. And it is advisable to perceive this can be a course of that individuals should undergo as a result of it’s simply very difficult to consider that this know-how may even exist.
MOLLY WOOD: After which lastly, within the medical subject specifically, is there one thing, is there a moonshot that you simply assume you actually need this know-how to tackle?
PETER LEE: You realize, once I take into consideration what’s crucial factor to perform, there’s a idea in medication referred to as real-world proof, RWE. The dream there may be, what if each healthcare expertise that each affected person has might feed immediately into the development of medical information and science. And so right here’s my favourite instance from the pandemic. Within the first yr of the pandemic, some medical doctors all over the world had been randomly discovering that if they’d a really sick COVID affected person in respiratory misery that they might generally keep away from having to intubate that affected person by having the affected person keep susceptible for 12 hours, keep on their stomachs for 12 hours, and they might begin to share that information truly on social media. And so different medical doctors began to do the identical factor, but it surely was very random and advert hoc. A number of months later, a community of medical analysis establishments all over the world banded collectively and fashioned a medical trial, a medical research, to check this. And a yr and a half later, they decided that, sure, for some sufferers in extreme respiratory misery that this labored. That year-and-a-half hole is one thing that, first off, results in 1000’s of sufferers being intubated when perhaps they didn’t should be and a few of these sufferers dying needlessly. What if we had methods that might observe each single expertise in each single medical encounter that sufferers had? And that feeds in immediately into the storehouse of medical information. That’s the dream of real-world proof. And once I see what AI is changing into right now, I can not escape the sensation that some points of that dream of RWE are literally inside our grasp. And that’s the place I’d wish to see the world result in.
MOLLY WOOD: Peter Lee is President of Microsoft Analysis. Thanks a lot for the time. That is phenomenal.
PETER LEE: Thanks, Molly. It’s been nice to talk.
MOLLY WOOD: Should you’ve bought a query or a remark, please drop us an e-mail at worklab@microsoft.com. And take a look at Microsoft’s Work Pattern Indexes and the WorkLab digital publication, the place you’ll discover all of our episodes together with considerate tales that discover how enterprise leaders are thriving in right now’s new world of labor. You will discover all of it at microsoft.com/WorkLab. As for this podcast, please fee us, evaluate us, and comply with us wherever you hear. It helps us out a ton. The WorkLab podcast is a spot for specialists to share their insights and opinions. As college students of the way forward for work, Microsoft values inputs from a various set of voices. That stated, the opinions and findings of our company are their very own, and so they could not essentially replicate Microsoft’s personal analysis or positions. WorkLab is produced by Microsoft with Godfrey Dadich Companions and Affordable Quantity. I’m your host, Molly Wooden. Sharon Kallander and Matthew Duncan produced this podcast. Jessica Voelker is the WorkLab editor.
