Video: AI & The Augmented Age In Health Care
Kerrie Holley: 00:16 Hello and welcome. I'm Kerrie Holley, technical fellow here at Optum, working on artificial intelligence, and other emerging technologies. Thank you for joining me today at the Optum IQ Help and Focus event. We live in some pretty exciting times. Technology advances that were once consigned to science fiction are quickly becoming reality. And, the potential to improve our world is growing every single day.
Kerrie Holley: 00:39 We've been busy putting a lot of these advances to play, solving some of the most critical healthcare challenges. We use curated data, leading analytics, applied expertise to design various aspects forms of artificial intelligence to help automate, optimize processes, identify at-risk populations, and serve up actionable insights to leaders in organizations across the healthcare ecosystem.
Kerrie Holley: 01:07 That keeps us pretty busy, but every now and then, it helps to take a step back and really understand where the future is going, think about the technologies how they'll evolve, how they'll change healthcare, and today we've invited one of the foremost experts in artificial intelligence and robotics, to join us, to share its perspective on a changing world, a new era of human productivity, and creativity that is just beginning.
Kerrie Holley: 01:35 Maurice Conti is Chief Innovation Officer at Telefonica's Alpha, Europe's first moonshot factory. He and his team are responsible for proof of concepts, proof of technologies, prototypes. They go on to become full moonshots at Alpha projects that will change the world. Projects that will affect hundreds of millions of people be a force for good on the planet, and grow into impactful businesses for Telefonica. He's worked with big companies, start-ups, government agencies, artists, to explore the art of what's possible to explore things that matter to us in the future, and to build solutions to get us there.
Kerrie Holley: 02:13 Finally, he's been called a visionary, a leader for change, and a benevolent mad scientist. Ladies and gentlemen, please give a warm welcome to Maurice Conti.
Maurice Conti: 02:31 Morning. Um, I'm um, thrilled to be here. Um, I'm really looking forward to exploring a little bit of the future of technology um, and discussing how that uh, might impact healthcare. Let's start out by talking uh, a little bit about change. Um, the pace of change, how technology is going to shape our future in the next 10 or 20 years, um, and how that might change the way that we think about healthcare.
Maurice Conti: 02:58 Just how fast are things changing? Um, I think in the next 20 years, we're going to experience as much change to the way that we go about doing our work as we have in the last 2,000. If you think about it from a historical perspective, um, there've been four major eras that have defined the way that we work, right? Um, the first one was the hunter-gatherer age. It lasted several million years. Then came the agricultural age, it lasted a couple thousand years. Then, the industrial age lasted a few hundred years. The information age has lasted just a couple of decades, right? So you can see the acceleration. In fact, things are starting to move so quickly, that it's getting hard for us to envision the future.
Maurice Conti: 03:43 I don't work in Hollywood. Most of the time um, I work in AI, and um, with my teams I've built real things in the real world. Um, and so I can tell you that the immediate future is going to be a little bit different than what um, Hollywood has shown us. Uh, but my argument today is that we're heading into a future where our natural human capabilities are going to be radically augmented, and they're going to be radically augmented in three ways. We're going to have computational systems that help us think, robotic systems that help us make, and kind of a digital nervous system that's going to connect us to the world way beyond our natural senses.
Maurice Conti: 04:20 So let's start with cognitive augmentation. How many of you are augmented cyborgs? One, two, three, okay. Not bad, good crowd. Um, I would argue that we are all already augmented. Um, imagine if the holiday season you're at a party with your, your co-workers, and somebody asks you a question that you don't know the answer to. If you have one of these, you can know the answer. Almost instantaneously with very, very little effort, right? This is just a primitive beginning, actually. Uh, because even tools like Siri and Alexa have one very interesting thing in common with our very first tool, the stone hand ax. Because for three and a half million years ... It's a long time. Three and a half million years, our tools have been entirely passive. They do exactly what we tell them, and nothing more, right?
Maurice Conti: 05:12 So, our first tool only cut where we struck it, an artist chisel for instance only cuts where he points it. Even almost sophisticated tools though, are useless without our explicit direction. Um, and in generative design, the computer system basically uses the human's goals and constraints, and then does the rest. So, I'll give you an example. Uh, this is an ariel drone chassis. Um, in this project, the designer just told the computer that she was building a thing that was about this big, it had four propellers, and she needed a chassis that was as lightweight and aerodynamic as possible. And then the computer goes on and explores the entire solution space that meets those requirements. It actually knows how this thing is going to be made, this particular one was 3D printed. Um, and it explores millions of options, and then returns the very best uh, geometry.
Maurice Conti: 06:02 And it's returning things that humans would never have imagined, right? A human could not design something that, that looks like this. Uh, no human ever drew anything. Uh, the designer just told the computer this is what I want to make, these are my constraints, now go for it. By the way, it's no coincidence that the ariel drone chassis looks a lot like the pelvis of a flying squirrel, because these algorithms work the same way that evolution does, they just go a heck of a lot faster.
Maurice Conti: 06:28 So, we have computers that can generate new things based on our very well defined problems and constraints, but they still, they're still not intuitive. They don't really learn. They don't know what's going to happen next. Unlike Maggie. See, Maggie is smarter than our most advanced computational systems on the planet today. Why is that? She does know what's going to happen next. So, if her owner picks up that leash, Maggie knows with a high degree of certainty that it's time to go for a walk, right? Trivial for Maggie to do. Maggie even knows that it's time to go for a walk if she hears the words dog, and walk coming from the next room. In fact, she even knows it's time to go for a walk from her owner's body language um, without any words or any leash, right?
Maurice Conti: 07:11 Um, and so how does Maggie learn? She actually only has to do three things. She needs to pay attention, she needs to remember what happens, and then she needs to form and retain a pattern in her mind. Interestingly, this is exactly what computer scientists have been trying to do for the last 80 years in artificial intelligence. Yet, we're still nowhere near as smart as Maggie. But, things are happening really, really quickly.
Maurice Conti: 07:35 Back in 1952 they built this computer, and it could beat a human at tick-tack-toe. Very impressive. 45 years later, 1997, Deep Blue beats Kasparov at chess. This is a big deal. Seminal moment in AI. 14 years later, in 2011, Watson beats these two guys at Jeopardy. Now, Jeopardy is much harder for a computer to play than chess, because it can't work from predefined recipes. So, Watson had to use a kind of reasoning in order to beat the two humans. Then, a couple of years ago, DeepMind's AlphaGo beats a human at Go, which is the world's most difficult game. Go is not advanced chess. Okay, there are more possible moves in Go than there are atoms in the visible universe. So, uh, AlphaGo actually had to develop kind of an intuition, and in fact there were a couple of moments during that game where AlphaGo's programmers did not understand why it was doing what it was doing, why it was making the decisions it was making, which is interesting.
Maurice Conti: 08:41 But for me, the most exciting thing is the rate of change. So, in less than a human lifetime, we've gone from a child's game to the pinnacle of strategic thought. So what's next? Is it data? How many of you guys think we're going to get data in our lifetimes? So, an anthropomorphic robot that can walk around and do stuff. How many of you guys think? Lots going on. Things are going fast. Raise your hands up high, so I can see. Okay, you guys are all wrong. Um, I, I don't think this will happen in our lifetimes, and, and here's why.
Maurice Conti: 09:12 So first of all, I'm um, perplexed by our fascination on modeling artificial intelligence on human intelligence. This really um ... Especially for the near terms. The next 10 to 20 years. Um, first of all, who says human intelligence is the best intelligence, right? No seriously. There's lots of different kinds of intelligences, and why do we have to be so anthropocentric? Um, second of all, we don't really even understand how our own intelligence works. So in the last 18 months, two, maybe even three years, there've been some phenomenal breakthroughs in neuroscience, and in the deepening of the understanding of, of how our brains work that even with those, I'd say we understand less than, than 10 percent of the processes.
Maurice Conti: 09:56 So, it strikes me as odd, if not presumptuous, that we would stand up and say, "We are going to create a super intelligence in our own image like gods when we don't really even know how um, things work for ourselves. And the third reason I don't think we're going to get data uh, any time soon, is that strong AI, a general purpose artificial intelligence, is really, really hard to do. Um, so the folks I'm working with who are uh, working on this problem say that a general intelligence that can operate at the level of a B is still decades away, and they're working actively uh, on the problem. So what are we going to get? We're going to get squirrels. And we should be super excited about this. Um, so this squirrel is way smarter than I am. Um, this particular species actually goes out and collects acorns, um, and hides them all over his territory. Thousands of them. And then a year later, he can go back and find all of those acorns. You or I could never, ever do this. In fact, I couldn't find my phone in my hotel room this morning. Um, I'm a little bit embarrassed to admit how often I use find my iPhone when my phone is in my house. Um, so in this one narrow dimension, the squirrel has what is effectively a super power, um, compared to my own general intelligence.
Maurice Conti: 11:14 And for me, this is the vision for AI in the next 10 to 20 years. They know tools that are incredibly powerful and when deployed the right way can become extremely valuable. That, I think is the future of cognitive augmentation that's, that's relevant to us. So, we can augment our minds, but what about augmenting our bodies, with robotic systems? So, I'm sure you've heard there's been a lot of talk about robots taking people's jobs, um, it's actually nothing new. It's been happening for at least 60 years, more depending on how you count. And by the way, do you guys know the joke about the factory of the future? So it has two employees, a man and a dog. So the man's job is to feed the dog. The dog's job is to keep the man away from all the fancy machinery. And I think that's absolutely the wrong way to look at it. Because the successful companies in the future are going to be the ones that figure out how to put humans and robots together, to do things that neither can do on their own.
Maurice Conti: 12:11 This is the applied research lab that I founded when I was at Autodesk in San Francisco. And, one of the big areas of focus there is advanced robotics, and specifically human robot collaboration. This is Bishop, one of the robots uh, in the lab. And as an experiment, we set Bishop up to help a human working in construction, doing repetitive tasks. And here, the human is talking to the robot in natural language, the robot can talk back. There's some simple hand gestures. And, um, the robot just executes on, on instructions. So, we're using the human for what he's good at. General awareness, the human knows what's going on. Uh, navigating the construction site, and making decisions. That's what the human's good at, and the robot, we're using him for precision and repetition.
Maurice Conti: 12:54 And so together, they can do something that a human uh, by himself could never do. That partnership, that's the key for me uh, to the augmented age. So, we've got computational systems that are going to help us think. We have robots that are going to help us make, but what about connecting to the world around us? What about augmenting our own nervous systems? So, our human nervous system is pretty good. It tells us what's going on in our immediate surroundings, uh, in very high resolution, very fast. Um, it's really good. But the nervous system connecting us to our things, and our things to each other, I would say is rudimentary at best.
Maurice Conti: 13:32 So, a car doesn't tell the city's public works department that it just hit a pothole, it fit the mission. A building, for instance, does not tell the people that designed it whether or not the humans enjoy being inside of it, or are productive. And, a toy manufacturer, for instance, has no idea how their toys are being played with. When, where, are they even any fun? They would have no idea, because they don't have a nervous system connecting them to their product, or to their service. But this is changing, because we are developing these nervous systems right now. And they're connecting us to people, to data, and to things. Um, I would actually say this is probably one of the trends that's most interesting for the healthcare industry. Um, how many of you guys talk to AIs today? Yeah, you all do. Um, I, I'll tell you a story. I have a friend who's kids are a little bit younger than mine, they're seven and 10, and they have Alexa, uh, at home. And, every night, uh, at bedtime, the kids say goodnight to Alexa. And Alexa says goodnight back.
Maurice Conti: 14:37 Occasionally they forget, they get in bed, because little kids forget stuff. And, they realize that they forgotten to say goodnight to Alexa and even though they're tucked in and warm, they will get out of bed, run downstairs, and say goodnight to Alexa, and then go back to bed. I think this is a profoundly important evolution in human behavior. I think this is a big deal to see this happen.
Maurice Conti: 14:57 Um, there's a prediction that by let's say 2025, half of our conversations will be with non-humans. Um, so I think right now is the time to start thinking about how these new digital nervous systems are going to shape healthcare. I think the time um, for this particular set of, of technologies is actually right now. And, I'm super optimistic about what's coming. We're going to be able to achieve more than ever before as organizations, and individuals by partnering with these technologies, that are going to augment our ability to think, to make, and to connect to each other, and to the world around us.
Maurice Conti: 15:34 Mahatma Gandhi said, "The future depends on what you do today." And the future of healthcare is in the making right now. That's the opportunity in front of you. To embrace this technology, and to augment your organizations and the humans in them, so that they can go about building a future that we can all look forward to. I think that's the magic of the augmented age. Thank you very much.
Kerrie Holley: 15:57 Thank you very much Maurice.
Speaker 1: 16:03 So how do-
PART 1 OF 3 ENDS [00:16:04]
Kerrie Holley: 16:00 And this, okay. So, how do you see the augmented chain, the augmented age, effecting healthcare?
Maurice Conti: 16:09 So how is the augmented age gonna affect healthcare? I think that's a great question and effectively impossible to answer. Um, but seriously, uh, I think the technologies that make up the augmented age are something like, um, a rising tide. They effect everything. Um, and so asking that question is a little bit like asking if, back in the day, um, how is the printing press or the invention of, um, uh, written or spoken language going to affect the future? It's gonna affect it in every way. Um, you know this is gonna be foundational.
Maurice Conti: 16:43 Um, and so the interesting thing is, um, sure it's gonna affect, um, everything at a macro level. That's not a very useful answer. Um, interestingly, I think you have to immediately dive down, into the details. And that's kind of going back to this idea of, these are narrow, um, tools, narrow solutions to very specific problems.
Maurice Conti: 17:01 So there's kind of no middle discourse, in saying well you know, AI is going to generally. And it's like, yeah, generally it's gonna make things much more interesting. Um, but if you want more detail, you need to immediately dive down, to the specific example and say, well here's how AI is going to be applied to this particular problem. And that's making that jump quickly, I think is important. Um, can save you a lot of time and money. Um, and just you know, just getting down to brass tacks, um, without trying to kind of have a, uh, a broad general vision too far ahead of time.
Kerrie Holley: 17:28 Is ... Another question for you, I get asked this a lot, which is, uh, what is AI, how would you define AI?
Maurice Conti: 17:33 Um, I think there's two sides to that question. Um, the first side is, you know on the tech, on the tech part, um, we're starting to see what's, what's coming beyond, um, ML, uh, machine learning. And, um, uh, seeing these reasoning systems and sort of, things that are behaving a little bit more generally, a little bit more like strong AI, even though it's not, um, strong AI. So that's coming, um, machine learning, um, deep learning is gonna get more and more powerful. And you know we're already starting to see the ways it's being deployed. Um, you know get, get broader and broader, so that's interesting. Um, and, and there's lots out there around that.
Maurice Conti: 18:09 What is less talked about, is I think this next wave is going to have a human component to it. So, in those previous waves, you did see humans up on stage but to me it's sort of, more like window dressing than anything else. Um, the humans weren't really doing anything interesting. Um, and in this next phase, um, the, the output of these technologies is gonna be the result of the technology and a human, in a loop together. Um, it's not just the technology being deployed by itself, it's this cooperation that's gonna give us, um, some new things that are gonna be very interesting.
Kerrie Holley: 18:40 So let's, let's talk about this augmented age, a little bit more. And maybe two questions here, what is the augmented age and, and what, how do you see it affecting our lives, our health?
Maurice Conti: 18:53 So I mean, put simply the augmented age is just this idea of partnering humans and technology together. So the technology is augmenting humans, um, not just off doing its, its thing on its own. So together, humans and this technology, um, are, are greater, um, than, than the sum of the parts.
Maurice Conti: 19:09 How is this gonna affect healthcare? Um, you know again it's a, it's a, healthcare already being one of the broadest, most complex, uh, multidimensional, uh, industries on the planet. Um, that already makes it difficult to answer the question. And pairing a technology that it all, is also multidimensional, complex and, and very varied. Um, I think the answer is, you're gonna have a universe of, um, connection points between these two things. Um, that are gonna be very interesting. Some of the, um, sort of general qualities, uh, of, of this tech. So one is that, um, largely I think they're democratizing. They have a democratizing effect. Um, so for instance, I could see, um, you know class of technology being deployed that gives more people, better access to higher quality care, um, without necessarily increasing costs.
Maurice Conti: 20:00 So, uh, telemedicine for, um, you know communities that maybe are outside of the reach of high, high quality care. Um, suddenly because you've deployed this piece of technology you've, you've democrasized, democratized access to care that, that wasn't there before.
Maurice Conti: 20:16 So just this theme of democratization, um, carries through pretty well throughout this, this tech stack. Um, the other is this notion of, um, so right now we're pretty good at, um, and, and you hear the discourse you know, we're gonna automate the mundane stuff, so that the humans can focus on the more creative stuff. And I actually think, I sort of believe that about 70%. It depends on, on the application. But I think in healthcare this could be more true than not. Um, because healthcare maybe more than any other industry, um, there's a human component that is intrinsically important. Um, you know the touch, the, the, the empathy. Um, you know we're not gonna automate that for a very, very long time.
Maurice Conti: 20:59 Um, and, um, and so getting other things out of the way, so that you can put, um, the more human qualities at the forefront. Um, for the patient and even the practitioners, um, I, I think that's a big opportunity as well.
Kerrie Holley: 21:12 Excellent do you, do you see advanced AI or how do you see advanced AI changing the consumer, the patient experience?
Maurice Conti: 21:19 Well, I, I, I imagine that, um, you know the friction that exists in the system, um, no one likes, right? The, the, um, the providers, the administrators, uh, don't like it, it's expensive. It slows things down, difficult to deal with. Patients, um, probably don't like it. It, there's no direct connection between my headache and the paperwork I'm filing out. Um, so uh, moving that friction out of the system, which coincidentally something that these machine systems are, are pretty good at. Um, I think will result in, um, again a more human experience for the patient. And, uh, an experience that has just, just generally end-to-end, less friction.
Kerrie Holley: 21:58 No, I, I completely agree. I think we're gonna see this era of ambient intelligence, ambient computing. We're gonna see a more contextual experience with, uh, with, with healthcare, uh, in our homes, wherever we happen to be. We're gonna see it, as you say, be personalized, be localized. And a lot of that's gonna be because of the advances in a lot of technologies. And a lot of those technologies fueled by this, sort of general purpose technology that we call artificial intelligence.
Maurice Conti: 22:23 Yeah.
Kerrie Holley: 22:24 What do you think? Should people be excited about the potential of AI?
Maurice Conti: 22:28 Uh, well I mean I think they should be hugely excited. Um, (laughs) uh, it'd be hard for me to get out of bed in the morning if, if I wasn't so excited about it. Um, I, I think the question is, um, so here's why they should be excited about it. And maybe this is, this is where, um, it gets challenging for folks. Um, I think there's sort of three steps to this. Um, so one is, understanding the technology. And I don't mean reading the sort of pop, you know popular scientific, um, article that describes AI in a page and a half. Um, but really getting a sense of what the stuff can do. Like go visit the labs. Go, um, see the technology being deployed in adjacent industries or industries that are really totally orthogonal to yours, um, because there's a lot of cross, um, pollination opportunities. Um, and really get a sense.
Maurice Conti: 23:15 Um, I, I personally saw this, not happen, um, when you guys remember sort of seven or eight years ago, when 3D printing was all the rage. Um, so additive manufacturing, everyone is talking about additive manufacturing. And you saw all like, you saw startups and huge multinationals, um, come up with these projects where they were just gonna 3D print it. Whatever it was, they were just gonna 3D print it and that made it cool. Um, that got VC funding and so forth, which I thought was the more inane thing.
Maurice Conti: 23:42 Um, and, and you could tell like, all of these people had never even touched a 3D printer. They had no idea. So, um, you know what we did is we built a bunch of 3D printers from scratch. Some of the world's biggest, most sophisticated printers. And we were able to go back and say look, this is, this is how this technology is actually gonna be valuable. Um, and so understanding the technology is super important.
Maurice Conti: 24:02 The second part of that, which was also the second part for us is, understand the problem you're trying to solve. This is what people call it, design thinking. Um, but getting, getting a really sharp understanding of the problem you're trying to solve.
Maurice Conti: 24:14 And the third part is, making the right connection between those two things. So connecting the appropriate problem with the right technology. So not trying to 3D print houses, which is like the silliest thing. Because 3D printers print small, highly complex, expensive things, really well. Houses are not that.
Kerrie Holley: 24:31 Right.
Maurice Conti: 24:31 Um, so, um, you know, understand those two things and then the magic happens when you put those two together. And you find a problem that is important to the company. And can't be solved any other way but with this technology, that's magic. Because then that technology is unlocking something that didn't exist before because nobody could do it before.
Kerrie Holley: 24:49 Right.
Maurice Conti: 24:49 Until you have this technology, so that you know, that's where I would hone in.
Kerrie Holley: 24:53 Okay, you bring up some, uh, interesting points because obviously we're talking about artificial intelligence but I know you and I share this point of view that, you really can't solve a problem by using the words AI. You really have to get underneath.
Maurice Conti: 25:05 Exactly.
Kerrie Holley: 25:05 And, and really think about, uh, understand the specifics of what, as you said, understand the problem you're gonna solve if that's designed thinking whatever method of madness you choose to use.
Maurice Conti: 25:16 (laughs) Yeah.
Kerrie Holley: 25:17 Uh, but you've gotta understand, uh, is this something that reinforcement learning can help me with? Or is this something that I can, um, use deep learning or, or natural language processing with deep learning. We really have to get underneath, uh, the label of AI. That was one thing you said. A second thing you said was this sort of weak and strong AI.
Maurice Conti: 25:34 Mm-hmm (affirmative), mm-hmm (affirmative).
Kerrie Holley: 25:35 That we're, we're probably a bit closer on the weak side.
Maurice Conti: 25:38 Yeah.
Kerrie Holley: 25:38 Than on the strong side.
Maurice Conti: 25:39 And weak but sorry to interrupt.
Kerrie Holley: 25:40 Which is-
Maurice Conti: 25:41 But weak by the way, is not pejorative. Like these narrow AIs are incredible. There's, there's no reason to think less of them, poor guys. Um, and, and we call it weak, weak and strong but, um, that is not a pejorative label. Um, this is absolutely what's happening, the immediate future. It's great, um.
Kerrie Holley: 25:58 Well you bring up another point, this narrow versus general AI. Because the, the media would have us think and, and some tech companies would have us think that, I can build an AI system and it can do all of these general things. But the reality is, that we build because that's state of the art, we build narrow AI systems. So if I build a system to diagnose diabetes, it probably can't diagnose cancer. The data is different, the, um-
Maurice Conti: 26:25 It probably can't tell the difference between Type I and Type II.
Kerrie Holley: 26:27 That's true as well.
Maurice Conti: 26:28 Right.
Kerrie Holley: 26:28 Unless yeah, exactly. So that's the state of the art, uh, narrow versus general. Another question for you. Our OptumIQ team did a survey. Uh, we surveyed a bunch of folks, uh, industry leaders.
Maurice Conti: 26:40 Mm-hmm (affirmative).
Kerrie Holley: 26:41 And interesting enough, 25% said they had not put together any plans to implement AI. What do you think they need to do? Where do you suggest they begin?
Maurice Conti: 26:52 So, so I think-
Kerrie Holley: 26:53 What do you think about that?
Maurice Conti: 26:54 (laughs) So, so I think the other 75% are, are fine. Um, the 25% I would say, it sort of depends on, um, their, their vision of the future of their company. So if they wanna be in business in 10 years, um, I, I think they're in trouble. Um, I, I would be in a panic, uh, especially in healthcare, to be figuring out how, um, you know my company was gonna deploy these technologies. Um, they're gonna get left behind very quickly and at an accelerating rate.
Kerrie Holley: 27:24 Another, uh, interesting question for you. When you think about various, let's say AI models that have been developed independently, um, can they arrive at the same conclusions? You know we have different hospitals, different health plans. They you know, how will they difference them, differentiate themselves and, and will these models derive, derive at the same conclusions?
Maurice Conti: 27:43 Yeah and it, and, and maybe your question also goes to, um, commoditization of the algorithm. Right, so how do you, how do you maintain differentiation? Um, if, if I can quickly write an algorithm and get the same results you can.
Kerrie Holley: 27:57 Right.
Maurice Conti: 27:57 You know what's, what's the big deal? Um, and I think you know the, the proof in the pudding isn't in the deployment. Um, yes the, the systems are, are difficult to write, uh, an engineer. Um, but deployment is often the most difficult part. And so, even as they become, um, commoditized. And by the way like, a lot of this if, if you guys know, a lot of the development of the stuff that you, um, that you read about, I mean like deepfakes and you know whatever, whatever's, uh, on, on your feed. Um, a lot of that stuff is really just collecting libraries that are already out there. Um, and that, that people share. And stitching them together in, in a sort of, little bit of a custom way.
Maurice Conti: 28:36 So a lot of the work is already done and it's kind of assembling stuff. There's, there's the groundbreaking science but in, in the applied, um, world there's a lot of reuse. So this is already like, there aren't these silos of secret, secret sauce. Um, that's kind of not the way the industry is, is evolving.
Maurice Conti: 28:52 Um, and so it turns out what's really hard is, is deployment. Often, it's like the humans that screw it up. Because they you know, like oh they don't like this, they don't like that. Um, and so, um, technology is relatively easy to predict. It's very difficult to predict how the technology's going to, um, evolve and live when it, when it hits the ground you know.
Kerrie Holley: 29:10 It, it raises another question for me. We, um, we have a lot of talk in the industry that, I think many of you, uh, in the audience have seen around explainable AI.
Maurice Conti: 29:18 Mm-hmm (affirmative).
Kerrie Holley: 29:19 Uh, in our world here at Optum we have clinicians who, rightfully so say, "You know I can't use that model because, uh, it's a black box."
Maurice Conti: 29:28 Yeah.
Kerrie Holley: 29:28 And that's sort of, an interesting, um, dilemma because there's a bit of a mythology around ... I always use the, uh, the, uh, the scenario of a stethoscope. Most doctors don't know how a stethoscope works but they use it nonetheless. Uh, and of course, uh, many doctors don't know how these quote, "AI systems," work. But we're at, not at the same maturity. But the mythology that I'm speaking about is that, when we talk about explainable AI what we're really saying is that, the layman who didn't build the AI system, doesn't understand how it came at that conclusion and therefore, is uncomfortable. That's different then saying that, I as an engineer, can't explain what I do. So we can build interpretable interfaces to explain the model. And you used the example of deep go, which the engineers at first glance.
Maurice Conti: 30:21 Yep.
Kerrie Holley: 30:21 Don't know how it came up on the outcome. But if you ask them to explain it.
Maurice Conti: 30:25 Yeah, they can go back.
Kerrie Holley: 30:25 They can go back and figure out well, the reinforcement learning caused, they can explain it. So that's something we really need to clear up, that, that these black boxes because any black boxes in the world, the engineers who built it can take it apart and tell you how it, uh, how it puts together. The same is true of, of our AI systems. So.
Maurice Conti: 30:45 Yeah.
Kerrie Holley: 30:45 Something to.
Maurice Conti: 30:46 Yeah I, I agree, um, especially in terms of where we are now.
Kerrie Holley: 30:50 Yeah.
Maurice Conti: 30:50 So I think, um, you know we have, we have AIs that appear as black boxes that, um, that are explainable today.
Kerrie Holley: 30:57 Yes.
Maurice Conti: 30:57 Um, I think in the future and there's a bunch of people thinking about this. I think it's super important that we, um, are able to hold AIs accountable for the decisions they make. And as the AIs are becoming more sophisticated, that's gonna be harder and harder. So we will, in the next handful of years, start to have AIs that are opaque, um, unless we design them otherwise. Um, that, uh, you, you know we, we can't know exactly why it made a decision versus another one. So there's lots of people working on that problem right now is how, how to make AI accountable. Um, and then you know there's also the approach of, just put a human in the loop.
Kerrie Holley: 31:31 Right.
Maurice Conti: 31:32 So the AI might make a recommendation, um, it might, um, do a preliminary, um, diagnosis on a piece of medical imaging. Um, but then you have a radiologist that looks at that and go, okay I get your recommendation but I've also been talking to this patient for the last three years and I know this, this and that. And, uh, as a, as a result, I actually agree or disagree.
Maurice Conti: 31:52 If you have a human in the loop, um, and, and the decisions are not too complex, um, I, I think you can mitigate that for, for the next you know while.
Kerrie Holley: 32:01 Okay, awesome. Well, um, first of all-
PART 2 OF 3 ENDS [00:32:04]
Maurice Conti: 32:00 ... next, you know, while.
Kerrie Holley: 32:01 Okay, awesome. Well, um, first of all, if you guys are thinking this, we did not coordinate our outfits. Just uh-
Maurice Conti: 32:08 Hey, he has great taste (laughs).
Kerrie Holley: 32:10 And do does Maurice. Let's take some questions from the audience. Have we got a question here in the front row?
Speaker 2: 32:17 What advice do you have for aspiring future clinicians or engineers that want to enter the healthcare field?
Maurice Conti: 32:25 Okay, so the, the next generation. Um, and it's funny, I get asked um, similar questions a lot, especially from nervous parents of um, high schoolers who wanna give their, their kids unsolicited advice on what they should study. Um, so I think uh ... Well, look, um, generally speaking, there's gonna be lots of tech in the future. And so, any course of study that includes technology is probably better rather than worse. Um, I would, uh ... I mean, that's kind of general thing. It doesn't mean they all need to be computer scientists, but more tech is probably better than less tech. Um, but we're not all wired that way.
Maurice Conti: 33:03 Um, we're already seeing um, hyper specialization, so um, you know, I think um, folks being passionate about some narrow thing is probably going to fit well with the evolution um, of the space. What I haven't heard talked about as much, which I think is interesting is hybridization. Um, and I think, like, the super power um, individuals in, in the near future are gonna be hybrids. So, imagine um, a nurse, um, who has 20 years of experience dealing with patients; touching them, listening to them, um, but also has facility in some area of computer science. Um, that's an incredible tool.
Speaker 2: 33:46 Right.
Maurice Conti: 33:47 This is someone who, at the very least, would, would be able to deploy technological tools faster, better, deeper than um, someone who doesn't have this hybridization. Um, at the most interesting end of the spectrum, these are the kind of people who are gonna invent the tools of the future, invent the really disruptive tools that are very specific to healthcare. Um, because there's no way that a, that a software engineer who doesn't really know that much about medicine is gonna make the ultimate tool. It's gonna be a hybrid, whether it's inside of an individual or it's part of a team, um, that are gonna, that are gonna unlock the real potential.
Maurice Conti: 34:19 So, if I was, um, um, coming up, um, I, I would be really interested in, in, in investigating, um, different areas that are potentially complimentary, but feel, you know, bizarrely different. Um, and uh, and, and, you know, and think about how, how these two different ways of thinking could be brought to bear.
Kerrie Holley: 34:39 That's a great uh, a, a great answer. I also think that um, we're, we're seeing this today, that, that, that programming is becoming as important as reading and writing. You don't have to be a professional writer or a professional programmer, but we're beginning to see uh, a great many doctors who know how to program.
Maurice Conti: 34:55 Yeah.
Kerrie Holley: 34:55 I think uh, we'll see in the future, just as we're beginning to see with genomics, uh, that we need to actually educate doctors on the ... What are, what are genomics, the power of genomics, so they can actually do better patient care.
Maurice Conti: 35:08 Yeah.
Kerrie Holley: 35:08 Uh, we're also gonna, I think, see doctors who are, uh, gonna need to understand, uh, artificial intelligence at a various uh, level. And um, I also think that, uh, I forget the movie that had this quote, that math has become useful again. Not that it ever wasn't, but uh, uh, we're gonna see, uh ... You know, you're gonna brush off your skills in linear algebra.
Maurice Conti: 35:30 Right.
Kerrie Holley: 35:30 And uh, and, and doctors are going to uh, continue to do what they do, but they're gonna, they're gonna be even, uh, richer.
Maurice Conti: 35:37 Yeah, and, and, and I think the cool thing is that that barrier is actually coming down really fast. So, the technology is becoming more and more approachable to the layperson.
Kerrie Holley: 35:45 Exactly.
Maurice Conti: 35:46 Um, you know, you've got, uh, in genomics, like CRISPR-Cas9, you could sort of boot up a wet lab in a corner, um, of your building, and, and start fooling around and make glow in the dark plants. Middle school kids do this.
Kerrie Holley: 35:58 Exactly.
Maurice Conti: 35:59 Right?
Kerrie Holley: 35:59 Exactly.
Maurice Conti: 35:59 So, um, so, learning, um, messing around, not that, not that hard. Um, you ... Um, Unity, which is a game engine, so a lot of the um, video games out there are programmed in this environment called Unity. It's super easy to use, especially to do basic stuff. Um, like to do full immersive VR prototypes is almost trivial. Like, you don't need to be a programmer. It's, it's all visual drag and drop kind of stuff. Um, and so, you know, somebody in, in traditional healthcare has these tools that are quite approachable, and kind of fun and, and creative. Um, so I think that's, you know, the direction the future is going, but, but also the direction that the tools um, are, are going [crosstalk 00:36:36].
Kerrie Holley: 36:36 Think about 10 years from now. We've got all this augmented age, this ambient intelligence, ambient computing. We've got these palettes-
Maurice Conti: 36:44 Mm-hmm (affirmative).
Kerrie Holley: 36:44 ... that doctors can, can basically build their own AI.
Maurice Conti: 36:47 Right.
Kerrie Holley: 36:48 And uh, and but they're still a doctor, and they're augmenting their practice, they're uh, they're providing, uh, better patient care.
Kerrie Holley: 36:54 Additional questions from the audience?
Speaker 3: 36:57 Maurice, you have a lot of experience developing and nurturing new concepts. Can you share any lessons learned when it comes to building a solid business case when it comes to investing in artificial intelligence?
Maurice Conti: 37:07 Interesting. I suppose it depends a lot on who your investors are and what's on their mind. Um, what might make it easier, uh, today is the general climate. Like, I don't know that there's a lot of people that would argue with you, um, uh, if you were making the point, like, "This is a valuable technology and it's not that hard to see, um, the potentially positive impact."
Maurice Conti: 37:27 Um, again, I'm kind of back to this, um, you know, narrowness thing. Um, which, which makes the, the argument to an investor or, or, um, leadership, um ... Because you need to get to specifics very quickly. If you're talking general terms, you're, you're not talking about anything. Um, and especially in this, in this context.
Maurice Conti: 37:46 And so, the, the need to get to specifics gives you a specific discourse, which generally, at least in my experience, I've found um, investors react very positively.
Kerrie Holley: 37:54 Okay, awesome. Other, uh, questions from the audience?
Speaker 4: 37:59 Hi. Uh, which roles in healthcare do you see evolving, and how do you see them evolving as AI and robotics free up peoples' time to spend elsewhere?
Maurice Conti: 38:11 I, I mean, I think this goes back to the first question, um, around what's, what's coming and what, what should the future generation focus on. Um, again, I think there's going to be specialization, um, partly fueled by the evolution of the, um, of the industry, but partly fueled by the, um, delivery of these technologies. Um, you know, it's ... As the technology and the knowledge piles on, there's like more stuff to know. Um, and one person can't know all that stuff. Um, and so, you, you just sort of naturally have this, this specialization that needs to, needs to happen, because, you know, just, just addressing this narrow part of, of the industry fills up my head, I can't do anymore, right?
Maurice Conti: 38:52 And that, that section is gonna get smaller and smaller. It's actually ... It doesn't get smaller, but there, there's more of them. And so, um, specialization will need to come out. And then the other one is hybridization, whether it's, um, human machine or um, human machine and other humans, so ... As part of a team. Um, so for instance, um, you know, highly productive teamwork in, um, complex situations and complex problem solving, I think that is a, um, super important skill, um, for the future. And so, um, I think a lot of folks' jobs are gonna evolve, um, to a stage where, um, that is key in, in their success, is, is, um, relying on and collaborating with, um, other folks in order to achieve their goals, because these goals are beyond what, you know, a single person or even a small team can, can achieve.
Kerrie Holley: 39:40 And we may see the emergence of a digital doctor, and AI doctor, just like an anesthesiologist or a radiologist. We may start seeing more specialization where the blend of multiple fields come together. And maybe, uh, both of those parties have to be involved in patient care, or involved in the operating room. Uh, we'll see.
Speaker 5: 39:59 Um, AI has made amazing leap frogs in algorithms and deep learning in recent years, uh, but knowledge representation has really been lagging. Uh, we still have a lot of people storing data in spreadsheets and disconnected siloed databases. Uh, how can AI help us, uh, move out of a, a backward knowledge representation into a more knowledge, uh, representation that is actionable for our AI agents?
Maurice Conti: 40:23 It's interesting, you know, it depends on your lens. Like, when you zoom in and out in time ... Like, if you zoom out a little, like, it feels frustrating now. But if you zoom out and you look back on this period of 50 years, it will look like it happened really fast. We're in the middle of it, and so it's frustrating. You're like, "Wait a minute, somebody just sent me a fax? Like, really?" Um, uh, but you know, as, as we zoom out, these will look like, um, you know, step, step function and changes.
Maurice Conti: 40:45 Uh, you know, in terms of advice, I think, um, part of it commitment. So, I'd be, I'd be curious to hear where is the friction coming ... Like, who thinks it's a bad idea? Who is, who is rooting for the faxes and the spreadsheets? Um, and I would really wanna understand them, um, and, and their process. Um, and then again, look at, look at the body of technology and say, "Okay, how can we fix this very specifically and very effectively?" Um, rather than say, "Well, this ... That's silly. Excel, that doesn't make any sense. We should just apply some AI to that." Like, that doesn't actually make any sense. Um, what kind of AI? Why, why are you applying it? What part of the process are you trying to unlock?
Maurice Conti: 41:23 Maybe, maybe Excel has nothing to do with it. maybe it's something else going on here earlier in the process that makes Excel just, just not relevant, right? Or a fax. Um, if your ... If you ... If your, um, touchpoint with a consumer is an app that's always on, um, the ... Like, the fax never happens, um, anymore, because they're, they're constantly, uh, interacting with you through, through a whole new interface.
Maurice Conti: 41:45 Um, so it's a little bit hard to answer the question without knowing the details. And so, I would say the, the, the magic is in defining those details and understanding them deeply.
Kerrie Holley: 41:52 And you ... If ... As you know, when you look at any technology, look at the automobile. We can, uh, find pictures in the 19th Century where we saw a horse and buggy on one corner and automobile on the next. It takes time. Uh, not that we can't accelerate it, but uh, change is, uh ... You know, takes time to-
Maurice Conti: 42:08 Right, it takes time. Well, we had electric cars back then, too.
Kerrie Holley: 42:12 Uh, that's true.
Maurice Conti: 42:12 And we had autonomous vehicles.
Kerrie Holley: 42:13 That's a very good point.
Maurice Conti: 42:14 They're called horses (laughs).
Kerrie Holley: 42:16 Yeah, most people don't know that uh, we had autonomous uh, tractors in the 40's.
Maurice Conti: 42:20 Yeah.
Kerrie Holley: 42:20 Uh, self-driving. Not horses (laughs). Real, real, uh, engine, uh, engine, uh, based vehicles. Questions?
Speaker 6: 42:28 The healthcare industry is notoriously late with adopting new technologies. Where do you feel that, uh, advantages of the augmented age will first be implemented?
Maurice Conti: 42:38 I mean, it's, it's a little bit going with the, with the previous question. Um, you know, I have to say, as, as an industry, um, you're incredibly well placed to take advantage of these technologies. Um, I, I would probably argue the best placed industry to deploy these technologies, um, in the world. So, beyond, um, manufacturing, construction, transportation, food. Um, like, like, all of the pieces are in place. Um, you have everything you need. You're right on the, on the precipice. You couldn't be in, in better shape. So, um, I'm not sure what the rate limiting step is, because you've got all of the, um, cards in your hand.
Kerrie Holley: 43:20 And I would also, uh, assert that, uh, we're at an interesting point in time, because for the first time, we're beginning to see very clear signals that the healthcare system will be disrupted, that friction will be reduced. Uh, there's just a tremendous number of players. I mean, Telephonica, they're a TelCo company and one of their moonshots is in healthcare. Uh, this is unprecedented. And it's all because of a new general purpose technology. If you haven't heard that term before, a general purpose technology improves over time. It gets better and better and better, because of more inventions.
Kerrie Holley: 43:58 But this is the ... I would assert that this is the first general purpose technology that really threatens the encumbrance in the healthcare system. And I think that's a big motivation for change.
Speaker 7: 44:11 So, based on your experience, uh, what were the attributes of a problem which was successfully solved by the AI?
Maurice Conti: 44:19 Yeah, I mean, uh, broadly speaking, um, the, the most successful deployments have been ... I mean, this sounds trite, but the kinds of problems that AI's are good at fixing, um, there's actually not that much out there, um, in AI today, like in the wild that, that can be deployed. Um, so if you have even a basic understanding of the tech, it's usually pretty obvious. Um, uh, you know, take ... What I think is interesting, we're, we're, uh, we're not disrupting jobs, for instance. We're disrupting tasks. We're automating tasks, right?
Maurice Conti: 44:52 So, um, and this is a, a problem that spans the pyramid. It's not just sort of, um, entry level, um, work, because, um, a fast food worker would be almost impossible to automate, because they do so many different things. That would be an extremely different ... Uh, difficult problem for a machine system to, to automate. Um, if I was a lawyer or in finance, I might be a lot more worried.
Maurice Conti: 45:14 Um, so understanding what the tech can actually do, um, I, I think is this critical first step. And the, and the set of tech is not super wide today. Um, and so, it's actually, like, not a big ... Like, you know, a day long workshop, um, you'd be able to walk out of that room with some pretty clear ideas about, "Oh, here are the problems that we can apply these, um, technologies to."
Maurice Conti: 45:37 So, it's a little bit of a thinking exercise. Um, some exploration, some familiarization with the tech, and then introspection into, um, into what, what you're trying to fix inside of your organization.
Kerrie Holley: 45:47 And so, obviously we can see a lot of examples in the real world today with self-driving cars, with facial recognition, and the areas of, of crime and security. Uh, in, in our particular, uh, company, we have a ton of examples. We have Wave Zero, Wave One and Wave Two examples. Uh, some of the folks in the audience here have done projects where we've been able to go from Wave One to Wave Two, we've been able to show that we can predict with greater accuracy, and that has a profound effect on our population health, being able to diagnose with greater accuracy. More people would be getting diabetes ... Uh, that, that's pretty good, be able to do multi disease prediction, which is actually more important. That's pretty good.
Kerrie Holley: 46:29 We've been able to ... Uh, with case readmissions, uh, we've been able to reduce those, automate those, take friction out of ... I could go on and on. And we've got a, a ton that we've actually put into production already, um, in all of these different waves.
Kerrie Holley: 46:42 Thank you. I apologize, we don't have time for more questions. This has been great. Let's give another, uh, round of applause to Maurice.
Kerrie Holley: 46:51 So, on behalf of the Optum IQ Team, I thank you for this awesome event, and I thank all of you for tuning in, and have a great day.
PART 3 OF 3 ENDS [00:47:16]
AI and the Augmented Age in Health Care
In this OptumIQ™ Health in Focus event, two AI leaders with unique expertise, Maurice Conti, innovator and “benevolent mad scientist,” and Kerrie Holley, Optum® technical fellow, dispel myths and provide insight into the impact of AI in health care.