Podcast #228 Pediatric Moonshot

 TRANSCRIPT

Ken: Good day, and welcome to the Momenta Digital Thread podcast series. Our listeners know that Momenta has emphasized our industrial impact, driving positive change towards sustainability, resilience, and human centricity across critical industries through digital technologies. As we lean into this topic in our future podcast, I could not think of a better role model than Dr. Timothy Chou, the founder of Pediatric Moonshot. Dr. Chou is considered one of the early leaders in moving enterprises to the Cloud, beginning with his tenure as the first president of Oracle's Cloud computing business. In 2004, he published the landmark book, "The End of Software," which foretold the rise of enterprise SaaS and featured three startups: Salesforce, NetSuite, and VMware. In 2005, he launched the first class on Cloud computing at Stanford University, cs309a.stanford.edu. Today, he's the chairman of the Alchemist Accelerator and serves on the board of Teradata and Oomnitza. He recently came out of retirement to lead Pediatric Moonshot, the mission of which is to reduce healthcare inequity, lower healthcare costs, and improve outcomes for children locally, rurally, and globally by creating privacy-preserving real-time AI applications based on access to the data in all 1 million healthcare machines in all 500 children's hospitals in the world. To get to the moon, they've engineered a new rocket- a privacy-preserving, real-time Edge Cloud. Tim, this will be an exciting topic, so welcome to our Industrial Impact podcast.

[00:02:19]

Timothy: Ken, thanks for having me. It's going to be fun.

[00:02:22]

Ken: It will be, and I remember the first podcast. We featured you in episode 132 in March of 2021, covering your accomplishments as an entrepreneur, educator, executive, and author. We ended on the topic of your groundbreaking work with Pediatric Moonshot. Today, under the theme of Industrial Impact, I'd like to go headlong into that topic. Let me start with a few questions I asked you last time because so much has happened in the last couple of years. What a difference three years makes. But let me ask: what new topics are being discussed in your legendary Stanford course and/or that you're seeing at your Alchemist Accelerator these days?

[00:03:04]

Timothy: Well, Ken, it shouldn't be too much of a surprise. We all know that a little over a year ago, the whole gen AI, ChatGPT thing arose. I only do this class once a quarter. Last year, we had the CEOs of Intel, Franklin Templeton, and HubSpot- I mean, the AI conversation threaded through everything, whether that's how it transforms software development or what that means for next-generation chip technologies. Yeah, I don't think it should be surprising that that conversation is part of every conversation right now.

[00:03:48]

Ken: You might substitute Nvidia for Intel for this year's course.

[00:03:59]

Timothy: Well, that's chasing them down. He's on a mission there. To make a point of it from a national point of view, it's quite remarkable what Intel is attempting to do, and important. It's great to give Pat a stage to talk to the next generation of kids who will affect much of this change.

[00:04:25]

Ken: Yeah, which I've always loved about your course. It's so forward-looking to have discussions like that with industry leaders at a major university like this and to do so, especially in Silicon Valley, where everything's pretty much forward-looking. Last time I asked you a question, "What would the 2021 Tim Chou whisper in the ear of the year 2000 Tim Chou?" To see and put a reinforcer on what has changed, I can re-ask the same question but obviously with a 2024 lens. What would the 2024 Tim Chou whisper in the ear of the year 2000 Tim Chou?

[00:05:03]

Timothy: There'll probably be more change than you could imagine. Those of us who've been in the tech business are somewhat used to change- it's part of the fabric of working in this. In 2000, I would have said, "You haven't seen anything yet." Whether that's COVID or gen AI or- frankly, the rise of the cellphone is even in this category. I mean, I don't know that any of us realized the impact all this could have or will have, so a lot of times I say to people, I got to be there at the beginning of "Cloud computing," which, if you go back in time, we're now 15, 18 years into this. We're one year into the gen AI thing, maybe. It's hard to imagine what- go forward another 10 or 20 years, what it will mean. Yeah, I would say, "Buckle up. It's going to be bumpy and interesting." How about that?

[00:06:16]

Ken: The famous story, of course, is Albert Einstein while working in the patent office in Switzerland before he became super famous, right? Quoted as something to the effect of everything invented probably has already been patented, etc., right? My quote's probably way off, but the point is, generally, as tech prognosticators, VC investors, and everything else, we saw more and more refinement of existing things. IoT platforms, AI, etc. But man, gen AI came out of nowhere in some sense, and it started to strike me literally- it was social conversations. My wife and I were out with friends; they're in HR or whatever profession, legal, etc., and they're talking about using ChatGPT to write contracts or, in some cases, code. It was like- and this was March a year ago, so I agree with you about how fast this has come up, and now we have started to understand and appreciate the impact. I've been to several industrial conferences like Bosch Connected World recently, and in every session, gen AI, gen AI, gen AI, right? Living in such an interesting time is amazing, yet, as you say, we're just scratching the surface of this impact. What a great time to be alive in that regard.

[00:07:37]

Timothy: Amen.

[00:07:38]

Ken: As promised, let's go ahead and jump back into your founding work on Pediatric Moonshot and, of course, a project to create privacy-preserving real-time applications based on access to data in all 1 million healthcare machines in all 500 children's hospitals. That is quite an ambitious goal. Can you remind us of the overall Moonshot goal here and probably a little bit of the DNA of how you got to this?

[00:08:07]

Timothy: The origin story starts in my Stanford class. This is six or seven years ago, and a rather unique student who has an MD, an MPH, an MBA, and is Chief of Pediatric Cardiology at the Children's Hospital in Orange County showed up. His name is Dr. Anthony Chang. Anthony kind of adopted me because up until that moment, I had a passing knowledge of healthcare, but I've never been immersed in it. By adopting me, I learned things like- you're kidding me; they're still using CD-ROMs to pass data around, which is true even today. Then, at the other end of the spectrum, because obviously, I live in the tech community, we all know that to build next-generation deep learning applications, you will have access to large quantities of diverse data. As COVID begins to descend on us, I'm thinking, "Are we going to sit around and watch Netflix all day?" I said, "No, maybe we can do something useful." As you have stated, I brought a team together with the mission to reduce healthcare inequity, lower costs, and improve patient outcomes for children locally, rurally, and globally. How are we going to do this? By creating real-time, privacy-preserving AI applications based on access to data in all one million healthcare machines in all 5000 children's hospitals in the world. That has been our mission, as you already alluded to. We had to build a new rocket. We've engineered an Edge Cloud, enabling the Cloud servers to be in the building. In fact, when we talk to the clinical community, I don't actually use the word Edge Cloud because they go like, "What Edge?" I go, "In the building Cloud service." right? Okay, that is plain English. The Cloud servers are there, so why must they be in the building? Well, the only reason they need to be in the building is that the only way to talk to them, and I'll use ultrasound as an example of this. The only way to talk to the ultrasound is you have to be on the network in the building with the ultrasound. You can easily guess that the very first conversations with hospital IT were followed by NFW.

[00:10:42]

Ken: Been there, done that, my friend.

[00:10:44]

Timothy: Been there, done that. We're a team of infrastructure. We've been around this conversation for years and knew and engineered over 30 security and privacy features. In fact, one of our core team members is a former student who has 15 years of privacy law experience, so we knew security and privacy were not something we wanted to add at the end. It was core to the system's architecture, and we used that to build a capability since some of your listeners are somewhat technical. In a sense, we engineered what you could think of as EC2 and S3, so compute and storage capabilities at the Edge in the building. On that, we deliver a particular kind of application called a digital twin. I know that word is used in many ways in the IoT space, and the way we use it is basically to say that for a class of machines- so I'll use a specific example. Philips ultrasounds, from their big Epic 7s down to their portable Lumifys, the digital twin will replicate the data in the machine. When I say data in the machine, what is that? Well, that's the static data. What's the serial number? That's the dynamic data. What's the last error code it threw out? Where's the environmental data located? What's the temperature in the room?

Finally, what we call the nomic data, the actual echocardiogram or blood analysis or MRI image or whatever are all presented in what we call Edge data services. Edge data services are how we standardize access to the data to enable authorized applications to use any or all of what I just stated as the data coming from the machine. That's the fundamental architecture for those who are not technical; I explain it as it's kind of like what Apple's done with the iPhone, meaning they distributed infrastructure around the world that gave people access to a development environment, to a camera, to GPS information, and then told the world, "Please go build Instagram, build Lyft," all these apps that we are aware of using that data. It's the same fundamental idea, except our camera is an ultrasound or blood analyzer; fundamentally, we have GPS information. It's just one of the applications, whether cardiology, orthopedic, and neuro-radiology applications. That's been our journey; I'll just quickly- and then jump into questions. As an engineering team, we said, "Let's leave PowerPoint," so we deployed Edge zones in eight children's hospitals on three continents. We wanted to be global on day one. Our work has pivoted into applications in the past six to nine months. To use the analogy again, if Apple had just released a development environment, we'd all have gone, "That's kind of cool, but what does this thing do?" They had to build maps, Safari, and all these first apps, so we're focused on two major areas. We call it The Mercury Program, which allows non-children's hospitals to share images with experts at children's hospitals with a primary focus on emergency medicine and a secondary purpose, which is to allow for second opinions. Your listeners may not know this, but not many pediatric experts are on the planet.

In the US, I'll stick with cardiology; we have about 3,000 pediatric cardiologists. By the way, none of them live in any rural county in the United States. In India, we have 300 pediatric cardiologists. In Rwanda, there's one guy. That leads you to the second part, which we have dubbed The Gemini Program, which is to build AI in cardiology and orthopedics. The challenge we are on to technologically is how do you do this in a privacy-preserving way? How do you build AI applications in a decentralized architecture? This is not how ChatGPT or other more commonly understood AI applications have been trained. That was a long answer, Ken, but that gives you a little glimpse of why we did it, what we've done, and where we are today.

[00:15:43]

Ken: I appreciate it. You hit several of the questions I was naturally going to want to ask. Common sense would normally be, let's take a very low overhead use case and develop a technology solution for it because if we can make it work there, we can begin to raise it so that it can work in probably a more regulated solution or environment or more mission-critical or whatever. You have chosen one of the most mission-critical, privacy-challenged use cases to go after- pediatric data in the medical profession touching on North America. I think it's beautiful because of many of the things you talk about in terms of your platform: the design, the engineering, on-prem- as you say, in building. I could literally change the industry from medical to digital manufacturing, and a lot of the same needs would emerge naturally, and your platform would likely solve those. You've chosen the biggest, hairiest, and most audacious goal; thus, Moonshot is probably appropriate. I'm going to ask a couple of tech questions, though, because I'm very curious, and again, there are some analogies here to the whole industrial IoT as we like to focus on it. Number one is this idea of- you mentioned the digital twin, about getting clean, contextualized data from industrial equipment and systems. How are you solving for that in Pediatric Moonshot, i.e., are you leaning on the OEMs to actually do the digital twin or on the end users, or are you guys doing this voluntarily to bring all this together?

[00:17:23]

Timothy: Let's describe architecture, and then I can talk about capitalization if you like. We did not want to depend on the OEM for the digital twins for a machine class because even if we could go, "Oh, let's give us private access," we all know those are unsupported and will break in the end. All machine interfaces are through supported interfaces; therefore, we don't need the OEM to do anything because those are supported. On the standardization/harmonization front, because we've come at it bottom-up rather than top-down, which I think is a little bit of an advantage for us, what we decided to do is we said. I'll use blood analysis as an example. Siemens builds blood analyzers, and Beckman builds blood analyzers. The way the digital twins work is that all Siemens blood analyzers, I'll say all the data they will deliver through the digital twin, will be identical for all Siemens blood analyzers globally in our architecture. All Beckman data will be standardized across all machines globally. Now, we leave the rationalization of the Beckman and the Siemens blood analyses as an application-level question. We're not trying to crack through that one. We're trying to establish the foundation for a machine class to standardize the data and leave it to others with domain reasons. This is where we're not thinking from an application point of view but from an infrastructure point of view. We'll let the applications figure out what data they need to rationalize between, in my little example, Beckman and Siemens. Does that make sense?

[00:19:27]

Ken: It does, and perhaps a few minutes ago, you mentioned capitalization models. I'm as interested in how you guys, when you speak at a foundation, what the underlying business foundation is. How are you working with all the supply chain members or ecosystems there? Is it a not-for-profit, or are you doing it for profit? How are you guys set up for this?

[00:19:50]

Timothy: No, we decided- I mean, at the end of the day, we wanted to build something sustainable. If you think of it from conventional non-profits, all of that is donation-oriented, which is an every-year exercise that doesn't generally lead to sustainability, so no. BevelCloud, founded to deliver the Edge Cloud services, is a regular old corporation. I'll explain the business model fairly simply, which is it's an essence, a mimic of AWS. In fact, internally, we call it BWS. Here are compute and storage instances, and there are so many dollars per month. We're more in a reserved instance model of the world. Like AWS, which started with just a couple of EC2 and S3, today, they have 37 variations of that and 400 different services. We also see the same thing. We, meaning BevelCloud, see the same thing: creating additional services. Obviously, we've been working a lot in decentralized or federated learning, so that's an example of how we develop capabilities that go beyond just the conventional computing and storage services. That's very straightforward. As I said, it is a fairly simple business model. From a capitalization point of view, I have funded our efforts to date. Our venture capital friends are not very interested in the pediatric healthcare market. In fact, I have some interesting data. I forgot who put it together. The amount of money invested in pediatric healthcare last year was 30 million. De minimis amounts of money. Because obviously, in health care, it would be more lucrative to go after old rich people as a market rather than kids, which is obvious. I said it to somebody this morning. If you and I were money investors, why would you do- if you're going to do all this effort, go after old rich people, it's obvious what you should do. We've known that for quite some time, or I've known that for quite some time. We have sized the next level of what we're working on, what I described as Mercury and Gemini at roughly 112 million. As I said, we're not going to venture- we're not going to the children's hospitals. I've now deep-dived into some of their economics. It's crazy. UCSF Children's over here, 20% of their operating budget is made up of donations. The hospital runs on donations, which is just sad. We're not going there; there's no money there.

We are fundamentally on two paths. One is government sponsorship. Some of your listeners may know that the origin of the Internet was really the United States government under the aegis of ARPA, which basically funded the foundational backbone of the Internet. We, in essence, are building a network as well. If you listen to how we're talking about it, I mean, 500 children's hospitals, and obviously, as we extend this out into adult medicine as well, we're building a network. When you build a network, you can't do it one note at a time; you have to do it together, and so conveniently enough, or maybe a blessing, there's a new agency that got started called ARPA Health that was allocated $2.5 billion about a year and a half ago. Hence, we are deep in conversations with ARPA Health. We've also opened up conversations with the government of Singapore. We're going to have some introductory conversations in Norway. If you think about it, we're fundamentally building infrastructure, and it makes sense for governments to fund infrastructure. The other major axis is what we call corporate sponsorship. I'll have a little fun with you. My friends at Juniper Networks, I was talking to them, and I said, "Hey, you guys. You spend a million dollars to put a logo on a Formula 1 race car." They went, "Well, it's way more than that." I said, "Okay, you're helping me make my point." We have various threads going on with leaders in various industries. I'll give you one example, which is kind of curious. I had the CEO of AGCO come and do my class for the first time four years ago. The new CEO came this time, so I posted something on LinkedIn because I know the AGCO guys. In fact, they're a chapter in the book "Precision" that I wrote about IoT many years ago. One of the guys who's the global product manager for their Fendt product line sent me a note. He says, "Hey, we haven't talked in a while." I said, "Oh, Matt. Let's get together." We did a Zoom meeting. I described what we're working on and said, "Well, Matt, we were thinking maybe AGCO could sponsor the state of Georgia."

One other idea we've been playing with is 50 states; maybe the largest company in each state sponsors for $2 million, which gets us close to the 115. I said, "AGCO is a major employer in Georgia. 2 million, what about that?" Matt says, "Well, I have a much better idea." I went, "Okay, I'm good with that." He said, "Well, our customer and your customer are the same," and I never thought about it very much. Still, obviously, what we are working on while if your mom works for Google and you live in Palo Alto, it could make a difference; it makes a huge difference if your mom works on a farm in southern Illinois and you're 300 miles away from any children's hospital. By the way, the rural communities are also where all the farming equipment and ag equipment get sold. "My customer is your customer" really resonated. We're actively having conversations in the agricultural vertical, which, to your earlier point, has an interesting overlap coming from a product side because, at the end of the day, all of that agricultural equipment- they're trying to do the same things. As we know, they've been working on preventative or predictive maintenance, which, from our point of view, is dynamic data about the machine. But in the end, what's really important is the agronomic data. Like, what's the boron level, what's the weather? These things have similar challenges from a privacy and a real-time perspective, trying to aggregate all the data centrally to learn on. My point of view is in healthcare, it's never going to happen privacy, preservation, etc., and the data volumes. But I think, in the end, in many industries, the idea that I'm going to put all the data in one location and let you learn on it, it's going to work for some consumer stuff, but I don't think it's going to work for a lot of other industries.

[00:27:30]

Ken: I think you've undertaken one of the largest digital transformations in one of the toughest industries in that regard, again, which makes you applicable and useful for many other industries. You mentioned agriculture in that regard, so it's a beautiful story. I'm glad we've had this conversation to bring that out. Let me bring back the industrial impact element, which is getting down to the impact you've had on some of the people whose lives you've changed by this work. Can you tell us a story or two?

[00:28:03]

Timothy: You've made a comment about Moonshot being ambitious and big. I also tell people this is my last great project. It's a confluence of people I know and technology, and I understand that, weirdly, comes together for a game-changing purpose. Our hope is we can solve problems. I'll give you two stories so people can think about how technology impacts this. There's a kid, a friend of mine, who went in for optional shoulder surgery in Arizona. There were complications during the surgery to such an extent that they decided to airlift him out of Arizona to the Children's Hospital in Los Angeles. He arrived on the helicopter, but the CD-ROM of his CT scan did not. He actually died later that day. Now, whether or not that image could have made a difference, none of us really know, but it certainly couldn't have hurt to be able to have that. We've spent time with Howard Memorial in Willits, California, population of 5,000. To this very day, at this very minute, when they transport a kid to emergency at Stanford, UCSF, Shriners Burn Unit, etc., they send a CD-ROM, which often does not make it for various reasons. By the way, even if it gets there, they can't read it, and it's just stupid, given the type of technology we have in front of us. That's one. It's super simple, right? We could change that on a nationwide and global basis. Let me give you another one in which we can make a difference. There's a kid in Florida right now who has had seizures two to three times a day. At night, he wakes up screaming for the past 12 years. Now imagine being his parents or his siblings. This is life, right? The MRI imaged him early on but didn't see anything. They went down this progressive path of every pharmaceutical. Some he could afford, some they couldn't afford, none of them effective. They were getting ready to do electrical implants, and they re-MRI imaged him. Now, what they believe is that he has a condition called focal cortical dysplasia. It is a brain lesion. I can show you a picture of it, although we all look at it and go, "I can't see anything. It looks as fuzzy as the next thing." Now, the cool thing is, focal cortical dysplasia, you can actually surgically remove that lesion, and the kid's cured for life. Freaking amazing, right? The good news is that there are only 2,500 cases in the United States annually. Okay, that's good news. The bad news is that no pediatric neuroradiologist can see enough of these to go, "Oh, I can see focal cortical dysplasia." Now, this is trivial from our point of view. You go, "Well if I had every MRI image of every kid across the planet, do you think I could build a focal cortical dysplasia application that could diagnose it and run it in real-time on every MRI machine?" Hell, yeah. That's not hard to do. We can do that. I think the promise of whether it's to make it so that we can use modern technology and save time and lives in emergency medicine, all the way to diagnosing what you could term rare diseases globally, is there. It's there to be had, so I think that's inspired our larger team. In fact, just as a comment for those listeners, please www.pediatricmoonshot.com, check it out. We do a monthly newsletter. We're on LinkedIn. We've actually just started a TikTok campaign. We're doing a series of podcasts under Pediatric Moonshot. For people interested in following along or becoming part of the crew- like what we like to say, please reach out.

[00:32:40]

Ken: What a great set of stories and the impact you're having- it's beautiful because the technologies you're developing are truly cutting-edge but also very much aligned with state-of-the-art and bringing to this critical domain and the critical lives there. New solutions, new capabilities. I think it's great that you're getting both a combination of government sponsorship and corporate sponsorship, and I would encourage our listeners to learn more about this initiative but also look at it as a pattern for change, for truly transformational change that's out there. There is so much we could talk to, and I didn't even get to the gen AI closing question, but I'm afraid we're out of time, which means we may get to do this a third time, Tim. We could discuss so much more here, but hopefully, it won't take us three years to get back together to learn more about this. With all that, Tim, thank you for sharing these great insights with us today.

[00:33:38]

Timothy: Ken, thanks for having me. I'm glad to come back and be a guest again. Thanks.

[00:33:44]

Ken: I'd really like that, Tim. This has been Dr. Timothy Chou, founder of Pediatric Moonshot. Thank you for listening, and please join us for the next episode of our Industrial Impact podcast series. We wish you an impactful day. You've been listening to the Momenta Digital Thread podcast series. We hope you've enjoyed the discussion, and as always, we welcome your comments and suggestions. Please check our website at momenta.one for archived versions of podcasts, as well as resources to help with your digital industry journey. Thank you for listening.

[The End]