Can we get an open source medical AI that saves millions of lives?
Everywhere on the planet, health care is a problem. In the U.S. it’s too expensive. In Canada, it’s too busy. In much of Africa and India it’s too rare, and in many other countries including Western Europe, health care systems are overwhelmed and overrun. This might seem like a rest-of-the-world problem for Americans, but the U.S. is particularly challenged:
The US spends twice as much as other rich nations on health care, but it has …
- The lowest life expectancy
- The highest death rates for treatable conditions
And the highest maternal and infant mortality
AI could help, but only if we can get our act together, says HIPPO AI foundation founder Bart de Witte, who is our guest in this week’s TechFirst with John Koetsier …
(subscribe to my YouTube channel)
Subscribe to TechFirst with John Koetsier
Apple Podcasts
Spotify
Google Podcasts
Overcast
Other podcasting platforms
Full transcript: Open source medical AI
Note: this is AI-generated and lightly edited. It could be mistaken. Check the video or audio if in doubt.
John Koetsier:
Can AI save healthcare? Quite literally, can it save our lives?
Hello and welcome to Tech First. My name is John Koetsier.
The US spends twice as much as other rich nations on healthcare but has the lowest life expectancy, highest death rates for treatable conditions, highest maternal and infant mortality. In Canada where I live, the healthcare system is universal but it’s stretched far too thin. Wait times are up, treatment is essentially rationed. In Europe and rich Asian countries like Japan, Korea, healthcare is very good, but costs are growing everywhere. And the rest of the world, huge portions of China, huge portions of India, lack good quality healthcare. Africa, South America, many of the big countries in Asia as well.
Question is, how are we going to deliver good quality healthcare to billions of people at a global scale? Today we’re chatting with the founder of Hippo.ai. It’s an open source medical AI foundation. He’s formerly from IBM and SAP. His name is Bart de Witte. Welcome, Bart.
Bart de Witte:
Thank you for having me John. I’m looking forward for this conversation. Small correction, it’s hippoAI.org, not hippo.ai.
John Koetsier:
Oh, and I thought that I got there. Maybe Google is just smart enough to make up for my idiocy. I’m not sure, but I’m really, I’m really happy to have this conversation too. I’ve been looking forward to this for a long time.
We’ve been connected here, there, everywhere, sort of for it’s gotta be years upon years. And I’ve had this growing feeling living in Canada, seeing the healthcare system really stretched, being traveling to the States a lot and seeing so many people who don’t have any healthcare, nothing. Number one cause of bankruptcy in the US is a medical emergency that somebody didn’t have the right amount of insurance for or none or something like that, right? And then you just see the problems globally as well.
Can AI help us here? We’re gonna dive into all that stuff, but maybe let’s start a little bit at the beginning. You’ve been at IBM, SAP. How did you come to establish an open source medical AI foundation?
Bart de Witte:
Yeah, thank you for asking the question. I think because I didn’t see the hope that when I started my career in the health tech space, I expected that digitalization or healthcare would lead to a democratization, but healthcare is a bit more like Game of Thrones. It’s full of kingdoms fighting each other and there was no democracy.
So when we start turning everything into a digital system and starting to transform these systems, then data and AI is gonna play a huge role in terms of equity and access. And then I saw that we started to adopt the main principles that, or the same principles that we’ve seen with the internet economy, where we start turning how data, our data, into capital value. And then I said, like, this is probably not gonna lead into a positive direction if we start doing that, because if we equal held data to capital, then data accumulations will follow capital concentrations and that will lead to the same asymmetries that led to the power asymmetries that lead to inequality.
So from that perspective, I said we need to do something similar to what we have seen the last 30 years with open source software. There was the Linux Foundation, the Apache Foundation. Coming from IBM, I’m very aware of how IBM fought Microsoft 20 years ago.
I think a lot of people are not aware that when you wanted to learn to code 30 years ago, you could not go to GitHub, you could not go to libraries, you could not learn that from home. You had to go to these companies to learn how to code well. And that became democratized, and it’s part of that whole revolution that we saw.
And I think if we don’t do that with health data, making it accessible, and we don’t do that with medical AI, and we don’t democratize this, this path of continuous growth of costs and as well the gap between the hefts and the don’t hefts will just grow larger and I think we need to change this. And that’s how I started.
John Koetsier:
It seems like a good place to start in Europe. And I’ve thought the same for Canada as well, because we have what they like to call in the States socialized healthcare systems, where everybody is more or less treated equally. You come, you need healthcare, it’s provided, it’s covered, it’s cared for.
There’s gaps and I’m sure there’s inequalities within those systems, I know in Canada, I’m sure in Europe as well, and there’s private stuff that you can get and all that. But in those places, governments have a financial interest as well as a humanitarian interest and an economic interest in providing great quality, affordable healthcare.
And you would think that in those scenarios, establishing large databases of safely acquired privately maintained healthcare data that an AI could learn from, see, and help contribute to would be a massive bonus.
Bart de Witte:
Yes, like a Wikipedia of data would be definitely a massive bonus. As the data that we use needs to be refined and curated and engineered, there is a layer that comes in between that. And the question is who pays for that?
And today it’s mostly the financial driven industries that say like, well, we will have something similar to that very bad technology with oil, we will refine it. And then when it’s fine, we can own it and then we can sell it. And that’s kind of one of the… starting points.
It’s like we need funds that are funding the creation of common goods or data commons. So we can turn that unrefined data into curated data and make that accessible. You would think that a government would do this. No. Here, even in Europe, they see all these data sets as a hidden value that they can resell to the industry so they can create… better public finances.
And I oppose this because it’s similar to scientific research that is being sold as a PDF that you download for 60 euros, which is funded by public resources, where the author doesn’t get paid for it. And then Elsevier and Springer are getting more profitable as well as Google. And we know that was a mistake. We are fighting back for open access. And we’re doing, again, that same mistake when it comes to data.
John Koetsier:
Mm-hmm. Let’s broaden the conversation a little bit. Let’s say that we had a goal of delivering a certain level of quality of health care globally to 8 billion people. How could we reasonably achieve that in a reasonable time scale?
Bart de Witte:
Well, I think that is a very simple question, but a very difficult topic. But I think it’s all about getting access to life-saving knowledge.
And if knowledge is an extraction out of data, I think that’s where the starting point is, allowing people to access these artifacts that allow you to build medical products, to write your own digital stories.
I sometimes compare it with giving access to the alphabet so we can all write our own books. And I sometimes compare today’s value or the situation that we are living in that if Gutenberg, who invented the printing press, would have lived today, he probably would have been advised to patent the printed alphabet.
Now, if he would have done this 300 years ago, we know that Luther would never have written his Bible. We would still have been buying our letter of indulgences, and we would still believe that health or disease was because we had sinned and we would not have seen the enlightenment. So I believe that some things need to be accessible for all and these are in technology called as general purpose technologies but in data it’s a bit difficult to talk about technologies but I think that we need a common layer that allows us to innovate, to cross-pollinate, to collaborate and I think to answer your question, the only way to do this is to collaborate and not to compete.
John Koetsier:
I have to think that there’s a role for AI to play here and a couple things bubble up to mind as I think about that. One is I recently did a full body MRI scan from Prenuvo, found a bunch of stuff, it was interesting, didn’t understand it, put it into GPT-4, got an explanation. That made sense. That was nice.
There was a story that came out, I want to say a month ago. somebody who got a diagnosis for their pet, I believe it was a dog, and nothing was working, put the symptoms into GPT-4, came out with three possibilities, took them to another vet. The vet said, oh yeah, that makes sense. This is probably the one. Let’s do a test, confirmed it, boom, saved the dog’s life.
How do you see medical AI developing? Obviously it needs access to the data, and the data is core and foundational, but how do you see medical AI Developing we’ve seen already like it was what’s three years ago a medical AI or a medical Smart system in China passed there The the level of being the test to be a doctor, you know, obviously there’s lots of caveats around that But how do you see AI developing for health and medicine?
Bart de Witte:
I totally believe that AI will allow us to improve healthcare in dimensions we haven’t seen before. But there was a prerequisite here because if we see for example insulin, which was invented by a Canadian, or like it was not invented, the synthesization process of insulin was invented by a Canadian, who … who sold it for one dollar because he said like this belongs to humanity.
Now 100 years later the price of insulin has increased in the last two decades by 1,200%. Last year 100,000 people in the US died of diabetes.
And that is not because we don’t have access to insulin, because insulin was not invented. It is because of the extraction in the financial markets. We are living in a system where if you have something that is an intangible asset, which is knowledge, and you turn that into something that is protected by legal, political barriers, that is IP protection, you allow others to increase the price if the demand goes higher.
So it’s not like a commodity where the price goes down when you have a higher demand. No, it’s the opposite. And you have seen that in many, many cases. So I believe if we… allow a system like OpenAI that is very close, that is similar as insulin or a drug, is protected with these legal barriers, that at the beginning, yes, this will be cheap, but once you have a monopoly in healthcare, the prices, and that can be in five or in ten years, will go up.
So you always need to look at the foundation of what we’re building up, or what is going to be the consequence in five to ten years, or in two generations. AI needs to open in that sense and turn it into a common good. We won’t see these benefits and we will design a system that is unequal as the system of today.
John Koetsier:
I had to smile when you said the name Open AI because names are funny things, right? Open AI. And of course your hat says, make AI open again. There’s been some controversy … it’s like the democratic Republic of North Korea is not democratic at all. It is probably not a Republic. Right?
Bart de Witte:
Yes.
John Koetsier:
Names throughout the history of countries. You know, if you put Democratic in it, it’s probably not. Now I’m wondering if you put open in it. Is it not either?
Bart de Witte:
Exactly.
It’s an open coffin. Now, I think OpenAI is the biggest joke out there because nothing they published in GPT-4 in that 49 pages paper was actually open. Like there was no in, like they didn’t mention which training data.
They probably used comments like Wikipedia, which is a common layer and taking that comments and appropriating that, closing everything down. And then turning that into a sort of closed black box. And I’ve always warned about this, like, don’t worry about the technical black boxes. We should worry about the legal black boxes, because AI models are protected by trade secrets, which is different as with patterns, patterns are time limited, trade secrets are not time limited.
And if you have non time limited protection, then this can turn into a very powerful … platform over time where, to give an analogy, to go back into time, which is a bit comparable to the early days in Europe where we had the Vatican, which was a moral instance that had a secret library that we were not able to access, the Apostle Library. The Latin language was not accessible for the people, there was no democratization of knowledge, and we called it here in the western world the Dark Ages because there was the time where we had less progress, we were living in feudal systems and we didn’t have access to knowledge.
And there was this term tech feudalism and I think open AI is a feudalistic platform that over time can use that power, knowledge is power, use that power for their own means and perhaps government means and others. I think it’s a very dangerous development.
John Koetsier:
My hope, at least in the medical space, is that the nations that are literally spending hundreds of billions of dollars annually on health care will see the light. We’ll see … we can actually learn better, we can provide better outcomes, and we can actually provide better access to all kinds of healthcare by centralizing our data in a safe way, in an anonymized way.
That’s pretty critical, that’s pretty important, and sharing it, letting researchers access it, letting AIs access it, and integrating smart systems into our healthcare system with doctors. And there’s a lot to think about there.
There’s a lot to do, you know, what will doctors do? How will they treat patients? How will they be in the loop or even out of the loop in some cases? Will we trust that? Is it like full self-driving from Tesla? You know, why don’t we trust it to a certain extent, but then there’s some checks.
That’s gonna be an interesting next five to 10 years because the governments are not gonna have a choice. If I’m looking at Canada where I live, we’re spending tens of billions of dollars annually and the outcomes are decreasing for a lot of different reasons. And we’ve got to start thinking at a different level to solve problems that we can’t just solve by shoveling money at them right now.
Bart de Witte:
My hope that governments will solve this has decreased.
I live in Germany, the country that always points with the moral fingers to other countries. But then when we were living during COVID, it was also the country that created what I call vaccine apartheid, where we were talking about our third vaccination, where the global South didn’t access to any vaccination, where at the end it was a team from Texas University and the team from Peter Hotez who created an open source vaccine that was developed for only 7 million US dollars that then was used by Indian biotech companies to be put through clinical trials and distribution.
So there was even the German government celebrating BioNTech, of course because the tax income was quite high. That’s something that I didn’t realize a lot that if you turn everything into flourishing industries that if like Germany used to be called the pharmacy of the world in the ninth end of 19th century that there was this dream within the governmental mindset that through the tax income things can become expensive, but if you create and get taxes through these companies you can kind of afford your health care system, and that’s a very old way of thinking because that means that you are designing a system that always will lead to the global inequalities that we’re seeing.
John Koetsier:
100%.
Bart de Witte:
So there was this missing global perspective. And I don’t think the World Health Organization with COVAX, which was a program that tried to create health equity during COVID, didn’t work. I don’t think these things work up-to-down. I think that they will need to work from a bottom-up approach, where we allow local innovators to build solutions.
And to be able to do that, we need… to give them access to these assets.
John Koetsier:
And we need to start collecting these assets in a systematized way, in a national way. I think that’s really critical, especially in the States, which has a very federated health system. Your HMO might have a lot of data on you, but you go someplace else, they don’t have any clue about you.
In Canada, separated by province, we have provincial healthcare systems, not a national healthcare system. But if we can get that data… I think that’s really critical and I think that there’s a role to play for wearable tech as well, right? I mean a critical part of health care is prevention and if we can stop people from contracting very preventable diseases and illnesses, we can also preserve dollars — critical dollars — and bed spaces for those who have something that they can’t control, that happened to them.
So I’m hopeful that we’ll have some system in the future that will integrate data like this and then provide incentives to have a healthier lifestyle rather than just treating what happens no matter what happens.
Bart de Witte:
I used to think that way as well, but like one of my co-founders is the former general manager of the National Health Record in Austria. And during COVID, Austria was the first and I think the only country that mandated vaccines for a very short while. And he lost his job, like he didn’t lose his job, it was his decision, because the political power… was mandating him to give access to all the non-vaccinated data and hand it over to the police.
And this case is also being described in public, shows that as soon as you start centralizing data … with the appetite comes the hunger for power.
And so living in Europe where data, centralized data decided in the Second World War if you would be deported to concentration camps or not, because that’s where the high data protection comes from in Germany. Like there are in Holland double as many Jewish people that were deported as in France because the census databases were destroyed in France and in Holland not.
Seeing the history and knowing that sometimes that centralization of data is leading to absolute abuse of power is, I think, a dangerous to wish for scenario. And we saw it with COVID. There was even …
John Koetsier:
You can’t think of a safe way to do it technologically for protecting privacy?
Bart de Witte:
I think we need to work more towards edge cases, like edge AI, edge storages, and then create systems which are data trusted, can be private organized, that allow this. But I’m not a fan anymore from centralized governmental systems that have that power, because we saw… especially during COVID, how that power was abused and how suddenly there was such an transparency in these systems.
And I oppose this now because of the learnings. And I changed my mindset as well because I used to be a quantified selfer and I had all sort of wearables and tracked myself.
There was a book that I read from a German intellectual, her name is Julie See, and it’s also translated in English, it’s called The Method. And it’s about Germany in the year 2058, which became a health dictatorship.
So the scenario is about the government dictating that and monitoring everything that you all need to live healthy, because it was like putting prevention so far that creating incentives that if you don’t live healthy, you get punished in that sense. So your toilet was monitored. And if you threw up in that book, if somebody threw up because he was drunk, she got in jail. And when that book came out in 2008, I hated the author. I said, how can you be so negative? How can you be so dystopian? This is all good.
Now, when I attend pitch events from startups, I see everything that was written in that book being now proposed.
Then I read the European HAL Data Space proposal, which is part of the Data Governance Act as a new digital legal framework. And there was written in that proposal that in case of a pandemic, that private companies need to give the government access to that data. So we are putting everything in place to create what was written in that book.
Now… I’m not as dystopian in that sense, but I think we need to avoid these sort of systems allowing people to have that power. Because what is prevention? Are you going to mandate lifestyle? Are you going to mandate these things? I don’t think this is a healthy way to look forward. I think by education and by giving people higher wages and higher life quality, I think there are better food quality. All these things, that’s how you can influence health, but not my mandating prevention, or sleep tracking, or all these things.
John Koetsier:
Yeah, it’s a great point because nobody … we don’t want Big Brother. We don’t want Big Brother. We’ve seen that.
Bart de Witte:
We have Big Brother.
John Koetsier:
To some degrees in some places, exactly, but not to the level that they’re monitoring my toilet. Let’s put it that way.
Bart de Witte:
No, hopefully not. You never know.
John Koetsier:
We do check wastewater, but it’s not individual.
Bart de Witte:
No, it’s not individual, but yeah, true. But we started to check wastewater.
John Koetsier:
Sure, you can see where COVID is still going in the world by checking wastewater. I’m in the … Florida coronavirus subreddit and you see the ebbs and the flows there Absolutely, so we don’t want that so we do want however is a system where we can teach smart systems. We can provide health care better.
We also … I mean there’s hard questions here. There’s really hard questions here. Let’s just think about them: we have somebody who’s a smoker, a drinker, and eats more than they should. That person is going to take 10X more healthcare resources than somebody who doesn’t drink much, doesn’t smoke, reasonably works out or exercises and doesn’t overeat.
There’s edge cases. There’s you know, Winston Churchill lived till his eighties and did all those things, right. But they’re the general case.
And so there’s issues of equity. People have to make hard choices about who is going to get these healthcare resources, right? So I don’t believe in mandating healthcare prevention. I don’t believe in big brother. I don’t believe in monitoring. I think that there could be some solutions to guarantee individual privacy.
And it’d be interesting if you had some solutions around healthcare that were similar to some of the… car insurance systems that you see coming out over the last few years where if you’re a safer driver, your insurance rate goes down. It’s tough, it’s challenging, lots of questions of monitoring, privacy, all that stuff, I totally get it. I don’t know the answer, but I suspect there is one.
Bart de Witte:
I think for me I go back to education and that sounds as well to the social determinants of health. I think if you mentioned drinking and smoking, if you look at today with today’s knowledge who drinks and smokes and who is living and healthy, these are not really the people in the high classes, these are people in the lower income stream mostly because they have a lot of social issues or have their … Yeah, there are challenges in life and I think everybody sometimes has their challenges and that’s why we have insurances which are based on a solidarity principle.
If we start now to… to personalize insurance and we cannot call it insurance anymore because we don’t distribute the risks anymore of a population. We start to create individual risk profiles and then you’re going to have to pay more because you are living unhealthy. That is quite a social policy if you start doing that. How far are you going to drive it?
Are you going to go to the genetic level where you have a predisposition of having a disease, which is what happened in the US where if you have once you had cancer you hardly can get an insurance in the US anymore because you were diagnosed once with cancer?
So I started as well over the years to turn a bit more that we need anti-discrimination laws because we’re going to go more and more into that transparency that data will bring. And we need more and more solidarity in these systems, not more personalization.
So I don’t think that the car example is one you can put into healthcare because a lot of these decisions you can’t influence. will certainly change your lifestyle and you’re going to be punished in that same sense.
So these are really tough questions and I think the only answer, like part of the answer that I try to work on is really making things cheaper and in healthcare… Like you’ve been in tech since a long time as well and you saw through Moore’s Law how prices fell down. And now there is in Africa there was a company called Marathon and you can sell an equivalent of or buy an equivalent of an iPhone 7 for 58 US dollars, which is 100% produced in Africa.
That’s beautiful. Certainly it changed the lives of millions of people because Moore’s Law drove these prices down.
Because in healthcare we have seen everything increasing in price exponentially, but the performance didn’t go up exponentially. Because we are living in this extraction world that serves the financial interest of the market because they know that when you have a disease where you’re going to die, you’re going to spend your last cent before you die. There was this perverse incentive in these markets that is leading to these higher costs.
So I think if you talk about health equity, we need to democratize technology. We need to drive these prices down. We need to stop talking in the age of AI where machines are going to invent, not allow machines to create inventions that can be patented. We are not allowing Dali to create images that can be patented. There was a Federal Trade Commission case where somebody tried to create copyrights on images and she couldn’t because it was a machine. that created this event. Now, imagine machines start to create new synthesized proteins, it start to
John Koetsier:
And they will!
Bart de Witte:
And they will, they will. Now there was a huge lobby that could still be patented. And I think we should fight that. If an artist cannot protect his work, why should we not do that in healthcare either?
Because that’s gonna drive prices down. If we don’t open source, if we don’t… create new copyleft licenses on data and do everything that we saw with open source, if we don’t apply that, we’re going to miss a huge opportunity to democratize healthcare.
John Koetsier:
It’s very persuasive what you’re saying, and I will totally go that way rather than risking Big Brother in our lives and in our toilets, on our wrists, in our eyes, on our faces when we have our new face computers from Apple and all that stuff. If we can democratize that …
Bart de Witte:
But Apple, I support Apple because they’ve understood that the currency of interaction is trust.
And Apple now with the new iOS 15, they announced that there was a very little message that nobody or not a lot of people saw that they implemented now in the iOS and LLM, a large language model, in your phone on an edge case where it’s doing auto-correction but also suggestion of your typing and they want to speed up your typing with a factor of 10 or something. Which is the best case for large language models because it’s a prediction of what you’re going to write.
But it’s implemented on your phone with your local data and it’s not going into a cloud server. And I think that was so beautiful to hear that Apple implemented it in a privacy preserving edge case. And it was a pity that not a lot of people picked it up because that’s for me the future of AI.
John Koetsier:
It must’ve been the announcement in iOS 17. And I did see some of those …
Bart de Witte:
Or 17, yes.
John Koetsier:
I did see some of those things that Siri is going to get better and that auto correct is going to get better, which it could hardly be worse, frankly, on iOS right now. So that’s good news.
Um, we have to land this plane and, and I want to, we talked a lot about policy and that’s pretty critical obviously, um, because it drives a lot of other things. Talk to me about the technology.
Do you see medical AI coming in the next five years? Useful medical AI, whether that’s open or closed, I don’t know at this point. I prefer it to be open, but do you see a usable, good quality medical AI coming in the next five years?
Bart de Witte:
It’s already there, it’s not being applied yet.
So I think that there’s a problem with adoption and that problem is that physicians are aware that in the way how it’s implemented today that knowledge can be industrialized, their knowledge, which is something that they owned. And I sometimes make this comparison that if physicians don’t… support open source, it’s called speaking for my case, that they will go through the same fate as the pharmacist in the 19th century because the pharmacist in the 19th century owned the knowledge of creating drugs and they were the creators of that knowledge and then it turned into a big pharma.
Now the same thing can happen now with medical knowledge into a big tech. That’s why a lot of physicians are refusing to start playing in that system. That’s something I learned while being at IBM and seeing Watson Health fail.
Nobody talks about why Watson really failed. There were a lot of fuzzy arguments out there and a lot of wrong stories, but most of the resistance came out of the model that was applied. And I think I’ve been discussing with many, many physician associations, if you turn that into open source and you are a provider of a digital health solution. and you tell them that the model is open source and can be co-created with physicians, then you will see adoption.
And I’ve been testing this and similar for me, if I prefer Signal over WhatsApp, then I go for Signal because it’s open source. In healthcare that even counts more, because it defines if physicians still have control over the knowledge or if the industry is going to have control over the knowledge.
That is a power play, that’s pure game theory. And we should apply more of these game theoretic models in that Game of Thrones to understand how we can then accelerate the adoption of AI. Because there are so many good cases out there, they’re not being applied just because of the power plays behind the screens.
John Koetsier:
I hope that we’ll be able to cut through those and we’ll be able to democratize the data in a safe way and a non-Big Brother way as well as make access to the technology because literally billions of people have their lives at stake. Many of them from treatable things, many of them that just maybe a little bit of knowledge would be what’s necessary and then maybe just a little bit of medicine as well.
And I wonder if we won’t see a bit of a divergence in the future in the role of a doctor. We already have many different species of doctor, a GP, a general practitioner, surgeons, and then all these different, throat, nose, podiatrist, all that stuff. Well, it might be the case that a very good medical AI could be almost a GP or at least the front line and then other doctors can come in and see and apply their expertise. I don’t know. We’ll see how that goes.
Bart, I want to thank you for this time. It has been a lot of fun.
Bart de Witte:
Thank you John and let’s hope for a better future and democratize everything that is related to life saving knowledge.
TechFirst is about smart matter … drones, AI, robots, and other cutting-edge tech
Made it all the way down here? Wow!
The TechFirst with John Koetsier podcast is about tech that is changing the world, including wearable tech, and innovators who are shaping the future. Guests include former Apple CEO John Scully. The head of Facebook gaming. Amazon’s head of robotics. GitHub’s CTO. Twitter’s chief information security officer, and much more. Scientists inventing smart contact lenses. Startup entrepreneurs. Google executives. Former Microsoft CTO Nathan Myhrvold. And much, much more.
Subscribe to my YouTube channel, and connect on your podcast platform of choice: