'How to Stop Automating Broken Healthcare Systems' with Leo Anthony Celi

AI is not failing because of bad technology. It is failing because of bad questions.

Dr. Leo Anthony Celi, critical care physician and researcher at MIT, has spent years exposing a problem that the machine learning community would rather not face. The datasets training our most powerful healthcare AI do not just underrepresent certain populations. They reflect, and quietly reinforce the inequalities already baked into our systems. Building bigger models on broken foundations does not fix medicine, it scales the dysfunction.

In this conversation with Jean Gomes and Scott Allender, Celi makes a counterintuitive case; that AI could become a Trojan horse for genuine transformation, not because the technology will save us but because confronting its failures might force us to ask better questions about who gets seen, who gets studied, and who gets to shape the tools that increasingly govern our lives.

Dr Celi shares why he is bringing musicians, indigenous philosophers, and religious communities into rooms usually reserved for data scientists, what a basketball game and a tangled net reveal about who actually solves hard problems, and why the four qualities most needed from AI are the same ones most needed from leaders: creativity, humility, curiosity, and a healthy scepticism toward easy answers.

For leaders navigating the AI era, this is a conversation about what it really means to build systems worthy of human trust.

Reading from Jean Gomes and Scott Allender: 

Leading In A Non-Linear World (J Gomes, Wiley, 2023) 
 The Enneagram of Emotional Intelligence (S Allender, Baker Books, 2023)

Social:

Instagram           @mindseteconomypodcast
LinkedIn             The Mindset Economy Podcast
Bluesky            @mindseteconomy.bsky.social
YouTube           @TheMindsetEconomy
 
The Mindset Economy Podcast is researched, written and presented by Jean Gomes and Scott Allender with production by Phil Kerby. It is an Outside Consulting Ltd production.

Transcript

0:00

What we think AI should be doing is enhancing our humaneness, enhancing our virtues. Welcome to the Mindset Economy. I'm Scott Allender, and today John is in conversation with Dr. Leo Anthony Celly, a critical care physician who has spent years at the intersection of medicine, machine learning, and what he calls the brokenness of our systems. What makes this conversation unusual is where it leads. Leo starts with the provocation: AI, as currently built, is a tool of the powerful. The data sets that train these models are riddled with absences, and those absences aren't random; they reflect who gets healthcare, who gets studied, who gets seen. And yet, the machine learning community keeps building bigger, not better, to reflect the current demographic and economic status quo, but this isn't a doom-laden episode. Leo has a counterintuitive thesis. He believes AI could become what he calls a Trojan horse for genuine transformation, not because the technology will fix us, but because confronting its failures might. He's calling for AI that nudges us toward creativity, humility, curiosity, and less gullibility. He wants to bring indigenous philosophy, religious community, and musicians into the room where these tools are designed, and along the way he shares a memorable story about a basketball game, a tangled net, and why the people who created a problem are rarely the ones who solve it. This one will stay with you. So, you've described AI as this kind of Trojan horse that forces us to confront the brokenness of our systems. What's the most important shift do you think we need to make, so that we don't just automate that dysfunction so that it elevates us instead. Yes, I think that the focus on the what question might be what is going to prevent us from leveraging this very powerful technology, and as I have mentioned in the past, we are focusing now on the why, the who, and the how, because we think that if we at least get some answers to the why, the who, and the how, the what becomes clearer, and the Trojan Horse part I think is what makes us optimistic, because if you look at the state of AI at present, where the control and the power are really concentrated in the hands of a few, we think that AI is just going to become one of the tools with which those in power are going to preserve their status, so I think that if we invest a little bit more, not on the models, not on the platforms, not on the technology, and invest even greater time and funding to the ecosystem, primarily the people. I think we have a relatively good chance that this is going to end well for us. You've been working with historians to think about the precedence in technology adoption. What are the key takeaways that you've gained from those conversations? One statement that's really struck me from a historian is that one thing we learn from history is we never learn from history, and there's another saying that there is really nothing new under the sun, whatever we're complaining about, someone else has complained about last year or last decade or centuries ago, but are they, are we in a different position now? Because if we accept that the only thing that we could learn from history is, we don't learn from history, then we're doomed. But also, if you believe that it means that we learn something from history, if you act, so there is the, it's not really circular fallacy that if that is really correct, then it means that there's a way forward. So we have been looking at the history of the printing press, the history of the steam engine, and how that disrupted operations processes way of thinking at that time, and what we're hoping to do is to identify. Identify the things that work and really be mindful of the things did not, that did not work. What we realised that there are key differences between the steam engine and the printing press at one end, and then AI one is that the form and function of the printing press as the steam engine did not really evolve, whereas with AI, as we know, it's perpetually shifting its form and function, which makes it very unpredictable, which makes it harder to be two steps ahead, because we just don't know where this revolution is going to take us. When Chat GPT was introduced to the world in November of 2022 we had no idea that a little over three years later we will have all of healthcare disrupted, as we know. Open AI just released a lot of tools that are supposed to make this general purpose AI very useful as a health guide, and then within a few days, Claude and Anthropic did the same, I think, in time for the JP Morgan conference in San Francisco. We have to understand that no matter how good is the fine tuning of general purpose AI, it will never be good enough, or it will never be safe enough, not even good enough, safe enough, because for the most part the data sets that we use to train these models do not capture health and disease in the world as it exists. We all know that what is captured in mainstream media, and even what is captured in academic publications, is just a sliver, a sliver of the reality that we are hoping to use to choose the models that will improve health outcomes. There's so much missing data when you really dissect what is being utilised by companies in creating this general purpose models, and what is missing in the missing data, more importantly, that than just which patients are represented, is truly what we refer to as the provenance of the data, the context of the data, and the way we are building models now is a great example of the lack of critical thinking, where there's no attempt of truly asking the questions, why are some data missing, and missing data is not just the empty cells or whatever corpus of text that is not representing certain patient populations, it's completely almost diverting our attention away from understanding why they are missing, so the current attempt of the machine learning community to impute missing data to increase the weight of the data that is sparsely representing the diseases that we are hoping to improve, it's I think the wrong way of looking at things, because again, you cannot derive information about missing data on the data that is present in the same way you cannot impute predictions and classifications of people who have no access to health care by training them on data that we have, because the data that we have does not represent those who did not make it into the data set, we want the community to be thinking more about these things, and every time I go to a machine learning conference, or even a medical conference, it feels so frustrating because it sounds like a broken record being played with each conference, they're all talking about bigger data that they have, they say that, oh, we have representation. From minority patients, but it's not enough. The way data is being captured from the minority patients, it's tinged with structural inequalities that are not easily seen when you just throw that data to perform some exploratory data analysis, so we are at the beginning of the journey. Unfortunately, there's a big push to get this into the hands of everyone, and when we talk about healthcare AI data or health AI data, we are not just talking about AI that's being deployed in clinics and hospitals. The AI that will harm the most when it comes to healthcare is the AI that is represented by general purpose AI. So, the question to us in the healthcare community is, what is our role in overseeing general purpose AI? Because at present there is not much oversight of this general purpose of AI, a general purpose AI, and the reason for that is they're not framed, they are not branded as health AI. So, in the United States, they are not under the purview of the FDA. They are under the purview of the FCC. Of course, there are regulations, like you cannot advertise a product with false promises, but I think that that may not be enough to curtail any harm that could arise from the use of general purpose AI, even though they are fine tuned on well curated corpus. To me, I would push back, because your well curated corpus comes from research papers that were performed in high income countries with a focus on a majoritized demographic, so even fine tuning your general purpose AI on papers from PubMed is no guarantee, so as you could see, there's a lot that we need to do in the machine learning community, and unfortunately, none of these is being taught. We had a survey of machine learning courses, and very few - maybe I could count with my fingers - really spend enough time teaching the students about the provenance of the data, and the reason for that is because the professors themselves do not understand the provenance of the data, and I think it will be unrealistic to expect machine learning professors to understand the provenance of the data. This is why the way to teach this is to bring together faculty from across universities, but again we could talk for an entire day of why it is almost impossible to do that in the way universities and schools are set up. So I'll pause there, because I think that there are so many topics there that we can dive into. Well, there's a lot, and you caught our attention with this idea that we should be building humbler AI with the capacity for doubt, and we're really interested in this idea of being able to harness metacognitive capabilities in people, so that they're better understanding risk and uncertainty. You've written about the four qualities that you feel are critical to overcome the flaws that sit at the heart of today's systems, can you expand on those for us? Yes, what we have been saying to in the past is that AI cannot just be doing more of the same, but faster, and maybe a tad better, because if it's just more of the same, but just faster, it's not going to address any of the issues that we're hoping to really head on tackle, so the way that we are developing models now is that the main function of AI is information retrieval, knowledge retrieval, and that to us is not a full use of this technology, so what we think AI should be doing is enhancing our humanness, enhancing our virtues, because I think that a lot of the systems are flawed now because they were designed by humans, but maybe we could deliberately design it, design AI, so that. What makes us thrive as humans are facilitated or enabled, so we want AI not just as an information retrieval tool, but a nudger, a coach to advise us to be more creative, more humble, more curious, and less gullible. So, those are the four attributes that we're hoping to design into what we refer to as multi AI, multi human systems. So, we're very much critical of an individual working with her or his or their AI, because in all likelihood that interaction is just information retrieval. If we want humans to be more creative, that means that we have to surround that human with other humans who think differently, because creativity is something that's very difficult to do right now with the way we develop models, and I would often reference move number 34 when we talk about imbibing creativity into human AI systems, so if you recall, move number 34 was the move that beat Lee Sedol by AlphaGo, and when that move was made, all the humans who were watching the game were astounded. They've never seen that before, and some thought that that could be the end for AlphaGo, but it ended up being the winning move of AlphaGo, and that's because AlphaGo had access to the millions, if not billions, of possibilities of Go moves, and no one has had the ability to connect all of them with AI, what we're saying is that we are less likely to be creative if we're training AI within the same corpus of digitised text. What we need to do is we need to interface different bodies of knowledge writings from millennia of wisdom coming from different religions, different civilizations, different indigenous communities, and the advantage of using this as training corpus is that there is moral underpinning in the way those corpus was put together, and I think no one would argue that there is no moral underpinning in what we find in the internet, and that is the reason why we're finding all the the the risks of harm from the models that are being trained on internet data, so it's very important that AI is not just using the data that the creators see, because that's what's happening now, and that the data that the creators see, the ones that are very easily seen and extracted will be data from the internet, but we need to push the developers to look for data under the rocks, in the depths of the ocean, up high in the horizon, and this are data again coming from religious communities, indigenous communities. We have been trying to adopt indigenous philosophy in the way we develop and validate health AI, because the indigenous people have this view of health being relational, our relationship with the people around us, our relationship with the flora and the fauna, our relationship with the land, the water, the air, and that is something that should be the fulcrum of the way we create tools for healthcare. On that note, we are having an event tomorrow, and we're very excited. We're hosting our colleagues from the Vatican. We have a Jewish rabbi, we have monks attending, and we invited indigenous communities, and not only that, we also recruited musicians, so we have Berkeley College of Music here in Boston. We invited faculty and students from there, and the idea here is, can we brainstorm together? If we have a clean slate, what would learning look like? Because learning is so much different if you have tools in your fingertips that could access deep expertise, and for that reason, what would be our new role? When you hear the phrase humans in the loop, it's not about arbitration. Of the ground truth, I think humans and humans in the loop are fighting for their survival. What is it that would make give purpose to our lives? Because that is increasingly in question. I talk to medical students, I talk to computer science students, and I'm telling them that you have to reflect what are the things that you do on a day-to-day basis can be replaced, and if the bulk of that can be replaced, then chances are you will be disposed of, because AI does not need health insurance, AI does not get sick, AI does not get hangover, so of course, in the name of progress and cost cutting, they're going to replace you with AI. And then we've started thinking ahead, because it's very important for us to be two steps ahead. We cannot be reactionary, and I think all the governance frameworks and all the even ethical discussions are for the most part reactionary. They don't do a lot of mapping what are possible scenarios and come up with guardrails depending on what scenario we stumble upon. So going back to your your question, well, I guess the question that we're talking about here is humbler AI, and you've outlined a framework for what that looks like. Silicon Valley doesn't necessarily think in these terms, they think in a very linear mathematical way around building AI models. So, I'm interested in the conversations that you've had with the people who are building the systems and how are they reacting to what you're saying to them, so there are two extremes of reception there's the positive reception and gratefulness that we are surfacing these issues because they have been so focused on their key performance indicators, they're so focused on what will bring or what will increase shareholder values, and they have not been given opportunities to reflect what is it that they want to do. Why are they doing this? But I've gotten very negative pushbacks, I've gotten reactions that this philosophising is not really realistic, because their day-to-day, their ability to be able to put food on the table depends on whether they impress their managers or not, whether they impress their investors or not, and we have been struggling in trying to convince them that because there's a pushback, they say that there is no time to pause, there is no time to slow down, because it's unfolding, and doing that will put them at a disadvantage, doing that will maybe lead them to lose their jobs, so yes, they understand that we have this noble intentions, but they call it quixotic, this is not just what is practical at this time, and we have been talking about how do we reach people who automatically put us on mute every time they start hearing about ethics and the bias of the data and the provenance and what is missing in the data. Those are trigger words for them, and they don't want to engage with you anymore. So, initially we were saying that, how do you communicate with those who don't want to hear what you have to say, and we've realised that communicating is not the right word. The correct word is engaging, and maybe what we need to do is not to lay out everything that we want them to be aware of, but our plan is just to warm up the room by a degree or two degrees, so that the next time they hear this they're a little bit more open, but I'm learning so much as I represent our group, talking about the research that we're doing, because I get exposed to a whole spectrum of feedback, but I am buoyed up by the positive reception, I am buoyed up when I hear my words being spoken by the students that I work with, like, yes, yes, I have succeeded, and that, to me, are the wins that truly make us optimistic that there's a way forward, given how you see the world. If you were starting out again. Know if you were 23 or 24 you've just finished your masters or your PhD, and you're thinking about the challenges of automation and how the workplace is going to evolve in the next 20 or 30 years. What advice would you give to yourself from what you know now? What would you do differently? I think I would engage more with people who don't think like me, and of course that's very hard to do, because you're very focused on I have to finish my degree, I have to finish this course, and anything that I do outside of that is probably going to slow me down to what I consider as a sign of success, and of course it doesn't help if the whole system is also doubling down on that way of thinking, but I probably, or maybe not, I would have been more of an activist, a change agent, which I'm just discovering now, but at the same time I'm able to do this because of the stature that I've attained. So, who's going to listen to me if I were a budding doctor or researcher? So, somehow you need to be able to earn this gravitas, this authority to be able to make these statements, but I don't, I think it's never too young to start, and this is why we spend so much time in bringing with us the young people. One of my advice to deans of universities that I've been speaking with is to replace half of their current deans with young people. This is also my advice with health ministers that the people who are invited to seat at the table with decision making powers that should be shared with young people and I tell them that it's impossible that we ourselves in the system will be able to come up with solutions. There's this video that I saw, I think, last year or two years ago, and it was a video of a basketball game, a men's basketball game that got interrupted because the net got entangled, and they stopped the game, and each man started trying to out jump each other to disentangle the net, but they all failed, so they ended up saying that they're going to continue at another time. Within one minute, the women's cheering squad came in, formed a pyramid, and disentangled the net. And I thought that there are three lessons there. The first is competition almost always kills collaboration, which is what is needed to address complex problems. So, the men were all out chomping each other in machine learning for healthcare. You hear the words, my data is bigger than yours, my model is deeper than yours. I trained on 1 trillion parameters, and I don't think that this is the mindset that will allow us to benefit from this technology. The second lesson is that you won't solve problems if you're surrounded by people who think like you. So, the men never thought of different ways of disentangling the net, and I've been critical of labs where all the members are computer science students, and the professor is a computer science professor. You won't be able to solve any problems, because you're going to be seeing the problem, the issue from the same angle. You're like one of the six blind men trying to describe an elephant, and all they're describing is the part of the elephant that they could feel. And then the most important, I think, lesson from that video is that the solutions will not come from the people who created the problem, and that's what's happening now. I keep telling my colleagues we cannot solve these problems by ourselves, because we have doubled down on the problems, we need fresh pairs of eyes. Unfortunately, there's resistance to bringing in people who have not earned their keep. Why would we share our power and control with students who did not go through decades and decades of training getting into the top elite Ivy League institutions? That is just unfair. It is downplaying, discounting my value here in society, and I keep saying that perfect is not the enemy of the good, it's hubris, and it's our entitlement to what we think is our merit, and once again we start talking about the merit. The myth of meritocracy, that a lot of it is really your access to privileges, your access to different cultures, that got us to where we are, and that is something that is not equally distributed around the world. So, all the things you're talking about here are quite economically and politically charged, and you're demonstrating a considerable amount of leadership and courage in talking about them. How does that feel in terms of the personal risk that you might be taking in your position or your to your reputation? Because there are a lot of vested interests in keeping the status quo the way it is. So I just wanted to say that I'm not against capitalism, and I have been branded as Marxist for all these ideas that I'm spouting. So, capitalism, at its very core, is great if truly what is generating market value are values that are aligned with society's values. so in theory, capitalism is great if it's operating the way it's supposed to operate, and I always get pushback that well, socialism is not good either, or communism is not good, and I say that the ideologies themselves are theoretically all good, but it's the people who hijack those ideologies that are the reason why these ideologies don't translate into what we thought would be delivered to us if we adopted those ideologies, and it's hard to argue against that right now. They see they step back because the problem is people will hear what they want to hear, and the other thing that is worth noting is what Maya Angelou said, that again they don't remember what you say, they only remember how they felt when they listened to you. So I think personally I have grown in the way I communicate. Again, I'm not trying to communicate, I'm trying to engage. All I'm doing here is warming up the room and trying to see how could I engage with this person? What would make him realise that I am on his or her side in the end, that I want him to keep his job and at the same time also contribute to public good? It's not either or, it could be, and, and we're here to both explore how could it be and rather than or, because one time, as I said, there was a pushback that I got that this is not consistent with what's happening in the world, but I'm trying to, to push for, he said that I did not hear any concrete steps from you, and I said I did give a lot of concrete steps, which is bring in philosophers and social scientists into the team that is making developing AI schedule some times of reflection, but he did not hear all of that, and he said that those are not really concrete steps, because our managers won't buy into that, so and then he kept saying that his do, he's got a good heart, he's doing this because his wife is a cancer survivor, and right there you realise that this guy is having internal conflicts, because the fact that he's trying to really emphasise that he is here to support people who are suffering from cancer. So, how could I tap on that? How can I use that to enhance his sense of purpose? So, and this is the reason why we, we deliberately talk to people who might be offering some advice on how to navigate, tackle these very difficult conversations, and to me, machine learning conferences, clinical conferences, medical conferences should put these discussions at the front and centre. I don't want to hear that question again. Will steroids improve sepsis? Because every year you hear that is six cc's per mil better than eight cc's per mil when we're trying to ventilate someone like we need to, and of course those questions are important. Don't, don't get me wrong, but at the same time, we should also be talking about philosophical questions. What is the identity of a doctor now, especially in the United States, when we know that nurse practitioners and physician assistants can perform a lot of the procedures? Initially, we are performed by doctors, so I believe 70% of anaesthesia is delivered now by a nurse anaesthetist in the United States, and now AI comes in. AI is able to better diagnose better in terms of putting together a treatment plan. So, let me ask you, as you are just entering the medical profession, what do you think your role is going to be? And this is where the indigenous philosophy would come in. Maybe our role, if we're truly interested in improving the health outcomes of society, maybe our role is to connect our patients with the people around them, connect them with the planet, but of course medical schools would cringe, like we are not in the business of improving them as humans, we're in the business of teaching them how to take care of people. Once again, it's not either or, it could be, and so I have been pushing for medical schools to rethink what the curriculum should be. If it were up to me, I would say that half of medical schools should be immersing the students with people who don't think like them, students of humanities, students of computer science, students of engineers, students from nursing, students from pharmacy, but no, we wait for them to graduate, and then we start expecting them to very seamlessly work with this, all these members of the of the medical team, I would say that half of medical school should be on in quality improvement projects, give them unsolved problems rather than giving them homework on some questions on some OSCE that AI could probably easily pass, then because it means that that skills would be good for tasks that can be delegated to AI, right? So if we are going to truly train a generation that is able to leverage the power of AI, it should be teaching them and and testing them on things that cannot be done by AI, but I think that what we have now as part of curriculum is not really geared towards developing that kind of critical thinking, and there's another thing that we have discovered is that you cannot teach critical thinking in a room full of medical students, in the same way that you cannot teach critical thinking in a room full of C-suite executives, because they all think in the same way. Critical thinking cannot be taught, it has to be discovered, it has to be experienced, and the way to do that is to break down the silos within the universities, break down the silos across different sectors of society. Another recent take away that I truly, that truly resonated with me. I was having a conversation with economists and anthropologists, and one said that why are people talking about health policy? Every policy is a health policy. Every policy in education, every policy in transportation they all affect health, which means that the different ministers, the different cabinets, they should be aligning how they're doing things and not try to address their specific domain with their own tools, so what we're truly, what we truly need is a level of coordination and collaboration, which we've never seen before. And how do we get there? I don't know, but I think that this is the tectonic shift that we want in the end is a reimagining of how we learn, reimagining of how we think, reimagining of how we engage with each other, so you're talking there in the context that AI might liberate us to be able to do some of those things, because the dividend really is the ability to move to a more systemic level, more creative, more critical thinking, and so on, to be able to tackle things in that way, which at the moment we struggle with, because we're consumed with the silos and the minutiae busy work. So I'm a leader, and I've got several teams reporting to me, how should I be thinking about AI's role in the culture and the processes of these teams, so that I can take us on that productive path. Let me think about that. I was, I had some thoughts at the beginning, and then suddenly it got replaced by something else. So, how do we make AI develop other parts of the brain, right? Because what AI is replacing right now is our vernique, our comprehension, and our broker, our speech, but there are parts of the cerebral cortex that are contributing to intelligence, so even our definition of intelligence, I think, needs to be questioned right now. And we have introduced this construct of what we refer to as AI plasticity, that at some point, hopefully in the near future, we're going to realise that healthcare problems will not be fixed by AI, because the problems of healthcare are not informational problems, but in experiencing the failures of AI, in experience, in going through the process of developing AI, and being exposed to the depth and the breadth of the brokenness of the systems, we ourselves get transformed. It's a leap of faith. It's my half glass, it's my glass half full, but I feel it, as I said, I think maybe an instance or an example of what that might look like. I mean, the way we are even talking about the problems now to me somehow can be attributed to AI that we have never - I've never been in such rich conversations, and even the people that I talk to, who are economists and politicians, it's the same way - they feel that they have been more indul involved with much no one's conversations, and this is the plasticity that I'm referring to, we just need to scale this plasticity, and hopefully it, it is contagious. So, what we are hoping to do is, as we continue to question what is the impact of AI, as we continue to iterate on the design of AI, as we continue to witness that AI is not going to fix any problem. Our brains are transformed, our hearts are transformed, our souls are transformed. Suddenly solutions, real solutions materialise before our eyes, conjures before us. But how do we get there? How do we facilitate this? So we have been talking to behavioural scientists, we have been talking to psychologists and psychiatrists, and even cognitive scientists. How do we make sure that this plasticity that we are describing now truly becomes part of the journey, because I will be less worried that, or maybe no, that's not fair to say, because we are worried the faster we're advancing in this field, the faster we are melting icebergs, so one question that always arises is that what improvements would we say would be enough to say that it's worth the price, because the price is the environmental impact, the price is the extractivist nature that is widening the geopolitical divide. The price is what else, the workforce displacement, the hollowing out of the middle class. So we have been seeing the replacement of entry-level white collar knowledge, colour white collar knowledge jobs being replaced by AI, and we also know that if you wipe out the white collar jobs, you're going to be wiping out the blue collar jobs, because the blue collar jobs are dependent on the income that is being generated by collar jobs, so what we're going to be witnessing is the furthering of the wealth inequality. This has been described in Thomas Pickering's book Capital, where if the, the gains, the gains from innovations outpace the GDP of a country, so when the gains are not really distributed through all of society, what you're gonna see is the widening of wealth inequality, and we think that it's going to get so much worse with AI, but again, these discussions remind you that there's plasticity going on, there's no way I would be coding philosophers and economies in my former life, and it's almost like AI gave me a forcing function that this is the only way to be able to advance in this field. Welcome to the Mindset Economy, a place where we explore the human advantage. Age in a world where machines can think, so your job has become more interesting and quite different. It's pivoted quite significantly from the typical neuroscientist, and that's really interesting for the wider thesis of this show, because it's increasingly that our value, our economic value, will not just be about traditional skills and knowledge that we've built over the course of the past decade or two. It's about the mindset and the ability to enable our plastic brains to adapt to for more critical thinking, more creativity, and so on. I wonder, what you think that looks like for you over the next decade. Well, if I don't get assassinated, let's not hope I mean, I don't know if people have seen that movie Lucy with Scarlett Johansson, where they discovered a way of unleashing the human mind, and I really feel that, like, and I see that in the people I work with, you see them just thrive and flourish, and I think that there's no end. I always fear at night that what if, if I wake up tomorrow and I don't have any more new ideas, but then every morning with my first meeting we generate five six new ideas just within the first 10 minutes and I say maybe I shouldn't be concerned so this is a true renaissance that I'm hoping is not limited to a few people and I was watching the thinking game and feel free to cut it, cut this part of the podcast, and of course it's very impressive, but then you realise that these people who are who really believe that they are fixing the world, they actually don't understand the problems, and in our travel around the world, I have met so many people that I think are as smart as Demis Hassabis, and these are the people who actually live the problems, and I think the solutions will come from them, not from those who are at MIT, Google DeepMind at Stanford, because for the most part the people who got there won a lot of lottery tickets and prevented them from truly experiencing the problems, and I think the ability to solve the problems hinges on whether how, how you've internalised the problems, and that is, of course, shaped by our experiences. So, what we feel our role is is to redistribute opportunities, redistribute privileges, and access to the different types of economies, whether it's knowledge economy, whether it's mindset economy. How can we bring this to people who are just so distracted, but they don't even know that they are smart as Demis, they are smart as Sam Altman, because no one has given them that opportunity, so how do we convince elite universities that this is the way to make impact, and perhaps you will attract more endowment, perhaps you will attract more students if you invest in this and not invest on where am I in the rankings of universities based on the number of papers I've gotten more irritated by people who only think in terms of publications and citations, because I think that they're going to be wasting our time if that's their KPIs, if that's their north star, and unfortunately that's how the world operates now, unless we erect different incentive structures. One of the projects that we have been, that we've invested a long time ago is how do we replace this publish or perish culture? How do we replace this preoccupation with the H index with the impact factor, and one of the ideas that we had is to come up with not really alternatives but something to complement the H index, we called it Community Index. So, how are you contributing to building communities that will truly advance science that has impact? What is your contribution in terms of sharing your data sets, sharing your codes, sharing your educational materials? And it has been that paper. Has been rejected so many times by journals, and I think there's some conflict of interest there of why they may not want to publish this paper, so and the ideas for this again, I don't claim them, these are ideas coming from students that I meet around the world, so we've really been pushing for people to think outside the world that they know, but it's hard for them. They say that, oh, I still have to pass my course, I still have to submit my, my, my capstone project, and I say, well, that's fine, but somehow you need to set aside space, you need to set aside time in looking at what can be done differently. I keep telling students, I don't want to work with you if what you're proposing is what every other student is proposing, I want to see some crazy ideas out there, because I think that we don't want to contribute to the garbage, we don't want to contribute to the clutter, and I love what Ralph Waldo Emerson said, that don't take the path before you create new paths, create new trails, so that others might follow, so a lot of our projects now would start off as a paper, and I think the people who I've been working with for years would say, I know that the paper is not what your intent here, and they will say, I know that your intent here is to start a movement, start a revolution, create platforms, create tools, recruit more people, I mean, that's what science was all about before, but now science again has been translated into this false metrics, such as publications and citations, and we need to ask, what is our true intent? So we have another project looking at the publisher parish culture, and the very first paper that we found that used that phrase, publisher parish, was released in 1928 So we're now very close into 100 years of publisher parish, and then we pulled out all the reviews that really mentioned or treaty dissected publisher perish every decade, and every day decade it's the same complaints, it's the same lamentations. So our job is, how could we be different at this time? And one of the questions that arose during a recent meeting is, what is the root of publish or perish? Because most of the time we're trying to solve a problem, and in fact the problem is just a symptom, it's just a symptom of a cancer that is even more intricately part entrenched in the system, and unless you address that, all you're doing is just putting band aids, and I have.. I have that cancer. What's the fundamental problem? Oh, I think it's human nature is one of the fundamental problems. Our constant craving for validation, and we were saying that, and this is, of course, tied with hubris, right, publisher perish, hubris, they're all part of the same soup, and of course the initial intention of publisher of publication is good, it's, it's, it's noble, like you need some authority, you need some gravitas to be able to say something about science, and that can be measured by your publications and citations, but over time, again, as per usual, it gets hijacked by people who are very good at gaming the system because of their craving for validation, they're their craving for being the best, and this is a huge problem in medicine, right? The people that we attract and also the environment that they, we expose them to, it really almost invites you to be better than the others, and this is why we have invited the musicians, because in an orchestra, if you try to outshine the player next to you, you will completely destroy the orchestra, and their goal is to be able to listen to each other and to support each other all the time, it's not being better than anyone else, so I think, is there are we characters in a Shakespearean play where it's just, it's such a tragedy, because we always end up to saying that human nature is the root. Foot evil of all the system problems that we're seeing, but I still think that there's a way out. It's just not clear to me now, but the more I continue talking with people who don't think like me, I think it's going to start conjuring right before our eyes. I'm sure there's many people listening or watching the show are going to want to know more about what we've been talking about, so for the academics they know how to get hold of you and they know how to get hold of PubMed and find your work, but for people who are more in a management environment or in healthcare systems more widely, how can they get hold of your thinking? I am always waiting to be invited by people to give me an opportunity to warm the room by a degree or two. I was at a hymns meeting in Los Angeles, and I was so surprised by the positive reception, and I met a lot of people who have been thinking as well in this way, and these are, as you know, hims is very industry driven, but I think our job is to find other adults in the room, especially rooms that make decisions, because I know that they are out there, and we need to band together. We need to be smarter than the less than point 1% because they're always two steps ahead. I think I forgot. I think it's Church Hill who said that the people who love war are always smarter than the people who love peace, and we need to organise as well as those people, because right now they have the script, they're the script writers, and we need to wrestle the pen from them. We need to wrestle the laptop from them, and we need to write our story. What's your next piece of research focused on? We continue doing what we're doing, we're hoping to scale what we do, so the event that we are having tomorrow, where we bring in a ragtag of eclectic, we call them non-playable characters. They appear in the world, but they're not really contributing to the decisions about what to teach in medical school. We already have two planned in the works, one in Vancouver, Canada, in November, and then another one in Paris, France in December, and we're hoping to infiltrate professional societies, industry meetings with this kind of mindset. Well, Dr. Sally, we've come to the end of our time, and it's been a wonderful conversation. It's made me think on lots of different levels, and I think it's very clear listening to you about how you have evolved in recent years, that it feels like there's a transformation taking place in you, and that's fascinating. So, it would be wonderful to get you back at some point and talk more about how this has continued to evolve, because clearly it's a movement that you're trying to create here as much as anything else, and would love to see how that develops. So, thank you. Thank you for the opportunity. What strikes me most about this conversation is Leo's refusal to accept the false choice between idealism and practicality when he describes his approach, warming the room by a degree or two, there's no wide-eyed naivete. He's been strategic about how change actually happens in systems that resist it. There's a thread running through his work that connects directly to what we explore on the show, the recognition that our technical challenges are really human challenges dressed up in data and algorithms. The missing data in healthcare AI isn't a technical gap, it's a mirror reflecting who our systems were designed to serve and who they were designed to ignore Leo's four qualities for better AI: creativity, humility, curiosity, and scepticism toward easy answers aren't just specifications for machines, they're specifications for the mindsets we need to build if we want to shape where this goes, rather than be shaped by it. If this conversation sparks something for you, share it with someone wrestling with similar questions, and if you're interested in going deeper on how to develop the cognitive flexibility and critical thinking Leo describes, check out our Mindset Economy Manifesto. Until next time, keep questioning, keep adapting, and keep investing in who you're becoming, and we'll see you next time on The Mindset Economy, you.