Subscribe
Share
Share
Embed
🌍 Global Perspectives on Digital Health
A podcast unpacking the stories, insights, and innovation shaping health systems and underserved communities.
🎧 Listen on Apple, Spotify. Watch on YouTube
Shubs (00:00)
How is it that there is such a lack of representation of studies and research
led by people in low resource settings.
What are the barriers for people who are building, implementing, researching AI in global health? How are they linked to the kind of bigger challenges in global health and politics and power?
Does academia and publishing even matter anymore in the future?
Can we reconcile shifting attitudes towards evidence and expertise and knowledge and information gathering? Can we reconcile the decolonization of global health and how academia is part of that?
What does that mean for how we approach implementing and creating evidence that AI tools actually work in the field and how we talk about that as an industry. These are the types of things that we're going to be talking about in this episode. We are going to be joined by Peiling Yap and Rohit Malpani, who are editors of a new journal as part of the Royal Society of Tropical Medicine and Hygiene.
focused on AI in global health. We'll be talking about what gap this journal fills.
What is the future of academia? And so if you're building a digital tool in the global health space, if you're researching if you're implementing, if you're making policy for AI in global health, then you're not going to want to miss this one. you'll leave with a solid understanding.
of where academia and the field is headed in general and what this means for how we generate evidence,
and some bigger existential questions It's going to be a really great episode, so let's get into it.
Shubs (01:48)
Peiling Rohit It's such a privilege to have you on Global Perspectives on Digital Health. let's start with a little bit about yourselves, introduce yourselves, and then we can go into grilling you about the the journal and
Rohit (02:01)
Hmm.
Shubs (02:01)
kind of the bigger questions that we have. So Peiling, let's start with you.
Peiling Yap (02:06)
Thanks a lot, Shubs and thank you so much for having us on your podcast. It's our privilege and pleasure. just a quick note on my background. I'm a pharmacist and epidemiologist by training. I'm currently the chief scientist for Health AI, the Global Agency for Responsible AI and Health. So we look a lot with governments to capacitate them in terms of AI governance and AI regulations for health.
I'm also a board member of the African Population Health Research Center, ABHRC, and of course in my role today as the co editor in chief, together with Rohit, for the new journal Artificial Intelligence in Global Health, that was launched recently by RSCMH.
Shubs (02:51)
Thanks, Peiling. Rohit.
Rohit (02:53)
Well, first of all, thank you for having us here. It's really important for us to be able to talk about the journal and to get the message out. so my name is Roth Malpani. I'm a lawyer by training, which means I'm good at making up answers like a language model. but I have been working on artificial intelligence on the ethics-related aspects for about eight years now for the World Health Organization.
and I also do other work around access to medicines and I have a long history of working in civil society and somewhat in the private sector.
Shubs (03:28)
Super. so great
to have you both on the podcast. we'll we'll get a little bit into the journal, but maybe kind of just to frame it, it'll be great to understand the why of the journal. why why did this come about? There's lots of journals on global health, there's lots of journals on AI. and of course, kind of we framed this conversation as there are challenges in
representation of people who are leading research in low resource settings. Can you explain your experience of this or what you've observed in terms of AI implementation or research generally in global health in low resource settings?
Rohit (04:09)
Well, I can start maybe briefly about why the journal, and then maybe hand to Peiling about especially experience in low resource settings. so I think there's an institutional prerogative that's emerged, the Royal Society of Tropical Medicine and Hygiene, which sponsors the journal.
Already had two other journals that were receiving a lot of submissions related to the use of artificial intelligence. And to some extent, that made sense because you know this were sort of emerging from the ground up. They even had a special edition that they published about 18 months ago that was sort of trying to reflect this emergence of this research.
But there is a recognition that this was really becoming a much larger part of the landscape and sort of deserved perhaps its own place within their stable of journals. And so that was from their perspective institutionally why they wanted to do it. I think as you were alluding to, we are seeing obviously lots of journals in global health. We're seeing lots of journals that are dealing with AI, but of course, they are sort of, whether on the side of global health or on the side of AI, they sort of see these submissions that are trying to deal with both as exceptional.
And sort of perhaps as a special edition or as a one-off. And we thought it was important to be able to bring those together. But I think as Peiling can also talk about in more detail, to really show that this is occurring much more rapidly in low-middle-income countries and that people that are coming from these countries are starting to think about and are already developing a lot of research and implementation in this space.
Shubs (05:39)
Peiling, what's your perspective on this?
Peiling Yap (05:42)
Yeah, no, thanks Rohit for you know getting us started. I think and then I'll touch more now on the challenges, right, when it comes to AI implementation and research. I think obviously there are a lot of talent, let's say, in in these low resource settings for sure. what I'm seeing and we are seeing is really this kind of like wider ecosystem lack right? Be it the infrastructure, be it the governance architecture, and even still at
This moment, even the health workforce, the citizen kind of literacy, like general public literacy, right? I think those are factors that are kind of beyond the control of the researchers, but they do need them to kind of advance together with them so that eventually they can really implement and scale The other challenge that I see is really this bandwidth to be able to not just do research, but to be able to
publish and disseminate their results to the audience that matter, right? Be it to the public, to the policymakers, to the wider global audience. And I think this is really where I think Rohit and I definitely are committed to with this journal is to make sure that we invite or encourage really submissions from LMIC and really prioritize their presence in our journal.
And really try our best to reach out to various communities, various networks in LMIC settings that we we currently have and encourage our peers to do the same the other thing that we also then try and you know get around this whole implementation challenge is to really get that voice also from the private sector, from communities, from CSOs that are part of that implementation, but are usually let's say.
not well represented in scientific publishing. And then therefore really we have a section in our journal called Lessons from the Field, where we really try and encourage these submissions. So hopefully with the journal we can tackle some or or help to overcome some of these barriers, let's say, when it comes
Shubs (07:49)
Okay.
Peiling Yap (07:49)
to publishing work on AI implementation and research in LMIC.
Shubs (07:54)
so one of the the things I want to pick up on is what you mentioned about lessons from the field. in conferences I've been to in the past, we've led things called Fail Festivals. and we've had a call to action to the audience of people across
the global and digital health ecosystem of like we need to be able to share this. And part of what I see as lessons in the field is being able to share things that have gone wrong. And almost kind of going one step further sharing the things that gone wrong can be often more valuable than the things that have gone right. Is that part of what you're looking for in the journal as well?
Peiling Yap (08:27)
Yeah, exactly. I think we strongly want to encourage the submission and publication of negative results, right? The the crucial point here we want to emphasize is whether this whole study has been done regul rigorously, whether the scientific investigation has followed, you know, a very standardized, a systematic way of investigating and outlining the evidence. Because once you have that, then that makes it
It makes the negative findings as valuable as positive findings. Like you said, maybe even more valuable because people don't have to waste resources and repeat the same experiments. We also want to kind of sort of go against this trend that people think that, okay, what would our readership want to see, right? And what would people would like to see the sort of the glory, the the nice, beautiful, shiny.
objects, we want to kind of debunk that a little bit and and try to really push for, you know, again a focus and emphasis on scientific rigor rather than these kind of like outcomes that are perceived to be well received by our readership. So hopefully this will be again this is an experiment that we are doing and hopefully this will be also well received by the wider community.
Shubs (09:40)
Yeah, and I and I think there's there there is this I guess like bigger conversation on like yeah what do we value as people who are reading research, right? What do we value as an industry in terms of where like where we're trying to go? It's easy to be able to kind of pontificate in a podcast about, yeah, people should share their like negative findings. But you know, people who have
put a lot of effort and funding or investors who have who have said, hey, we're giving a small startup or a small research entity this grant to go and get this funding done. This is a lot there's a lot of pressure for someone to be like, okay, well, I even if the results don't look classically great, that I really need to put like this positive spin on this.
I guess like there's a lot of pressure on them to like not be able to do this thing that we want people to do. Rohit, do you have any thoughts on that?
Rohit (10:31)
Yeah, and I I think there are two broad concerns that that come from this pressure. One is that you know, we know right now in the AI field that there's a lot of concern around hype.
And so I think to the extent that we're constantly just getting sort of these, I don't want to say fantastical results, but ones that sort of say, well, we no longer need medical providers or that this will solve all of these infrastructure problems. Because look at the results we've gotten from this very limited set of data from a retrospective study. I think it continues to fuel that. And that sort of creates this vicious cycle then of each successive research team feeling that they have to sort of top the results that, you know, were were previously provided. So in a way, it's not trying to stop the hype cycle, but it is trying to
sort of create more
Just a a better ability to be able to mediate between the sort of hype and the risks. I think the other thing is that we don't want to lose sight of the importance of continuing to see the value in things that actually don't involve the use of AI. I think one of the biggest challenges that people in low and middle income countries are going to face, and especially policymakers, is not always going to an AI-based solution if it doesn't fit for their costs, for the health system in play. And so if all of the research keeps pointing in that one direction, I think it's going to be harder for them to make really good judgments about how to spend.
And scarce resources. And I think any journal has a real responsibility in helping policymakers make those decisions.
Shubs (11:53)
Hundred percent. You know, I think you know, I I I come back to this point a lot that we, you know, if you're really thinking about working backwards from our goals, and you think you then remind yourself that AI and particularly generative AI is one of a cornucopia of innovations, it reminds us there are there are maybe other more impactful and
Tools with better evidence base and maybe even cost effectiveness that don't need AI. I use the example of like just good service design is innovation, right? And so just reminding ourselves that not to start with a technology And that means we can be conscious and intentional about the tools that we use. We understand.
the consequences of outsourcing certain tasks with automation. And we can match the the technology to the actual goals that we're trying to have.
Rohit (12:49)
There was one other thing that I had forgotten to mention, which is this concern that one of the, I guess, powerful things of AI, of course, is that you you have these large data sets and you have your algorithm, and then you sort of can search for results that come from that that may not have been foreseen. There's a real danger in that, too. That essentially we're doing science by throwing as many things as we can at the wall and seeing what sticks. And and so again, with this sort of publication bias towards showing results, you don't want to encourage researchers to just be searching for something that emerges from the data.
in which, you know, again, we we don't know if it's correlation or causation. So again, it's it's trying to break this saying this is the question we were looking to answer. And it turns out it didn't work out and that's okay. It's not that we happen to find some other result that actually we never really were looking for in the first place.
Shubs (13:35)
Yeah, and I think we need funders to I guess like create the conditions to make that okay as well, right? So you guys are doing your bit as people in the editorial and distribution and making it visible process, as I guess like the output side of it, right? But in terms of the input side of it, in terms of people who are funding research and creating the conditions to say,
We want to fund good quality research. And it's okay if the finding is this didn't work. Like an example for me is something that came out of it was quite grounded, I think, that you know, the the Penda study, the using clinical decision support. And at least some of the findings were quite grounded, of like it didn't see the kind of big hype type of outcomes
But here were some kind of grounded findings there was value in this learning loop where people were then able to reflect on the gaps that they had. So there wasn't this like big operational cost saving that was like this big headline. And actually, like from what I've seen, people were people really liked that about that study, right? And so I think that's a good example of like it doesn't have to be
like our model is the great the greatest and best thing ever. we need we need better grounding on like, okay, here's what we're trying to explore. Wasn't it interesting that we didn't see that? But we've now found this other lever point or are this other interesting thing, and now we're gonna research that next, right? That like that seems like what we need more of.
yeah.
Peiling Yap (15:03)
I
think Shubs you just touch on the point where I wanted to go after from Rohit and it that's really like that socializing of this movement or or or push with stakeholders beyond the the journal itself, right? perhaps with other peer like minded journals or with the funders and really try and get them kind of used to this idea that it is okay to publish
results that were unexpected that were not delivering, let's say, you know, something that the funder is expecting, but it actually, like you say, has been grounded in true scientific methodology. that that's fine. It also reminds me of the time when we started when we were looking for funding at Health AI, right? To fund governance related work, regulation related work. And it was difficult because everybody wanted to invest in the shiny objects and the innovation.
and so it took years to turn the minds around and slowly we're still turning them and they are turned a bit now. But I think this push now initiated by our journal is hopefully something that will also lead to more sort of changes and and yeah, consensus as we go in the future. The other point I want to raise is really again, it's not just negative findings, but also implementation research and implementation findings. I think
I would be extremely happy if we can see really, you know, something that would excite me will be to receive an article that goes beyond describing how great the model is to actually how well it has been integrated or maybe how badly it is actually not able to integrate into the existing health systems, the clinical workflows, right? And and I think this is also something that we try and push for as well with the journal.
Shubs (16:48)
Peiling thanks for saying that. And you know, like it's not a plug or anything, but like I've just been gone through a being a guest editor on a special collection with NPJ Digital Medicine. this was focused on implementation low resource settings. And that I just wanted to say that resonated because that was like for me like really one of the filter points. Like, are you just is this talking about the performance of a model?
Or is this talking about the issues and challenges with implementations that we faced? and for me sometimes when you're kind of like, like, does this make sense? Is it relevant? And you kind of get lost in the weeds, like that to me became like a bit like a very, very clear filter to pass through. Like, this is what we care about, like this is what this was about. And it sounds like it was the same for you.
we've we've touched on it a little bit that kind of the academic world, particularly in terms of AI and digital health, has been very hyper-skewed to kind of like Western or so-called global north.
priorities and contexts. And what we think about with communities in low resource settings and part of the underrepresentation was thinking about impact and methodological robustness. I was talking to Shay Soromakun on the podcast, and she was talking a little bit about meeting communities where they are and kind of this difference between something being like
In terms of like what we think about in the so-called global north as like methodological soundness, statistical significance, and needing that as like one pillar, and then the second pillar being actually listening to communities and what they value. And sometimes that might bring those two into conflict because you know, in a lot of what we think of as low resource settings or communities.
Things that people value in terms of community
and storytelling and traditional medicine methods might be at odds with like the scientific method, right? And I would love to get your thoughts on that. How do you think about that? So we not only have a great methodologically sound paper, but it speaks to what matters to a community locally.
Rohit (19:01)
I think it's a great question. I I don't necessarily feel it's in conflict
because I think the types of research and research outputs we need are need to be more diverse. And so I think this also speaks to the types of research that the journal and others should be encouraging, which is that you know, this technology, and and of course we can look at social media before, have tremendous moral and ethical and cultural impacts wherever they're being implemented. And I think one of the the dangers that we saw from social.
media was an unwillingness to traffic and what that meant.
And I think what we're seeing now with artificial intelligence as a general matter or specifically in healthcare, that unless we are thinking about those questions with equivalent weight to sort of the, as you're noting, the traditional scientific questions, we're really not doing justice to what this will mean as it affects health systems, it affects patients and caregivers, and really the people sometimes who are making as much use of this as people are on sort of a scientific basis. And so I think that types of research will come from lessons in the field. It could frankly just come also from wider scientific.
studies that are actually seeking to measure this with more precision. And so yes, I think you're bringing up a current tension that exists, but I think it's it's kind of a false distinction because I think both of those are going to have to exist sort of as equal pillars.
Shubs (20:18)
I I'm glad to hear that. Peiling
Peiling Yap (20:21)
I I very much agree with Rohit that the two pillars, like you said, can exist, can coexist. and I think a few years back, I was part of a consortium with a group of, you know, wonderful researchers and we were working on this sort of idea of citiz citizen science, right? How do we encourage participation from citizens even in topics like modeling, something very abstract. And and so we did that.
sort of and that was for pandemic preparedness, right? So we did that in Brazil, Kenya and Vietnam. And to our surprise it was really encouraging to see that people want to talk about, you know, what impacted them during the pandemic, how it impacted them, and then kind of understanding also then if you were to build a model that supports decision making, how can their inputs and their insights actually contribute to the development of the model, right?
And and this kind of integration of perspectives that are non that are not part of the elite group, let's say elite researchers or scientists, I think that's extremely important. And I think we're seeing that hopefully more and more. I think we're seeing that trend sort of going up, at least hopefully get sustained because I also have seen examples of like citizen jury where you know people encourage citizens to really
not just participate, maybe even come up with research questions themselves, right? Because ultimately they are the ones benefiting or you know, suffering from the risk, right? So they should be the one kind of dictating what should be done and how it should be used.
Shubs (21:59)
And and I guess like that's
that's a driver of like changing the dynamics of like I guess like decolonizing global health as well. Like that's that that's the bottom up rather than the top down, right?
Rohit (22:10)
it's such a great point Peiling's making because the design aspects, of course, with artificial intelligence are very norm heavy. They're often not based on on really sort of any scientific method, and is really there where again the perspectives of people who are ultimately making use of this technology are going to matter. And it's sort of this idea that actually what we're seeing, I was reading in a a piece recently, we're not seeing AI becoming more super intelligent. We are seeing it becoming more super communicative.
And and so this idea then, therefore, that the the way in which these models are increasingly communicating as we're sort of making them more human-like and sort of prioritizing that has tremendous potential ethical risks as well as some potential health benefits. And to the extent that most of these models are trying to move in that direction, then the input of people who are actually going to use them and what impacts that has upon their use and their perception of the models will will become more important.
Shubs (23:02)
Yeah, absolutely. I think to me that's a really, really good segue into you've given us some breadcrumbs of like what you're looking for in the journal. but why don't we cover you tell us tell us more more specifics about the journal. I'm really, really curious to know, you know, who who is it aimed at? who are you looking like what are you looking for in terms of submissions?
Peiling Yap (23:23)
think I I can take the first part and indeed like who is who is it aimed for, right? you will notice that from our submission requirements, we are encouraging a lot of submissions from not just your typical researchers, but like what we mentioned previously private sector, CSO, healthcare professionals. I think another stakeholder group that
you know, we could potentially think about targeting is really also policymakers. I think for the longest time, when it comes to policymakers, scientific evidence could be quite challenging to access for them, or because of the way it's written, it's very difficult for their busy schedules to accommodate a reading of these publications, right? How can we make it accessible also for policymakers?
so as to encourage the translation of these findings into actual, you know, policy changes, into into policy implementation. I think that would be another stakeholder group that hopefully Rohit and I can can help to yeah to build.
Rohit (24:27)
Yes, in in terms of the intended audience, you know, I think I think our our focus is very much on building interest in low-middle income countries and and especially amongst researchers and scientists and policymakers. The architecture of AI, though, in the near term is still going to be very much dominated by the global north and of course then East Asia. And to the extent that we're talking about the need for resources to be diffused and to be shared and for decision making to be shared, it is going to be important that it also reaches those audiences. It's also also
This, the velocity of technology that's coming out of the global north. I think the other part to speak to paling in terms of the audience is alongside sort of the kind of traditional articles that you would see, it is very important from our perspective to have more policy papers, ones that are dealing with the ethics and governance challenges, as well as case reports and studies from the field, which really open up more space, hopefully, for both civil society and the private sector to provide more visibility, both in terms of the technologies that are coming through the pipeline, and then ultimately the types of
impacts that they're already starting to have in in real time.
Shubs (25:31)
Yeah, touching on the the last point just before we got into this, in terms of capturing what matters to people, right? to me that translates a big part of that translates as high quality, qualitative research, right? which itself is, I guess, historically hasn't been as valued, perhaps.
as others. How do you yeah, how do you see that?
Peiling Yap (25:54)
I think when it comes to implementation research, qualitative studies got to be featured. I I would think that, you know, I always think it's good to have the best of both worlds, right? A mixed method study design would always, I think, fare strongly, or stronger, let's say, than a pure sort of qualitative or pure quantitative. so so definitely I think implementation research-wise.
we would really look out for, like you say, qualitative findings. because that brings sort of and again, people like stories, right? That brings you the sort of the the anecdotes behind numbers. so definitely I think th this is something that we we will look out for as well.
Rohit (26:40)
And I I think one of the interesting headlines I saw like last month was the number of philosophers that a lot of the large AI companies are hiring, and to the extent to which a lot of their rule books or their model cards are being written by people who really think qualitatively and sometimes, of course, in much more sort of metaphysical ways. And and I think that's important to also translate that into the research. And and I think, you know, it's also going to be about having, as as Peiling's noted, not only mixed methods teams, but also just mixed teams in terms of the
types of skills that the teams are bringing in and also the geographic diversity and expertise. And and in my work with the WHO, that that's certainly been the strength of it is to the extent that you're bringing people with very diverse perspectives into to building out sort of what what how people should think about this use.
Shubs (27:27)
Peling, earlier you gave us a little insight into what types of submissions you'd really, really love to see that get you excited. Do you have any other examples?
Peiling Yap (27:38)
I think I would sort of follow up on Rohit's point, right? I would love to see something that is sort of really LMIC led for LMIC and not a typical setup where you know it's probably LMIC people contributing the data and the evidence and the research, but then it's led by, let's say, you know, HIC team that is kind of
maybe having more access to the writing and to the submission criteria requirements, right? I I think that that would be extremely heartening for me to see. And I hopefully with the sort of, you know, the the publication policies of our journal, with the waivers when you come from certain geographies, we hope that can can truly encourage that. And and also one other thing is is this like like the point that you made about
You know, it being truly grounded and taking time to synthesize the thoughts and the evidence. Everything is going so fast this day. Everybody just wants like, you know, even to read something and and I just get AI to generate like a one page summary for me, right? Like that kind of process, I think, you know, however it can be reflected in the paper, I think that would be really encouraging for me as well. Because the more we get AI to generate content, the more the less human.
content we have and the less sort of originality that comes from human thought process that we will have. So hopefully our journal can uncover some of these longer term, more grounded studies.
Shubs (29:14)
Perfect. and we will come back to this point that you've just talked about, Peiling. before we move on, anything else you want to add on that, Rohit?
Rohit (29:22)
Yeah, just just to kind of echo, I I think that the tendency again for a lot of the literature that's coming out is sort of what is the newest thing, what's the newest pilot? I think we're in this phase of pilotitis. I I'm more excited by something that has already been around for four or five years, for instance, you know, cervical cancer diagnosis, which is now sort of a tool that's been established, and seeing studies that actually look at how it's being implemented. So sometimes I think these sort of quote unquote older technologies are going to be, from my perspective, much more interesting. And then to reiterate
of the earlier points again things that are much more transversal in their thinking in terms of how these technologies either prospectively or already are having sort of these much more dramatic impacts in terms of how a health system operates in terms of the relationship between providers and patients so things that are trying to think systematically about these issues.
Shubs (30:12)
Yeah, a part of that to me is what are the lever points in terms of humans interacting with tools, workflows that go from this model works in you know, in silico, to people found it useful. It was used in a way that helped a workflow. And that workflow working well created actual.
clinical and operational outcomes and things actually improved for people. Like that's essentially what we're trying to do with good implementation, right? those are the building blocks So yeah. okay, why don't we zoom out a little bit then I really want to kind of go into some kind of bigger
I guess like meta topics around publishing.
We've kind of touched on the first one, we're in this we're in this world now, particularly after international development funding cuts, there's been a vacuum in funding. And this is on top of a global health system that has seen a shortfall of health workers increasing global burdens of disease and has already placed AI on a pedestal of like solving all of this, right? and then we had
USA and then other countries kind of pulling the rug for on funding various really impactful things within global health, including research, right? and part of this, part of the I guess like plan that we've seen develop over the last like I guess like 18 months has big investment from philanthropy into AI and AI research and implementation research.
and private entities have been part of this as well. do you see a world where you know the priorities of philanthropy, the priorities of big private companies who, particularly the big frontier AI labs who are now within and have potential to ha create big impact? how do we make sure that
That occurs and the funding is going in the right direction. but it's also going in the direction that governments and ministries and local communities and people in those communities want.
Rohit (32:18)
Well, I I think it's some it's it's a it's a big question, but an important one. I think in a way, part of the problem is I I don't want to put it just to that, there's now a loss of resources and a lack of capacity to respond. I I think the technology has moved fast enough that some of the institutions that generally are engaged in priority setting and setting up research agendas haven't really kept up.
And and I think that it also involves ministries of health and others that haven't necessarily worked together to identify those priorities. So I think there's a vacuum that exists that that you're noting that that doesn't solve the problem just because sort of a research agenda and priority setting is set out, but it definitely helps. And we certainly see that, for example, in pharmaceutical development. So there isn't necessarily coherence between what is missing and what will ultimately be developed, but you do see it having a certain signaling function. I think there's
Shubs (33:08)
Mm-hmm.
Rohit (33:09)
also just a a level of
transparency that whether it's the private funders or the philanthropies need to have in terms of you know what priorities they're setting, what the breakdown is, and how they're actually coming to those conclusions. Because I think with that then also comes the opportunity, including for people who then publish into different journals to sort of state whether or not that really makes sense from their perspective. So I think that sort of ability to have that dialogue or conversation doesn't yet exist.
But I think, you know, it's not just about high income versus low income. It's as you're saying between public and private. I think it's still a 90 to 10% distinction. And then we're not even bringing in the expertise and resources and data. So that's going to exist. And I think that is probably one of the most serious barriers to getting public health-driven research.
Shubs (33:56)
Any thoughts from you on that?
Peiling Yap (34:00)
Yeah, I I think how how to make research sort of locally relevant, right? there really I think we have touched on quite a lot of points. whether it's the design and and the the the initiation of the research by the local communities and then sort of the subsequent the the research itself and then also the subsequent dissemination in what kind of formats and what kind of
Yeah, in what kind of formats that would appeal or be more easily accessible to the local communities? I think these are very important key points. your point on the private sector, I think, they are publishing as well. But how do we encourage them to publish in a way that is not trying to cover some kind of bottom line that they have as a private sector?
But really publishing in the spirit of contributing to sort of real evidence and and and so being also accountable to the communities, right, through scientific publishing. yeah, so so I mean hopefully we can work on some of these topics or we can tackle some of these topics through our journal. but but yeah, very much agree that we need to start making also the evidence generation process more local, locally
Shubs (35:20)
Yep.
Peiling Yap (35:21)
relevant. Yeah.
Shubs (35:22)
And and then and there's two two aspects to this, because when we think of private, I guess like me, I fall into this trap as well. I'm thinking of like the big frontier models, right? And saying, how do we make sure that they do that? But private is also the smaller companies who are grown
Peiling Yap (35:35)
SMEs, yeah.
Shubs (35:37)
within within a country, and how do they get a equal opportunity? they have built local trust and local relevance. They've built specific
I guess like models that are relevant to that culture, the language, the cultural aspects. And so how do they get an equal opportunity? as equal opportunity as the funding that you've seen into I guess like Horizon 1000 with gates that is really focused on OpenAI or a partnership with Anthropic.
How do you make sure that there's also opportunity for growth for smaller private entities who are really trying to do the right thing and build local trust, right?
Peiling Yap (36:23)
Hundred percent, yeah.
Rohit (36:24)
I fully agree. This sort of bias towards the frontier models, the bias towards the thing that people see as the sort of pushing the envelope or as the most advanced model, I think is is it it continues to be something that we're going to have to deal with. I think this idea that something that's already been done in a high-income country setting no longer, therefore, has any relevance to report or discuss in a in another setting. So I think building more value in that. And I think there's just a function of being patient. So we are in the accelerationist phase of.
Of
AI. And I think a lot of the responsibility for ourselves and for others is to allow for this to grow at its own pace and speed in low and middle-income country settings. And I think unless we exhibit that patience and willingness to build that research and to invest in that research, then I think we're just going to repeat the same mistakes from previous eras of technological change.
Shubs (37:18)
Yeah, absolutely. We we do need the jagged frontier of progress, but we need the the rest of the things like basic good science, implementation, being human-centered, bringing people along, trust, ethics.
All of that needs to come. And we need to be okay that in certain places adoption will be slower. We just we're just okay with okay, the thing that will make the latest model work is actually getting the capacity right, actual transformation right, the human part right, right? And it's okay that that bit slower, because once we do that, that's like a foundation and we haven't like then built the future on like basically a house of cards, right?
Rohit (38:00)
Yeah, or
or developing something with a blank check in terms of your energy use and your water use and the amount of resources, the financial resources. There is real value, I'm not saying in like skimpy innovation, but basically that. Yeah, if you're going to build models that are sustainable or going to be built from the ground up it's going to be done differently. And and I think actually studying the changes or the differences and how that's going to be built should should be respected and should be featured.
Shubs (38:24)
Hundred percent.
Okay, so next question. let's talk a little bit about the future of publishing. I you know, I look at the challenges that academia is facing. We talked about is publishing even relevant anymore? People with many, many resources for marketing can just put out a a very flimsy white paper or a or a you know flimsy article out that's claiming all sorts of things.
we still need to care about robust research. but there are a lot of people who might be like, you know, I'm building a digital health tool, and I have so many competing priorities that I'm not gonna prioritize evidence generation or robust methodology because
I see all these other people investing in marketing and I'm gonna do that instead. do you see do you see a threat to publishing there? Like what's what's the future here?
Peiling Yap (39:19)
I think you touch upon I would say even a very philosophical question, right? Because ultimately it's what the society values. and what we have to continue to push for, however big our community is or however small our community, is to really to really push for scientific evidence and obviously then publishing that goes with it too, to disseminate evidence. Because I mean
And it's not that long time ago. I think it's was in the nineteen fifties. There was this the case of the Thalidomide right? Like before drug regulations and you know, clinical trials got sort of ingrained into our our minds or our system, the drugs were getting into the market without, you know, proper clinical trials. And a lot of that relied on lobbying, relied a lot on marketing, like you said, right? and then that tragedy happened, right, with the Thalidomide case.
Where you have lot of the babies born with defects and also, you know, resulting in death as well. We don't need that history to repeat itself. And I think going forward, especially with AI solutions, where we're still trying to build that trust, we're still unsure of how to actually generate evidence for it or its impact on our society. I think we definitely need to hold on tightly, I I would say, to to this sort of
tradition let's say of generating evidence, generating like impact assessments and stuff like that, to really understand the technology so that we can use it more sustainably sustainably and then we can trust it more as or use it more confidently going forward.
Shubs (40:57)
Yeah. Right.
Rohit (40:59)
Yeah, yeah, I mean, I would agree, although I I have a more pessimistic view, which is that the companies are just releasing the models as quickly as they can. They know
Shubs (41:07)
Yeah.
Rohit (41:07)
that they're going to be used in ways that they're they're even warning against using them, but they're not putting in the proper restrictions to prevent that from occurring. So I do see the value. I'm agreeing fully. But it's going to be for a future in which whether it's governments or regulatory agencies and others that finally decide to look at these as sort of, you know, products essentially that require some.
Some level or a lot of regulation to ensure that there's consistency between what the data shows and ultimately what people deserve. And so but I think we're in this period right now in which this push for market share is going to make it very hard for scientific publishing and the speed and deliberation with which it's done to sort of have its place in the sun. But I think in the meantime, we have to continue to produce this sort of information so that when they're ready, it it can be used.
Shubs (41:57)
And and I think another part of it. So so I think definitely one one driver is the the speed and people seeing this as a bottleneck. the other part is just I guess like story we touched on this earlier, like storytelling and scientific communication. what good marketing is, is great storytelling, right? and we know that science in general has to do a much better job of storytelling, right? and so to me.
I guess like the people who are who we might class as, for want of a better word, like the ones who are interpreted as being charlatans, they're using great storytelling. And actually what I guess like what we need is both good, solid scientific evidence and publishing. but then we're not done, right? We need to use that as like the bedrock of great storytelling. And that's like to me where like qualitative evidence and and storytelling to help
Almost if you're a if you're a company trying to lean into this in the right way, you're thinking about, I guess like your clinical evaluation plan, you're thinking about the evidence you generate as first, how do I want people to feel when they're reading this? And then how do I incorporate the the evidence that I need to make people feel that? And then how do I make sure that the evidence that I'm generating is then backing that up and supporting that, right?
people have very short attention spans, right? And so we need to hook them in with like a compelling story. And we're seeing it with like, you know, Nature Journal have now got a TikTok and people are now communicating science with like a paper behind them and they're green screening themselves and they're talking about that. Like to me, like you know, we're talking about the future of publishing. Like there's some something in there in terms of like combining, yeah, like solid methodological research.
With hooking people in with story and talking about what matters to them. I've talked a lot, but yeah, what are your thoughts on this?
Peiling Yap (43:54)
Shubs I would also argue that the reason why we fall into this trap where then, you know, it's all about the marketing is oftentimes the technology, they are developed because it's a cool thing. You know, it wasn't developed in with the with the mindset of like, okay, there is this need, and I'm addressing this, right? If the if the whole sort of the the design or the initiation of the development of the innovation has been done in a way that was like,
citizen centric or human-centric, right? Like actually understanding the system's need, the health systems, the the patient's needs, it wouldn't be that difficult to to find a story. There are plenty of patient
Shubs (44:31)
Totally.
Peiling Yap (44:32)
stories out there, right? So I I think and and maybe this is something that you know you know listeners or readers of your your work can can think about, right? It's I always tell my kids when you see a packaging in in the supermarket and it uses cartoons
Then you have to be careful of the nutritional value of that. Because probably it
Shubs (44:51)
Mm-hmm.
Peiling Yap (44:52)
contains no good. That's why it needs to hook you with something. And that's the same with the marketing. If it comes in a very glossy, very sort of like, you know, whoa, like shiny way, maybe take a pause and and really look behind the sort of the research that's gone into it. Because probably that's because there is a lack. That's why they need to put a facade in, right? So so I think this is also sort of awareness.
that we need to inculcate in in our public, in in our governments and everything, to sift through, to understand evidence and to understand, you know, where the right evidence are and yeah, how to look out for them.
Shubs (45:31)
Yeah, totally. That makes sense. let me paint a picture for you. If we are more and more using AI to write papers and do research, if the editorial process and the review process is using AI.
More and more. if then as a reader of a paper, instead of actually reading the paper, I'm just getting an AI summary of it, it's like AI all the way down. Like, is there any is that what we want? to me, the whole reason of doing that, the writing the paper, the doing the designing of the study, the the learning and the interpretation of the results, like that is the outcome, right? Like
what you learned through doing that. and I guess like what's ha what we potentially have a danger as of is having really I guess like tick box but superficial knowledge and experience because we didn't really understand what really happened. And because we didn't do the hard work on like actually writing the paper and designing the study. We didn't do the hard work on like the peer review because it wasn't reviewed properly.
Then there's like whole nother podcast on like hallucinated references, by the way. and then the whole reading of things, like we're just like building ourselves a house of cards as well. To me, that's some that's a big conversation around the future of publishing. Is there any point if that's like the way we're gonna do it? What how do you see this world?
Rohit (46:58)
Yeah, I mean, from my perspective, it comes back to some of the things we've talked about, which is this premium on speed, this premium on being at the frontier, the importance of sort of trying to one-up whoever has come before you. I think all of that sort of plays into the and then even potentially, you know, amongst journals themselves and others. I think all of that sort of plays into this churn of people feeling this pressure, perhaps, or this push to be able to use these types of solutions to be basically be able to produce and validate research as quickly.
as possible. So again, for me, a lot of it comes from first the cultural shift that has to occur before how people think about their relationship to a technology. I think that temptation is always going to exist, especially if you're in a place in which you feel you're in a race to basically develop to produce outputs much more quickly than the next person.
Shubs (47:48)
Pelling.
Peiling Yap (47:49)
I I think it goes back to the fundamental thing and and even questioning the role of AI in our society, right? Like what values do we place in our society as humans? How do we want to maintain the thing that made us that that is making us human, right?
But I think if we believe like certain values need to be kept, we just have to push for it in our own ways, right? Be it through, you know, for example the work that Rohit and I are doing in the journal, be it through the work that we're doing elsewhere. Be in the community, then we are interacting with our neighbors and and stuff like that. Right. So
it turns into a philosophical debate. Yeah.
Shubs (48:23)
100%.
And and if you're if you're watching this or listen to this and you're kind of feeling the same way, I highly, highly recommend you watch Ronnie Cheng's address to Harvard. I don't know if you guys have watched it. he uses this analogy. It's about 20 minutes long, and it goes into this like what do we place as value as humans? And he talks a lot about AI and people using AI as a shortcut to do things. and he says
You know, I gave a friend of mine a book called Buddhism Made Simple. And instead of like reading the book, he put it through AI and got a summary. And surprise, surprise, he was not enlightened, right? And it to me is an example of like the actual doing of th something or the actual doing something, making mistakes, writing is thinking, getting things wrong, going through that process is like much more important.
like than the output of a paper itself. To me, that's the corollary here of doing the research. And so I guess it's just like yeah, being intentional of like what out what's what tasks are we outsourcing? Particular and and also like we've got this like new potential body of evidence that we're building up, which is very I guess like flimsy, right? And we need to be careful w about that. A as a society
B as like a global and digital health ecosystem on like, yeah, is this really this future that we want? Like what how do we make sure that it's not just speed that we're thinking about, but quality as well.
Rohit (49:58)
I I think the point around science to me also is how many of our discoveries through history have been through serendipity. and this is something that AI removes. It's essentially this ability for people, as you say, to think, to be able to observe, to be able to write, and then to be able to formulate what actually comes beyond that.
Shubs (50:14)
Yeah, absolutely. we're almost coming to the end. I want to get a couple of top tips from you. Peiling, a lot of people who are listening to this are teams that are building tools and they're really thinking about, okay, how do I generate evidence in the right way? How do I think about outcomes? How do I think about like what my publishing strategy is or evidence generation strategy?
Do you have any tips for how they should approach it?
Peiling Yap (50:37)
Yeah, no definitely I think first of all, I mean, a few points here. there are a lot of existing guidelines. you know, for example, there's the tripod one, there's a consort, there's a spirit AI. These are all guidelines that are existing to promote sort of good research reporting practices, right? and and so it will also allow for the reporters to the authors to actually think through are they really
giving a very critical appraisal of the reported intervention and outcomes. So I would strongly encourage to like follow those reporting guidelines. And the other thing of course is to find like minded peers that are working in the same area. And here really we are quite fortunate at Health AI where we work, we are part of this group called the AI for Global Health. This is actually a research group funded by IDRC and also FCDO.
in Canada and the UK. And and it's it's really a group of researchers in in Latin America, Africa, East Asia, Southeast Asia, really doing very interesting implementation work. A lot of them may perhaps still in the pilot stages and everything, but looking to scale. So definitely I think reaching out to these groups, learning from each other what are the the the difficulties and challenges, I I think that would be really important.
Another point I want to make, which is related to your previous question shot, because it got me thinking about the future of publishing, right? It's really
Shubs (52:09)
Mm. Hmm.
Peiling Yap (52:12)
about the kind of evidence base we want to leave for our generations behind. I remember, you know, as a young researcher flipping through scientific articles dated like 1960s, 19, you know, early 19s, and being sort of
Amazed at how well the evidence can stand through time, right? Like can can stand through the evolution of of things So I think it's also us as researchers, as a scientist to really, you know, come back to ourselves again and say, you know, what kind of and evidence generation processes we want to put out there and what kind of evidence we want to leave for our generations, for our children.
so that when they themselves are doing research in the future they can come back and not just laugh at us for like, like you say it's a pile of cards like you know and then like what what are our you know, pioneers doing and and laugh about it and then yeah, so so I I think that's also another thing to to look out for.
Shubs (53:13)
Can you each give me one book or article about this topic on I guess like AI implementation, thinking about evidence, publishing that you recommend people should read? Rohit, let's start with you.
Rohit (53:26)
Well, I'll go with a book I am reading right now by a philosopher named Carissa Valise called Prophecy, which is sort of looking at the past, present, and future of prediction. so it talks about you know, essentially how we try to look at predictions as being the same as knowledge, but really it's a form of power and it's a form of trying to sort of have a form of social control. And she does a wonderful job of talking about historical uses of predictions by like the Delphi of Oracle and for example.
And the types of cultural assumptions that were made about that and the way in which societies were structured around those predictions, of course, to the present, in which we are increasingly relying upon artificial intelligence for predictions about our future and the consequences it has for autonomy and for the decisions we make. And so I think it's really instructive, especially in a lot of the examples it provides around medicine and scientific research, in terms of where the value comes from prediction and what the warnings are for its use.
Shubs (54:24)
how interesting. And there's something very meta about something talking about the future of prediction, by the way.
Rohit (54:30)
Yeah.
Shubs (54:33)
That that sounds really, really interesting. I'll I'll put I'll put a link on that in the in the show notes. How about how about yourself, Peiling
Peiling Yap (54:39)
I think one of the books that left a deep impression, at least I find it very relevant also to what I I'm doing, it's really the coming wave. I think it's Mustafa, I can't
Shubs (54:50)
Mr. Silliman,
Peiling Yap (54:50)
mean he called that he called this he used this term containment, right? for for AI and everything. But ultimately in sort of the global health world we have been talking about governance. and so I I think they're really
when we think about implementation, ultimately we always have to look at how we want to govern these things. So that book has been quite inspiring. Or yeah, it was quite good also to have him validate some of like our thoughts when it comes to governance. Another website I hope you will be able to share with your readership is really the website of the AI for global health community that I said. There we have like what we call AI for global health research hub.
we where we put in, you know, information about the projects that are being funded and also information about the various articles and that are relevant for AI implementation. So hopefully that will be a useful resource as well
Shubs (55:47)
Super.
Peiling Yap (55:47)
for for other researchers in LMIC.
Shubs (55:51)
this has been such a rich discussion. I pushed you hard on some existential questions. and it was really, really valuable to get your perspectives on this. you know, there are no right answers on this, the field is moving so fast. but it's really important to take the time to, I think, discuss these types of issues to ensure that we yeah, like we have a good grounding. And I really love what you said around.
You know, leaving a good evidence base for future generations and not having thinking, What the hell were you guys thinking? Right? in generations to come. Yeah, yeah. Yeah,
Peiling Yap (56:25)
We were just scrolling through TikTok and Instagram.
Shubs (56:28)
exactly. and yeah, and I think I think it was really, really helpful to hear about the journal and what types of things you're looking for. I'm particularly interested in kind of following up on the lessons learned part. That's that's that
particular area interesting area for me. So if you're working in underserved communities or low resource settings, particularly on AI implementation, this definitely looks like a good place for you to either read the submissions from or actually submission submit submit yourself. So Rohit Peiling, thank you so much
it was really, really valuable to get your thoughts here.
Peiling Yap (57:02)
Thanks a lot, Shubs
Rohit (57:02)
Thank you.
Peiling Yap (57:03)
for having us. Thank you so much.