Going viral: Can AI predict the next pandemic?

The next pandemic — it’s a question of when not if. Climate change is altering how and where diseases spread, and our insatiable love of travel means that viruses are now showing up in places they’ve never been before. Forecasting future outbreaks is becoming increasingly complex. But as infectious disease specialist Kamran Khan explains, this is where AI can help. Machine learning algorithms can detect patterns in data, model risk and project outcomes — and unlike humans they can work 24 hours a day. In this episode of Solve for X, host Manjula Selvarajah sits down with Khan to explore the connections between infectious disease and climate change — and how we can best harness the technology to help us prepare.

Subscribe to Solve for X

Kamran Khan: Humans are going to have to grow out of our infancy, and we’re going to have to actually be able to have that foresight and to think about risks that may be years or decades or even generations ahead of us.

Narration: I’m Manjula Selvarajah and this is Solve for X, a series where we explore the latest ideas in tech and science. We live in an age of ever expanding horizons. I think of my grandmother and how she never left the country where she was born, Sri Lanka, the small island off the coast of India. Meanwhile here I am, her granddaughter, and I’ve lived on three continents and travelled to over 20 countries.

How far we’re traveling is something that British epidemiologist David Bradley decided to look into more closely. Back in the 1980s, he mapped his own family’s movements, and he found something astonishing — each generation had travelled 10 times further than the one before. We have a seemingly insatiable appetite for travel.

If you look at July 6th of this year, it was the busiest day for commercial air traffic ever recorded. This increase in mobility comes with downsides — both for the planet and the spread of disease — and that’s the backdrop for the conversation you’re about to hear.

Kamran Khan: A thousand years ago, you could have had a newly emerging disease appear somewhere, but it didn’t get very far because we didn’t move a whole lot.

Narration: My guest today is Kamran Khan, he’s an infectious disease physician who was on the front lines of the SARS outbreak in Toronto in 2003. Nearly two decades later, he was one of the first people to sound the alarm on COVID. Kamran now spends his days trying to figure out where the next big outbreak might come from.

He runs a company called BlueDot that’s using artificial intelligence to track the spread of infectious diseases around the world. At the heart of what Kamran does lies an interesting tension. On one hand, it feels like we’re living in a world that’s increasingly unpredictable — and yet, we have technologies like AI that give us incredible power to pick out trends in data, model risk and project likely outcomes. But that challenge is getting more and more complicated. Things like climate change are shifting the patterns of how and where diseases spread. Soon we could see tropical diseases like malaria and dengue fever in Canada. There’s a lot to unpack here, and it’s a fascinating conversation. To start us off, I asked Kamran to take me back in time to when he first started working as an infectious disease physician.

Manjula Selvarajah: I understand that you actually started out, you know — some of your early days was doing SARS in Canada. Is that right?

Kamran Khan: Yeah, that’s in fact, you know — I’ll even rewind just a little bit earlier to that. I was doing my infectious disease fellowship training in New York City and West Nile virus first showed up in Queens — and we had heard about these cases and seen these cases of this flaccid paralysis and we were thinking: well, what is it? And then of course we found out this is West Nile virus. So during my training at an early stage, I started to realize — boy, these diseases are kind of jumping across continents. And then I moved back to Toronto, which is home, and then SARS CoV 1 happened just when I was starting at St. Michael’s Hospital here in Toronto — and that was actually 20 years ago right now.

Manjula Selvarajah: So when you look back at that time, the time of SARS and even the time of West Nile virus, I wonder — has the movement of diseases, has that changed? Is it different now?

Kamran Khan: Oh, for sure. Yeah, absolutely.

Manjula Selvarajah: What’s different?

Kamran Khan: I think there’s sort of a confluence of multiple different factors that has changed the global landscape of diseases. One of them is just human interaction and encroachment and disruption of wildlife ecosystems — and those viruses and microbes are spilling over into human populations. I think the second key thing is that human migration has changed. And we often don’t think about it, but humans are actually the vectors for the movements of these diseases. If you think about mosquito borne diseases, mosquitoes really just fly about a kilometre or so in their entire life span. But humans can carry these viruses, go to another part of the planet, get off of a plane, and when a mosquito bites us — we infect the mosquito. And then the mosquito actually now can transmit that to the next person.

Manjula Selvarajah: So we are the more dangerous vector in some way.

Kamran Khan: Well, when we think of how much movement there is across the planet, we are moving microbes around the planet. And then of course there are other factors like climate change. And I guess lastly, I would also say the world, I think, has become a bit more fragile when we think about inequalities. You know, infectious diseases can really exploit those inequalities. We often talk about these as the social determinants of health, and housing, and education, literacy, and basic income. And I think we’ve seen even here in our own country in Canada, but all around the world, these growing inequalities — and infectious diseases don’t really care who you are. So I think these are some of the underlying drivers that is changing the global landscape of diseases. And it’s happening in a way that it feels like it’s really accelerating.

Manjula Selvarajah: So here we have this picture that you’ve painted — I would say — not a pretty picture. And on top of all of this, we have climate change. As we try to deal with the effects of climate change on so many aspects of our society, tell me, what does a warming world mean for new outbreaks?

Kamran Khan: Well, first of all, I’ll maybe even flip the question around just a little bit and say that the relationship between infectious diseases and climate change is multi-dimensional. For instance, think about deforestation; we decide that we’re going to cut down some forests and we are going to do some industrial farming there. In doing so, we are disrupting wildlife; and then humans are either becoming exposed to or in closer proximity to some of these viruses and microbes found in wild animals — or the wild animals are exposing livestock — and then as we consume those livestock or slaughter those livestock, humans can then get exposed to those diseases.

So that’s one way that human action can increase our exposure to infectious diseases, but then again the use of that land is now increasing greenhouse gas emissions, which is contributing to a warming planet. And maybe one of the most obvious effects is that insects that transmit infectious diseases can start to thrive in new geographic areas. It wasn’t that long ago we weren’t thinking about Lyme disease in Toronto — and now we have to — so this is not something that our children have to think about, it’s happening in our own lifetime. And then if I think about people around the world… mosquitoes — I’ll just name one — Aedes mosquito, you may have heard of. It happens to be a particularly efficient transmitter of multiple different types of viruses: dengue fever, Zika, chikungunya, yellow fever, and others. And we’re just seeing this changing landscape of where these diseases are appearing.

Narration: The mosquito Kamran mentioned, the Aedes, it’s a species that was originally found just in tropical areas, but it’s now on every continent except for Antarctica. And it’s a really important vector for modeling the future spread of disease.

Kamran Khan: So those are some effects, but then there are others — things like environmental disasters, flooding — which can affect waterborne infectious diseases, cholera outbreaks, and so forth. So there are multiple different dimensions of how climate change is changing the landscape of infectious diseases.

Manjula Selvarajah: So what climate-related disease are you worried about right now?

Kamran Khan: Well I’d say, just here in our own backyard, I think about tick-borne diseases. And there are diseases that when I was doing my training in the U.S. — I would see much more often — and now they’re making their way further and further north. But if I look at it from a global perspective, not just me being concerned if I’m going for a hike somewhere, from a global perspective — diseases like dengue, like chikungunya — dengue is spread through Aedes mosquitoes, and we are just seeing this vector-borne disease expand its range across the world. We’re seeing outbreaks in parts of Mediterranean Europe, in parts of the United States. And these aren’t outbreaks where someone has returned from another part of the world and they have been a traveller found to be sick with that disease. They haven’t left. They’ve acquired the infection locally. And so I’d say that the main areas that I would be concerned about certainly are things like vector-borne diseases and water-borne diseases. But do keep in mind, when we have impacts to wildlife — that’s where we might be exposed to a virus that humanity has never seen before and that could trigger a worldwide event like COVID 19, or worse, that could just affect billions of people’s lives.

Manjula Selvarajah: Now, given all of these worries that you’ve laid out about a warming world and new diseases — maybe another pandemic even — how can AI help keep the world safe from climate-related diseases?

Kamran Khan: Yeah. On one hand, I realized when I’m talking about this, it is a little bit gloomy thinking about all the potential outcomes that could materialize. When I think about how we respond to these threats, I think about a few things. Number one is, we’re going to have to better anticipate what is coming next. We’re going to have to be faster at being able to detect outbreaks. We’ve got to understand where they’re going to, and then we’ve got to be able to mobilize responses in the right places, at the right time, in a very coordinated fashion because we don’t have infinite resources.

So when I think about the role of AI, I mean, what we have today — just to counter all of the things I’ve been talking about — is this enormous diversity of global data that is starting to become available. We can use things like machine learning to train a machine to, to look for important patterns in data. And then we can actually move data and insights around the world faster than an outbreak. I mean, if we think about going onto our phone and sending an email, that’s faster than an aircraft can move across the planet. So I think in many ways we have a lot of the scientific and technological and data analytical capabilities to be able to respond to outbreaks and to be in front of them — rather than being in the path of their destruction — which is the challenge that we face today because we’re always trying to play catch up, so we’re always behind.

Manjula Selvarajah: What are the limits you think to what AI can bring to the solution? Because I mean, you are dealing with global public health. There’s so many moving factors and where data is stored and how much access you have. There’s privacy. There’s a lot of noise. How far can it go? What can it do and where are the limits?

Kamran Khan: Well, I think you’ve highlighted some of the more, I don’t know, maybe I might even call them almost administrative challenges that we have with just being able to access and share data, to generate insights from data that may be in different environments and we have different custodians of the data. I think there are solutions that are becoming available, things like federated learning, which is where organizations can still maintain their own data but we can actually use technology to be able to gather insights from data that is not necessarily located in one place. We don’t have to have everyone put all their data into one server, for example. So I think there are some technical ways that we can address this data sharing, privacy — these are things I think we can, we can overcome with technology.

I think to me, the bigger challenge is that — I don’t see as necessarily overcoming in the immediate future is that — you know, machine learning today requires historic data. I mean, we have to train it based on something we’ve seen in the past. But when we’re talking about infectious diseases and in many instances climate change, we’re dealing with things that humanity has never experienced before. We don’t have any historic data. So when I was thinking about some of the work we were doing at BlueDot, and picking up that there was this unusual respiratory illness in Hubei province in China — a machine basically presented that to our team and said: look, this is something you should pay attention to. But how many other coronavirus outbreaks has the world’s seen before? We’ve seen SARS, we’ve seen MERS, we saw SARS CoV 2. We’re not dealing with tens of thousands or millions of events. We’re dealing with anecdotes. And so this is where I think human intelligence can augment machine intelligence. And I know there’s been a lot of discussions recently with large language models and GPT about, well, is this better than a human? And I kind of think of it as: what do humans do well and what do machines do well? And how do we get each to play to their respective strengths?

So I think that is one area in which there are certain limits to what kind of intelligence a machine can generate. But I think the most important thing is, we haven’t really dedicated enough time to think about human behaviour. And when I think about response to pandemics, behavioural science is really important. What is it that motivates people to take an insight and to be able to act on it? Or policy changes? Or other things? Ultimately, these insights have to turn into action quickly because when we’re dealing with infectious disease outbreaks — we’ve learned in the last 20 years or so — they move really quickly. And that means we have to be able to move faster. So I think these are some of the areas that I see as being really important. There’s a bunch of technological limitations that I think we will overcome.

Manjula Selvarajah: And then what to do when you have these insights, what happens?

Kamran Khan: An actionable insight with no action is a missed opportunity.

Manjula Selvarajah: The first time that people like me heard about the use of AI to do this kind of work was through the pandemic. You know, I remember hearing about AI models picking up the start of COVID in January, at a time when I myself said to multiple family members — who were worried about this — this is not going to happen. Our global health system is going to make sure this is not going to happen. And then of course, March came to me as a surprise, right? Which as a journalist is shocking to say, but it was the truth. I didn’t want to believe that this could be possible, even though there were all of these news reports. So there was this sort of early success, but do you think that AI prediction models did as good a job through the pandemic?

Kamran Khan: I think it goes back to that point we were discussing earlier about an actionable insight in action. And so, in fact, in many ways, I think the easier part has been to build this system to be able to monitor outbreaks around the world and to bring all these other types of data sources together and do this analysis.

Manjula Selvarajah: And say: here’s something interesting.

Kamran Khan: Yeah. And need to pay attention — here’s something that demands our attention now. But the harder part is influencing human behaviour. And so, it was in December actually, of 2019, that we had been picking up information about this respiratory illness in Hubei province. But you’re right. I mean for the longest time, the word was, this is a low-risk situation. And I think, you know, my thinking always throughout this was it’s low risk for right now until it actually arrives here. And I think we lost a lot of time between January and March where we kind of just watched things for the most part; and then as March came along and as we started to see, you know, this epidemic curve that was rising quite quickly, we scrambled into response mode. And I think what it really reinforces and highlights is that humans can respond to fast moving emergencies well, but we don’t do particularly well when it comes to slow moving emergencies. And that’s the case with infectious diseases, and I would say that’s exactly the problem with climate change.

Manjula Selvarajah: Yes, I would have to say that that is the problem that we’ve noted too, with climate change throughout this podcast.

Kamran Khan: If you ask any expert when it comes to pandemics: are we going to have another? The answer, of course we are, and we’re probably going to have more much sooner than we have in the past. But it also shows a human’s discount on the significance of future events relative to present day events. This is just kind of the way I think our brains are wired. Even if we know something is a certainty, we still tend to discount it because it’s in the future. And this is the big challenge around pandemic risk, this is the big challenge around climate risk, is how do we address that human bias and focus on prevention and preparedness and less so on response. This is probably just part of our evolutionary process, is that we just react to things that are quick because that offers us a survival advantage. But humans are going to have to grow out of our infancy, and we’re going to have to actually be able to have that foresight and to think about risks that may be years or decades or even generations ahead of us.

Manjula Selvarajah: It’s interesting that you talk about turning insights into action, because the other thing that I wonder about is if we have a hard time learning from past pandemics. What do you make of that?

Kamran Khan: What I’ve seen in my 20 years of working in this space is: an outbreak, a vigorous response and then very quickly just moving past. I think sort of with this feeling that we’re done, and this is not going to happen again. And what has been really interesting to me is to see that cycle happen with SARS, and then H1N1, and then MERS, and then chikungunya, and Ebola, and Zika, and this time I kind of thought: we’ve been locked in our houses for a year — I don’t think we’re going to forget this time. For one, there’s fatigue — I get that — but I’m already seeing this, just wanting to look past and just forget about it. And none of us are here to be alarmists — that’s certainly not my goal. But if we can’t remember the past, and if we can’t learn from it, we’re condemned to repeat it. And I think that’s the concern that I have: will our memories last long enough? Will we continue to recognize that we need to use every day of peacetime to be preparing for what is going to be coming, even if it’s not right in front of our face today.

Manjula Selvarajah: Let’s get back into that — into that technical portion; the first bucket for a minute here. There’s always been misinformation, for years now, but we certainly saw the rise of it through the pandemic. What is the impact that misinformation has on the efficacy of AI? Of AI prediction models of this kind?

Kamran Khan: Yeah, I think on one hand, when we think about training data sets — which are ultimately how we train the machine to be able to kind of look for certain types of patterns or outcomes — if we look at things like natural language generation, we’ve been talking about generative AI recently. On one hand, it can write things in a way that sounds like a human, but it also can generate text that of course is misinformation. And I think one of the reasonable and appropriate concerns is: will this amplify misinformation in ways that makes it harder to even know what’s real and what’s accurate and what’s not. This is why I think it’s so important, and this is sort of our modus operandi at BlueDot, is, there’s always human subject matter experts in the loop. This is not just “let a machine go and run with this.” I mean for one as a physician, I’ve taken an oath to first do no harm under the Hippocratic Oath, and so we really take that very seriously. We are looking at what types of data we are using to incorporate into our models and not just ingesting whatever is out there because we don’t necessarily know whether it’s accurate or not. We need humans in this to be able to qualify that this is scientifically valid information and needs to be incorporated.

So, these are, I think, some of the really important challenges that we’re going to see going forward. And I think for many of us, we don’t quite know what generative AI is going to look like. If we think about it, it’s only been months since we’ve been introduced to this. So, what is this going to be doing a year from now or three years from now? Still to be determined. But definitely a lot of power, but also a lot of concerns about misinformation.

Manjula Selvarajah: What would you say to someone who’s worried about the role of AI in healthcare, who has just read all of these things and been really thinking about some of the negative things that they’re hearing?

Kamran Khan: Well, it’s probably just like anything else. I’m just going to use the analog of a medication — it has certain benefits, but it does come with certain risks as well. And I think the key here really is to be thinking about how we utilize AI in a way that augments human intelligence and human capabilities — but making sure that we’re thinking and going with eyes wide open into the potential risks and we’re addressing them. You know, we just talked a little bit about generative AI, and the term is often used around hallucination, that some of these models will hallucinate, make up things. Very recently, I was reading an article in the New York Times about a lawyer who used GPT 4 to help with a legal case.

Manjula Selvarajah: Oh yeah, I read that one, too.

Kamran Khan: And then it turned out that everything that it said was just made up, none of it was true. So clearly, this would be very dangerous if you’re talking about supporting a patient, someone who has an illness, or you’re trying to promote someone’s health or prevent an illness. So these are things that we are just at an early stage, we’re beginning to understand what the potential benefits are — but we’re still learning about some of the risks. I often think about this in terms of augmented intelligence — it’s not just a machine or a human — it’s leveraging the capabilities of a machine to do things that a human can’t. Humans have to go to sleep, we get tired after doing 36-hour shifts or whatever it might be. But a machine can continue to process information 24 hours a day and just never gets tired. So these are just some examples of how we can play to machine strengths to be able to generate insights. And the approach we’ve taken here at BlueDot is for those machines to essentially find the needles in the haystack that we think we need to be looking at, and then it’s human intuition and knowledge and insight and reflection on — what do we want to do with that information? Are there any other concerns? Do we need to qualify if this is an accurate insight? Because these are models that are not perfect. And so we have to think about bias and edge cases and potential harms that may be coming from them. And I’d say in many ways right now, I think we’re probably obsessed with all of the potential. We’re just like, wow this is incredible. But we need to spend more time thinking carefully about some of the risks, thinking about regulation. How do we use AI in an ethical and a responsible way?

Manjula Selvarajah: One of the things that we’ve been thinking through this podcast is, when you think about climate change, you are always thinking of these decades in the future: what’s 2030 going to look like? What’s 2050 going to look like? Whatever space you’re talking about — be it electric cars, wastewater — it’s always something that comes up; you’re always painting a picture of what the future looks like in all of these different spaces. So paint us a picture of the future in healthcare — when it comes to healthcare powered by AI, what is that going to look like?

Kamran Khan: Well I think that’s a really nice way to shift this conversation because I think on one hand, you can easily get overwhelmed and be pessimistic and just feel like this is a problem that’s just too big — like I don’t even know where to begin. But when I think about data and AI and I would just say in general, analytics of data, I think we’re going to be able to have better situational awareness of what’s happening around the globe in ways that we haven’t in the past. When I went to U of T for medicine and got a great education — but you know, only a small portion of my training was actually in infectious diseases around the world. So we didn’t spend a lot of time on diseases that I might never encounter in my career or my lifetime. But the world, again, going back to migration now, anything from anywhere could show up in my clinic or in my hospital. So I think AI gives us the ability for a clinician to have global situational awareness in ways that we didn’t have 10, 20 years ago.

Now, this has to be done in a way where we’re not overwhelming the clinician, because if we just bombard the clinician with information, they will have to tune it out. But I think AI and advanced analytics gives us this opportunity to be anticipatory, to remind the clinician, you need to be thinking about the following right now, because there’s an outbreak halfway around the world and there’s 10,000 people moving from that location into your city — and, and, and — and here’s what it looks like, and here’s how you diagnose this, and here’s the personal protective equipment you need, and here’s the appropriate precautions that are required. These are some of the outcomes that I can envision with AI. And even going into the EMR and EHR systems and saying, a person’s liver function tests are abnormal, their blood counts look like the following, have you considered this as a possible diagnosis? These are things that humans…

Manjula Selvarajah: And EMR and an EHR are?

Kamran Khan: Oh, sorry. The electronic hospital records or electronic medical record systems where the patient’s laboratory results are stored. So we have to look for patterns: What’s happening around the world? Is this an unusual illness? AI has the ability to synthesize all of that and then present an insight to a clinician to recommend or suggest: Have you considered the following? And it may be a disease that we’ve never even seen before, it may even be a disease we’ve never even heard of before in our medical training. So this is where I see AI being able to really elevate our capabilities. We often talk about — the astute clinician recognized that this was unusual and then picked up this new disease. Well, I think AI can create astute clinicians across the board, and can elevate all of our — not just situational awareness — but our medical knowledge and our ability to focus on being human. So I see a world where human intelligence is vastly leveled up using machine intelligence. And I think that’s going to allow us to, again, move smarter and faster and deliver more effective and more cost effective and efficient healthcare than we’ve ever been able to in the past.

Manjula Selvarajah: On that note, Kamran, thank you.

Kamran Khan: For sure.

 
Solve for X is brought to you by MaRS. This episode was produced by Ellen Payne Smith. Lara Torvi and Heather O’Brien are the associate producers. David Paterson is the senior editor. Mack Swain composed the theme song and all the music. Gab Harpelle is our mix engineer. Kathryn Hayward is the executive producer. So far this year, the production team has visited 10 international cities, on three continents and racked up more than 106,000 kilometres in the air. As far as we know, nobody has brought back any new diseases. I’m your host, Manjula Selvarajah. Thanks for listening and watch your feed for new episodes coming soon.

MaRS helps entrepreneurs looking to scale solutions in climate tech, health and software. We offer targeted support through our Capital and Growth Acceleration programs. To learn more visit us at marsdd.com.