AI Infrastructure and the Future of Imaging | LSI Europe '24

Imran Hamid 00:06
All right. Well, hello everyone. I guess we call you after lunch, so please stay awake. You should be well fed and well rested. My name is Natasha Allen. I'm a partner at Foley & Lardner. We're a U.S.-based law firm, but the other role that I play is I'm also the AI co-lead of our firm's initiative. As we are aware, AI intersects with a lot of industries. Today, we're actually going to talk about AI and imaging. Just to give you a little bit of background, I'm going to ask the panelists to introduce themselves. We have a huge, I'd say, expertise in terms of geography and investment types on the stage—really well-rounded individuals. So I'll let each of them introduce themselves, and then we'll get started. You want to start? Thank you. Good afternoon, everyone. My name is Imran Hamid. I'm one of the senior investors at LifeArc Ventures, a UK-based charity that has quite a significant amount of capital to deploy across the full spectrum of biotech, health tech, and devices. We have a very healthy interest in AI, and yeah, looking forward to the discussion this afternoon.

Sahir Ali 01:12
Hello. Sahir Ali, I am the founder and general partner at Modi Ventures. For the purpose of this panel, my background is in medical imaging, digital pathology, and I'm an entrepreneur as well. We're investing in what I call the bio stack, and so that could be from human health all the way down to the cell.

Soyoung Park 01:31
Thank you. Natasha, hi everybody. My name is Soyoung Park from VU Venture Partners. We are a global fund focused on healthy aging. So we are looking for transformative technology to serve the most painful problems in the world at root cost level, with the most cost-effective way.

Vishal Gulati 01:56
Thank you very much for inviting me, and thank you, Natasha, for sharing this. My name is Vishal Gulati. For the last three decades, I've been a physician scientist and then a venture capitalist. In 2022, I founded a venture capital firm in San Francisco called Recode Ventures. We focus exclusively on AI healthcare companies or AI-first companies solving big healthcare problems. And that is my perspective on this great, wonderful topic. I think we'll just do a little bit of background. So, you know, AI is not a new concept. In 1950, Alan Turing wrote his paper entitled "Computing Machinery and Intelligence." Fast forward to November 2022, you have ChatGPT, and everyone thinks, "Oh, this is new technology. What's going on?" Right? I think that there are many reasons why people think that now is the time that AI is making a resurgence. But in particular, as I said before, we're going to talk about AI and imaging and really just AI in health and infrastructure.

Vishal Gulati 02:58
You know, some ways that generative AI is being used right now in the healthcare system include drug discovery, where you have it identifying new drugs and new targets. You have personalized medicine, where obviously personalized medicine involves looking at an individual's actual genes to figure out what medicines and what therapies work specifically for that person. Clinical trials, and then what we're going to talk about today is actually imaging. So it's being used to create synthetic medical images, right? Some companies are doing it for that to assess AI X-rays and MRIs. So, you know, we have a vast majority of knowledge in terms of investment right on this panel. And I'd be remiss if I didn't ask the panelists to really give us an understanding of what you think of when you're thinking about investing in a particular company in this area. I'm sure there's a lot out there, and there must be some type of differentiator between just, you know, investing in any company that you include in your portfolio. So I'll start. Let's just go random. Sahir, let's start with you.

Sahir Ali 04:06
Oh, wow. Okay. I guess what I'll say is, I mean, I'm a tech bio investor. So for folks who haven't heard the term, it's just a VC term for investing in transformative technologies that are disrupting things that were traditionally done in biomedicine. So AI drug discovery is an example of that. But in relevance to this conversation, I think it's with a general-purpose technology like artificial intelligence, and particularly LLMs, where you can just take a bunch of sequences and sort of put it into this general-purpose technology and find rules that govern that. I think we have an opportunity to start characterizing disease in a fundamentally different way, right? So if you take an example of an aggressive disease like cancer, for example, prostate cancer, where you have multiparametric MRI that characterizes the tumor.

Sahir Ali 05:00
But you also have radical prostatectomy through which you might get a pathology sample, which is getting digitized. It's a pretty common practice now in prostate cancer to also do germline genomics testing. And so if you think about what just happened with that disease, different dimensions and the length scales of the data are present, and different skill sets are looking at it. But you could now start to collapse all of that and really create a risk score of what that disease looks like, and ultimately what it matters. It comes down to what treatment and what dosage is at all. So I guess that's to me when it comes to, I guess, investment as well, there's a real opportunity here to create transformative technologies that are really geared towards finding the right treatment at the right time for the patient—the whole idea of precision medicine. That's particularly exciting to me when it comes to kind of the future of imaging and where I think AI could be ready to go—the unmet need. Vishal, do you want to go?

Vishal Gulati 06:00
Sure. So in my experience of investing in AI and healthcare, the value proposition and where we look for value or differentiation has moved on over time. So when I first invested in companies in AI healthcare in the 2016-2017 timeframe, when it was relatively new, what we used to look for was access to data. Because everyone was saying, "Hey, data is the new oil," and all of those things you might remember. And so that was one of the things. But what we found within this short span, that gap got filled because data providers became more aware of the value of the data. They started to do partnership-based partnerships; they started to trade data in many ways, so that no longer remained the only competitive advantage a company would have. So then the focus moved on to what type of AI they were using. So, you know, starting with very early on, the new neural networks and recursive learning—that was the first generation that got commoditized. And we started seeing more and more people using deep learning and unsupervised learning and things like that. So that became the focus. We started backing engineers this time that, you know, who are going to be at the cutting edge of this. And then again, as it happens, and this is one of the learnings from being in this AI space, is that things get commoditized really fast. And so now where most of the value is, in the case of tech bio, is in the assets—what assets you produce—because you can use other people's models if you have to, you can use other people's data if you have to, you can generate data more cheaply, but ultimately the value gets crystallized into that asset. And in the case of more imaging-based companies which are selling to hospitals, for example, in their case, it goes into the strategy and reimbursement—what their go-to-market strategy is. And that's I bore entrepreneurs who come to me to death by asking them repeatedly, "Tell me again, what is your go-to-market strategy?" That is important, and reimbursement is important.

Soyoung Park 08:18
Yeah, we are very similar in terms of investment. So AI, as Vishal mentioned, we actually saw AI a long time ago, even before the pandemic. At that time, AI was, "Okay, let's see how you can improve." We were just kind of watching it, and later on, now look at this very handsome gentleman called Sam Altman who put the cat outside of the box. And now everybody knows about AI. And now as an investment, we actually ask, "Do you really have AI technology?" I know you want to do that, but do you have it? That's the first thing I actually review. And second of all, do you have data? Do you own the data? Do you want to borrow data? Which one is more valuable? If you own your own data, it is unique. Data is very, very valuable to us. And then finally, how you train it. You train it is kind of raising AI. How you raise it grows in different ways. There are millions of people who have a different way of thinking. We cannot kind of say one AI, and we don't know if they say all right. So we want to ensure that you train it the right way, in the right direction, and then bring the right outcome. So those are the three areas we actually pay attention to for the companies who brought AI technology to us.

Imran Hamid 09:42
Yeah, sure. Let me just add to those comments. When we're looking at AI, one of the first questions I take a step back and try and figure out is, to what extent are we doing something that is incremental to existing clinical practices and workflows and then can accommodate some form of optimization or efficiency step? Or is this something that's truly category-defining that's going to change the way we do everything? And on that scale, where do we think we're going to get real adoption and scalability? Because I think you have to assess these businesses very differently, depending on where they sit on that spectrum. Then the natural question becomes, to what extent is there real, meaningful differentiation and defensibility? Vishal spoke about the fact that a lot of these businesses are very quickly commoditized, and having tracked a lot of the companies in the space, there are companies who have a lot of charm, and we sort of like them, and we take a step back and think, "Gosh, I can't see how anyone isn't going to replicate this," because the technology set is changing so quickly, the engineers are advancing the way they exploit their insight on this data so quickly. It's very hard to take a spot in a fast-moving market and feel confident that you're making the right decision. So in summary, I think one of the things that I look for across the team is, and this may be my bias as a former surgeon, is thinking about to what extent will there be meaningful change in clinical practice based on technology that's coming forward, and how can we look at that confidently, knowing that that is what is going to drive scale and adoption, versus all the benefit being in the technology itself, which may not be differentiated and defensible in the long run. And there are a number of companies out there that I think have done that well, but I think it's still too early to judge if they're truly going to be successful.

Imran Hamid 12:02
So then I guess this parlays nicely into my next question: what do you think are some of the challenges that some of your portfolio companies are facing right to break into health infrastructure, especially if they are AI companies? I think you touched on a few things, but I'd love for you to expand on that. And if anyone else has any comments, I'd love to hear what you think some of the challenges are.

Sahir Ali 12:35
Yeah, I'd be happy to pick up on that. So one of the most obvious things in this space is the tension between the imaging providers and the radiologists doing the analysis of the imaging. And this concept of, do radiologists really want to be replaced by AI? There's a natural question. They are paid very well to assess images, and the reimbursement is based on that. And you've got to think about, if I'm bringing a technology into the space, am I really trying to displace the people who are at the decision-making space on whether that technology is going to be adopted? You know, you're going to have to think about the political environment, the lobby environment, the medical organizations, and do they want to disappear overnight? No, they don't. So actually, how do you make thoughtful intervention in that space that builds up trust over time, so that they become the protagonists for wanting to see that change? And I think there's a company that I've seen do this actually pretty well, called Rad AI. I'm not an investor in the company, but I've tracked it for a little while, and what they spotted is that radiologists were suffering from burnout, sitting there day in and day out, having to report X-rays, CTs, MRI scans, and actually there's a huge amount of stress on them. And what they found is where AI can be used is to summarize the impression of the analysis they had done. So you're not displacing the radiologist, but you're helping them with their workflow. So that builds up the trust that the AI is my friend. It's helping me do some of the work, but it's not displacing me from the work. And I think the future could be that as that relationship gets stronger, maybe that AI does more for the radiologist over time, but you're not seeking to displace them. Now, some may argue that's a little bit incremental, but I look at this for the long-term thinking, "Well, if you got a substantive amount of the U.S. market using this type of technology, it buys you the right to then bring different levels of innovation thereafter." That's the type of play I think could be successful in the long term.

Natasha Allen 13:37
Yeah, well, you're talking to somebody who we are going to be displaced as lawyers. So it's fine, but I think you touch on a good point. It's education, right? Educating the workforce in terms of how that workforce can benefit from this technology, right? Using it as an additive, as a replacement instead. I'd love to hear anyone else's comments if they have kind of, you know, what are some challenges you've seen with your company?

Soyoung Park 14:08
I think in healthcare, decision-making is slow. That is a challenge for most of our companies. We also feel that it's very much dependent on reimbursement. I keep coming back to that. I think that a lot of companies that have done really well started with a very clear reimbursement strategy from day one, and the companies that didn't do that, that were more engineering-focused, have struggled a bit more. That's what my feeling is.

Vishal Gulati 14:42
Yeah, in my case, I want to go back to data. So human health data is abundant. It's a lot of data, and it's very, very noisy. So we actually don't know which one is actually showing the right one because everybody's part is different. Everybody has a different outcome. And if you say A doesn't mean it goes out to A+, it can be C or D or E. So we really kind of concerned. You are able to find very, very clean data—not to have garbage in, garbage out—to ensure that you actually bring reasonable outcomes. We can use it as clinical data. That's what I actually see as a challenge.

Imran Hamid 15:00
Sure, that you actually bring the reasonable outcome. We can use it as clinical data. That's what I actually see as a challenge. So, you know, they say when there's a gold rush, provide the picks and shovels. I think in the infrastructure space, there is a real opportunity. Rad AI is a great example of that. But if you look at another imaging modality, which is pathology, before the pandemic, the microscope was invented in the late 1800s, giving rise to cellular pathology. Since then, we kind of look at the pathology space in a very similar way—hematoxylin staining, and you have ISC, and nothing much has changed. If you look at the borderline cases, there is a 66% concordance rate, which means there's only a 66% agreement among pathologists on what is a scary number. And you know what the biggest sort of blocker to the adoption of digital pathology was? What is digital pathology? It digitizes the samples and looks at them on the monitor. It was approved by the FDA in the United States in 2009, and the friction point came from pathologists themselves. So forget AI. They just didn't want to change what they were used to until the pandemic. Then what? What it realized was that hospitals and pathologists who were on digital pathology were able to work remotely. Yeah, and now there's a huge sort of urgency in setting up these things. And so there's an opportunity there. Now, if you look at, I mean, again, I'm U.S.-centric, so I'll say FDA. There are 188 approved medical imaging algorithms sitting on the shelves. And guess what? They don't have ways to plug in. Everyone now has to figure out how they're going to integrate with PAC systems or in digital pathology. There isn't even a standard PAC. So I think there's a real opportunity to build infrastructure there. AI-first, a good example of that also is Deep Six. You know, it's a very simple model, but it does help, which is a traditionally spreadsheet-based system in the hospital, matching patients with trials. And also, I think there's a real opportunity in the infrastructure layer, where I think the real innovations are happening. Is what I just talked about—characterizing disease, better precision medicine. But I think these things have to plug in somewhere. So that's where I see a good opportunity.

Vishal Gulati 17:58
I think also on the infrastructure side, if I may continue on that, I think we may be at an inflection point in terms of infrastructure around GPUs, for example. And I think that there is a race out there—there is an arms race—and by and large, healthcare is losing in that race. So I did some analysis to try and see how many healthcare organizations have access to GPUs, and I found that the healthcare industry is probably at the bottom of that. So, you know, on the top, you have Microsoft and Meta, who bought 150,000 chips from Nvidia and spent $9 billion on it. Tesla is kind of in the middle with 10,000. And when I look at healthcare organizations, many of them don't even have access to any GPUs. And, you know, the only thing I could find was Mass General did a deal with Nvidia several years ago where they got access to, you know, HDX one, which is like the old version of what a 100 now is. So HDX is like 900 teraflops, and H1 100 is five petaflops. So basically, they're missing like 4 trillion floating points per second. That's some of the gap. So if you compare that to your current MacBook, you'd have to compare their server with the 2016 MacBook, right? That's kind of what the range is. So I think that what I find is that as we move to the new generation of AI in the generative space, pharma companies have greater access because they're using more biological samples, but I do believe that there is a massive gap in healthcare, and I think we may just miss out on a lot of good things that could come out of it.

Natasha Allen 19:39
That's interesting. It's interesting you say that because I was reading about the budgets for a lot of these healthcare organizations; it's not huge, right, for innovation and for technology. So if you want to expand, you know, into bringing your systems up to what we are to right now, they don't really have the funding to do it.

Imran Hamid 19:56
I agree. I think right now it seems that PepsiCo has more GPUs than all of the healthcare systems put together. And this is where we are.

Natasha Allen 20:00
So someone had touched on this, and it was really, what are the gaps? And I think you had addressed this too. And I think Sahir had talked about this, where can companies fill the gaps, right? If they are looking to build out a company that can help, you know, whether it is make things more efficient, help with the systems? Is there anything else you think of where companies could probably double down on to make themselves a little more relevant?

Soyoung Park 20:05
Well, I mean, literally, there's a huge opportunity in the healthcare sector with this AI technology. As you know, I don't know Europe very well, but at least in the U.S., a lot of them are still using paper-and-pen technology. Actually, they are using, and now we are moving towards the administration automation. And when you go to the clinic, you talk to the doctor before the nurse types everything. Now they are using AI technology. They just kind of talk—a regular conversation. It's all generalized. So those start from those kinds of real-life experiences. It'll change that we actually are going more towards digital clinics. In the future, we will not complete a different world. Right now, we are at the starting point of all this technology, which is not perfect yet, and you are the people who will perfect that and will find out who's the best. We want to see the future where people can be going to more prevention from the reaction with this technology. That's what I'm hoping for.

Soyoung Park 21:31
We can find the disease way faster. We can maintain our lifestyle healthier, getting help from the AI technology—all the intelligence combined. And we don't need to go to each teacher or professor. We just kind of ask AI, and then they give us all guidance. That's the future we want to see. And hopefully, we can be there yet. We are not there yet.

Vishal Gulati 21:56
Yeah, again, I talk about data continuously right now. Our data in our AI is using text and X-ray images. We call it unimodal AI. We are moving towards—we are hoping to move towards a multimodal approach so we can have a lot of different sources of data. It can be, of course, text, it can be audio, video, and all of this happens simultaneously, so we can actually review and we can process it simultaneously. All data will combine, and we can have a way faster, way more accurate results from those AI technologies.

Natasha Allen 22:59
I had a question. So I know some of the investors here invest in other markets outside of the U.S. Is there a difference in terms of the challenges? Are they the same across all boards? Right? In terms of when you're investing in companies outside of the U.S., are you finding that your portfolio companies are having the same challenges? Are they different? Let's take, for example, just let's focus on the UK. I think that's easier rather than saying the world. Do you find that the problems and the challenges are the same? Are they slightly different? I'd just be interested to hear a little bit of that.

Imran Hamid 23:31
I think that broadly, there are universal challenges, yeah. And actually, just to sort of pick up some of the points that have been made earlier, in terms of what the gaps are, yeah, I actually think the gaps aren't from the companies themselves. I think for the large part, the companies themselves have good technology, good teams. For the most part, it looks like the technology works. The gaps are actually at a system level. Yeah. So the gaps are a lack of urgency among the healthcare system to want to actually adopt AI, be it at the provider level or the payer level. I think there's clear benefits to adopting this technology. There's clearly investors lining up wanting to invest in the technology, but I think these actual companies themselves are struggling to get to meaningful scale because the providers and payers don't have any sense of urgency. And I think Sahir's point to the fact that there was better adoption during COVID is a great sign that people will make the adoption. I'd argue that it was probably the pathologists who realized that their revenue streams were going to diminish and suddenly wanted to do digital pathology. And I think we need to create that sense of urgency. I don't know how we do that, but I think there's an area around how that happens.

Vishal Gulati 25:00
And the other point is thinking longer term from the investor perspective: where do the exits really come from in this space? And that's something we debate a lot. It's not clear whether we're trying to create a whole new category of companies that stand up on their own, maybe get aggregated as standalone entities, or whether the major strategics are going to play in this space. And I think until we've got sight of that, I think it's going to be quite hard to invest, and I think those dynamics are universal whether in the UK or the U.S. Yes, there are some nuances, but I think they feel like the bigger issues.

Natasha Allen 25:30
And I was also reading that there might be some consolidation, right? We know that the investment market right now is a little tough, right, on both sides, if you're raising a fund or if you are a company yourself, right? Um.

Imran Hamid 25:56
Yeah. And so I was reading that maybe there will be some consolidation in the imaging area. I don't know what your thoughts are on that and if you think maybe some of the larger conglomerates will start eating up the smaller. Or if, you know, maybe some of these companies go into other markets, right, LATAM, just other jurisdictions to survive. But I'd love to hear that you touched on that a little bit.

Vishal Gulati 26:37
We recently sold one of our portfolio companies in the AI space and imaging called Caption Health to GE Healthcare. And I don't think that it is fair to say that GE is very acquisitive right now. They've made a couple of smaller acquisitions after Caption as well. But Caption was a very strategic play for them. Ultrasound is a very big market for them. They have very strong market share, and they want to dominate that. So that's so it is not—I cannot use that as a data point to say that there are these companies ready to buy all these AI companies. A lot of them are waiting on the sidelines to see what they can do. So you can look at many companies in, say, the breast imaging space. They've built their own AI, so that's competing with the companies that we have backed who are doing that externally. So this will shake out over a period of time. Either their thing works really well and our company will die, or their thing doesn't work, and then they will acquire. So there is that dance going on right now still, so I don't think that we are at a stage where there is a rush for companies to be acquired. I wish that was true.

Soyoung Park 27:15
Thank God there are three Es, and this is not my sexy term here, but that is effective, efficient, and equitable. I think there are opportunities in these areas. And so when I was speaking of digital pathology, we didn't invest in a company that realized in 2017 that, okay, well, we can't—yes, there's a lack of infrastructure, but it's not a must-have for hospitals because pathologists' biopsies were, even today, by the way, the stats are that we do 1 billion biopsies a year around the world. 10% are digitized, so it's a growing one. And if you look at the McKinsey Report, in the next five years, we go towards 30-40%, so you have a big market to be created. But they realized in 2017 that it's not a must-have. And so they built a platform for education purposes. So when you're a pathologist, you're looking at these textbooks and you have these biopsy images. And so they just started putting QR codes, and if you scanned it, now you had this browser opened up, and you can zoom in and tag things, and you could share it. And so with that, they built a sort of a cohort of 50,000 young pathologist students around the world. And then they went to the hospital system and said, "Look, this is what we have, and we can actually give you a commercial product." And they started competing with someone like Sectra, who was agnostic of this, and this is built for pathologists by pathologists. There are these opportunities. You've got to get creative. There are gaps. The last thing is, I think the equitable part has very interesting investment opportunities outside of the first world. You know, for example, we have algorithms right now that can do radiology triaging and protein sort of expressions through just H and E images better than human scans. But in the U.S., because of the reimbursements and everything else, we're trying to figure out what that looks like. But you go to certain continents and cities and countries around the world where there may be one pathologist or radiologist per million people, there's a real opportunity for the technology just to skip traditional things. And that happened with fintech, by the way. M-PESA did that, right? It just sort of leapfrogged to mobile banking. So there are some opportunities around the world as well, which follow these three Es. At least that's how I think about it.

Natasha Allen 29:01
Okay, another question I had. So some of the companies that you've seen in your portfolio that are a little more successful, do you think that there's a combination of—you talked about pathologists coming up with an idea that just works right for them and then commercializing it versus engineers trying to approach a problem that they think is in the industry. Do you think it's a combination of the two? So actual individuals operating in the healthcare system and the engineers. Is it one or the other? Or are you agnostic?

Vishal Gulati 29:39
Mostly agnostic because each sector has its own specialization. So I don't think I've yet figured out the formula. I actually see a lot of founders come to me, "Oh, I really want to serve this problem," and so I ask, "What is your solution?" A lot of times I say, "Wait, wait, wait, that's not what you want to do. You want to serve this huge problem." Then you try to target a certain very small area, literally, kind of not very impactful in the future. Do you want to spend your whole time and energy and money for that? Maybe not. Our huge evaluation is the end. I spend a lot of time actually helping them to think big and then think about what they really want to do. I really encourage all the founders here to think about what problem you really, really want to serve. Probably the problem is huge. Then you think, but what you can do or what you really try to do is not very relevant. And then you think sometimes we can be creating an engineering project, creating a science project. But, can you become a business? How many people can actually benefit from that, and literally, who's going to buy your business? Or can you go to IPO? We are in business. We are not just—we are not charity or anything scientific donation.

Imran Hamid 31:09
I know it's going to say, go to Iran very quickly to clarify, reinvest back into our charitable initiatives, not me. We want to ensure that you can actually make business. That's the most important to me.

Vishal Gulati 31:31
Yeah, counter, no, I was going to say anything. If you look at the history, I think it's very successful generational-defining companies have had folks who bring domain expertise and the business mindset. So you look at Amgen, Genentech, Pharmacyclics—these were a VC or sort of the business-minded person, you know, with someone who was just a domain expert. And I think that's a pretty good combination, where you, of course, need the mindset of just building tech and just sort of failing forward sort of thing. But you also need to build it with the right domain experts. And we've seen that where sometimes you create solutions that are looking for problems, right? So you don't want to be in that position. So you understand both just my perspective, and I look for that. Think if there is a strong team or the founders, if they can balance each other from just obviously, if you're a tech company, you need a strong tech presence, business mindset, and domain expertise.

Natasha Allen 32:44
Yes, ma'am, do you have anything to add?

Soyoung Park 32:57
Yeah, again, just reflect on what are the barriers to adoption here, and how do we really break through this space? We talked about there not being enough urgency from a provider perspective. And maybe we flip it on its head and say, actually, the providers are dealing with a whole set of challenges themselves that they can barely overcome. And maybe it's a question of actually, they've got 40 or 50 AI vendors knocking on the door every day saying, "We want to do stuff with imaging. We want to do stuff in pathology. You want to do stuff on your blood results and work for management." And if you're sitting there as a director of IT in a hospital, going, "Okay, how do I assess? How do I govern? How do I make sure it's done in an appropriate manner? What do the commercials look like around this? How do I ensure there's appropriate audit, safety, access?" You know, suddenly, from the simplistic investor perspective, where we think it's easy, you can see from a cost perspective, this becomes really very complicated. So are there plays to be made where there's almost like a middleware layer, an app store-type environment? We've got the accreditation and authentication of these providers. On one hand, there's a commercial layer in the middle, and there's the integration at the other end, where you need that type of infrastructure to exist to allow scale to happen. You know, to use an often-used example, the success of Uber was predicated on the fact that everyone had a smartphone in their pocket. There were plenty of people who had the Uber idea well before mass mobile phone adoption. So are there other structural sort of requirements for there to be success in the system that just don't exist yet? Yeah, and when we're not looking at it in the right way, I don't know, but I suspect that may be the case. And maybe we are still a little bit early, and actually, in a few years' time, some of these barriers will break down, and then we'll see more meaningful adoption.

Natasha Allen 35:00
And that's what I was going to touch on. Maybe we are so early that some of it is really just the infrastructure, like the big picture, like, okay, let's just figure that out. And then the specialties and the special, you know, addressing different departments comes with that as well.

Imran Hamid 35:30
Yeah, I think there's really no for both. It's just a point in time where we are, yeah, investment capital—you can only be too early. You can't be too late.

Natasha Allen 35:42
Yeah, very good. So we only have about four minutes, and I'm going to ask the big audacious question for all of you: what do you think is the future outlook for AI in the healthcare industry? And you could focus on imaging if you feel so, but I would love to hear what you think is like the next iteration we should be looking for.

Vishal Gulati 36:00
So I think, and again, you know, I might be a professional optimist, right? This is what I do. I think that the generational opportunity is in AI right now. I think that every generation there are new platforms. Before this, there was mobile. Before that, there was the web, and before that, I would say, mobility and so on. So there have been these generational platforms on which a large number of very big companies have been formed. And I think that if you look today, where that platform is, I would argue that that is artificial intelligence. And within that, if I just narrow that down for healthcare, even if you just look at what's happened in the last two months, the number of new models that have been released as open-source models. So I'm not even talking about the proprietary models by Claude and OpenAI. I'm talking about the open-source models, which is Llama 3.1. I'm talking about Sam 2, which is a segmentation model, which is multimodal, as you were talking about earlier.

Vishal Gulati 36:59
These are available at prices that are dirt cheap, yeah. And you know, issue evolutionary scale two, which was for protein design that was released, and I believe that this is a golden opportunity for founders to start looking for applications because the killer applications of generative AI are not going to be a chatbot. It's not going to be you making these videos of Kamala Harris and Donald Trump in different types of costumes. That is not the killer app. We are in the play mode right now. And what I'm hoping for this generation of entrepreneurs is to take benefit of this huge bounty of these models that are available to them and find the healthcare killer apps for our generation. And so I'm really excited about where we are right now in this history of the world, and I'm only talking about what happened over the last two months. Just imagine if we roll this forward several months because the speed of change has become so much faster. So I think that to anyone who has any spare time and desire to change healthcare, go play with those. That's what I would say.

Soyoung Park 38:10
Yes, amazing perspective. I can, I don't know what I can add on top of that, but yeah, again, just kind of reinforce that AI technology is not technology anymore to us. It's a platform. It will be the next internet in the new generation, and our kids will grow up learning how to use AI, and basically we need to accept that. And there's an opportunity for us; we can set the fundamentals for the next generation and then ensure that they can use it properly, in the right way—not to have wrong information. It's a very, very tough, very, very difficult situation we face—how what is really ethics? Here we are talking about those things. You are at a sustainable starting point to build that whole infrastructure now in healthcare, so I really think we really, you know, it's a very important point and a very exciting moment that we are seeing the paradigm changing in the world, and I really look forward to being part of that with you—all the founders and fellow investors here.

Imran Hamid 38:44
Absolutely not. I'll just quickly add that Craig Venter said when the Human Genome Project was finished that this is the century of biology, and I think we're talking about healthcare, but let's talk about just quickly health. I think we had the hardware technology in sequencing. We have this general-purpose technology that takes sequences without being known what that is. And if you look at biology and medicine, it is just sequences—amino acids of sequence pulled into proteins. You have the DNA sequence, which is A, C, T, G sequences. You look at protein expression sequences, you look at medical imaging pixels sequences. So biology and medicine have been made for this sort of general-purpose tech. Now you combine that with sequencing technology, I think we have a paradigm shift. Does it happen in five years or 10 years? But these are golden opportunities to build on.

Natasha Allen 39:29
Where are you taking us home?

Imran Hamid 39:37
Yeah, so I agree with everything that was said; it's hard not to. But then just the practical reality of where is the best place to start this huge transformational journey? And my mind comes to, are there some really simple use cases that are undeniably required by the healthcare system that there will be mass adoption around? And what do they really look like? And they could be deeply boring. So they could be around back-end workflow optimization. They might be around triaging. It might be around the summary of imaging. It might be something less convoluted and sophisticated that we sort of tend to gravitate towards. And I would urge people to look for those types of opportunities first, build a meaningful scale, and thereafter, you buy yourself the right to build in the really interesting innovation.

Natasha Allen 40:00
Great. Well, thank you all for your insights. Can everyone help me thank our panelists?