# AI, Innovation, and Regulatory Insights with Amy Abernethy

- **Show:** Raising Health (a16z)
- **Published:** 27 Jun 2023
- **Source:** https://raisinghealth.simplecast.com/episodes/ai-and-regulation-ZsyRE776
- **Transcript:** https://podcasts.opennash.com/shows/raising-health/ai-innovation-and-regulatory-insights-with-amy-abernethy-1d71e9/
- **Provider:** OpenNash Podcasts (machine-generated)

## In brief

Amy Abernethy argues that healthcare innovation only reaches patients if every stage of the pipeline speeds up together, including discovery, clinical trials, regulatory review and rollout. Regulators can keep pace with AI only by working closely with innovators early, regulating AI in reusable pieces, and using automation and AI themselves.

The central tension is a fast-growing flood of AI products heading into a regulatory system that has not changed much. Abernethy says adding more reviewers is not enough. Her answer is shared definitions of what counts as credible evidence, earlier regulator familiarity, and smarter, risk-targeted oversight. She notes that the FDA's January 2021 AI action plan had not yet become new rules.

## Key insights

1. **COVID showed uneven innovation speeds across the pipeline** The mRNA vaccine science moved astonishingly fast, but evidence generation fell back on trial designs from around 1995. Abernethy argues that fast discovery paired with slow trials and slow rollout simply creates a clog.
2. **Neutral forums let regulators and innovators agree on credibility** The Evidence Accelerator, run with Friends of Cancer Research, brought the FDA and companies together to test real-world data methods. It informed three FDA guidances at the end of 2021, so the rules were shaped by tested practice.
3. **Regulation should follow intended use, not the technology** AI for finding drug targets or easing conversations draws little regulatory worry. Abernethy says scrutiny rises as AI becomes the product meant to diagnose or treat disease, and software that gives medical advice moves into a very different space.
4. **Regulating foundation models once could speed approvals** Abernethy suggests checking a base model, such as one that reads chest X-rays, once. Specialized models built on top would then need only a standardized check, backed by periodic performance monitoring against labeled data.
5. **Regulators should aim scrutiny using risk-based, data-to-data review** In clinical trials, risk-based monitoring uses algorithms to flag where auditors should look harder. Abernethy imagines AI doing the same for regulation, with companies pushing data to the FDA instead of shipping hundreds of pages of reports.

## Questions this episode answers

### Why did COVID vaccines get stuck even though the science was fast?

Abernethy says the vaccines were developed very quickly, but proving who they worked for relied on clinical trial methods from roughly 1995. The slow stages after discovery created a bottleneck.

### What was the Evidence Accelerator?

It was a forum Abernethy describes where the FDA and innovators met closely to test new kinds of evidence, especially real-world data. Regulators could say what they trusted, and innovators could show what they were doing.

### Does the FDA already regulate AI medical products?

Abernethy says AI-based devices are already regulated under the traditional device framework, such as diagnostics or software as a medical device. The harder problems are the sheer volume of products and algorithms that keep learning over time.

### How could AI make drug and device regulation faster?

She suggests automating parts of the review, having companies send data directly to the FDA, and regulating a base AI model once so only the specialized layer on top needs checking. She also suggests using AI to target oversight where risk is highest.

### What is Verily trying to build, according to Abernethy?

Verily aims to unlock precision health for everyone by building a data bridge between research and care. That means software for clinical trials and for managing conditions like diabetes, connected through curated data that can also be used to build and check AI models.

## Facts

- **Amy Abernethy** [0:03:04]: "essentially sequencing the genome, publishing it for scientists all around the world to use within 24 hours, and then moving from there."
- **Amy Abernethy** [0:03:04]: "we hit the wall of evidence generation where we went back to circa 1995 clinical trials to actually try and understand which of the mRNA vaccines worked for whom and when."
- **Amy Abernethy** [0:03:35]: "we leveraged, for example, data from the Israel health systems to understand duration of effectiveness and safety signals and do you need a booster."
- **Amy Abernethy** [0:07:13]: "We called this the evidence accelerator. And regulators and others could say, I trust this, I don't trust that, here's what I want to know more about."
- **Amy Abernethy** [0:10:56]: "Friends of Cancer Research promulgated that thinking across the industry so that it wasn't just any one company, but rather all innovators having the chance to say what they thought."
- **Amy Abernethy** [0:11:30]: "That then informed the guidance that came out from FDA at the end of 2021. There were three core guidances around robot data, all informed by that process."
- **Amy Abernethy** [0:13:36]: "there's some parts of the application of AI in the healthcare and life sciences landscape that don't really create a bunch of angst. Identification of targets is a good example of that."
- **Amy Abernethy** [0:16:11]: "That was where my first real job was in 1982. And so that's a long time ago."
- **Amy Abernethy** [0:15:39]: "the pace of development in artificial intelligence ultimately happened at the end of 2022"
- **Amy Abernethy** [0:18:08]: "We can regulate them according to the traditional device framework, whether that's a diagnostic or a software as a medical device."
- **Amy Abernethy** [0:19:11]: "And I think it was like around January 2021, put out an action plan to specify, here are the details of what we think good's going to look like in this space."
- **Amy Abernethy** [0:19:11]: "We need, you know, good software design. We need to make sure that there is an easy way to understand the output of an AI-based product"
- **Amy Abernethy** [0:19:11]: "we need appropriate validation testing and we need some mechanism for performance monitoring across time."
- **Amy Abernethy** [0:19:45]: "that actually hasn't translated to a new rule set, as far as I know, beyond what we've previously had."
- **Amy Abernethy** [0:20:47]: "you ultimately package up the clinical trial, write hundreds of pages of reports, and ultimately ship that over the FDA and wait for questions to come back."
- **Amy Abernethy** [0:21:23]: "we were building the pipes at FDA to be ready for a landscape where really all you do is push your data across over FDA and FDA is ready to receive"
- **Amy Abernethy** [0:21:52]: "a foundational model of reading a chest X-ray. And then that's already done, a check for the foundational model."
- **Amy Abernethy** [0:25:37]: "pattern recognition and statistical programs were written about 10, 15 years ago to start to do that and better target the work of monitors."
- **Amy Abernethy** [0:26:33]: "our goal is to unlock the promise of precision health for everyone every day."
- **Amy Abernethy** [0:27:01]: "software for clinical research sites, software to conduct clinical trials, patient recruitment, as well as products in the healthcare delivery ecosystem"

## Transcript

**[0:00:01] Olivia:** Hello, and welcome to BioEats World from A16Z. I'm your host, Olivia, and I'm joined by Chris, a genetic engineer, science communicator, and, drumroll please, our new co-host.

**[0:00:13] Chris:** Hey, everyone. It's great to be here. Today, I have the pleasure of introducing our guest, Amy Abernethy, former Principal Deputy Commissioner at the FDA and now the President of Verily Life Sciences, who's here to help us understand how AI might be leveraged to streamline drug approval processes.

**[0:00:29] Amy Abernethy:** I kind of have this crazy thought that if we can have AI help write code, right? So AI helping to write software, then practically speaking, you know, the idea of AI helping us be more efficient in scale as we regulate doesn't seem all that crazy either, right?

**[0:00:48] Olivia:** AI is not just a sci-fi concept anymore. It's weaving into the fabric of healthcare, revolutionizing everything from research to patient care. It has the potential to create more streamlined and efficient processes. But the challenge now is how we adapt and regulate this ever-evolving technology while ensuring safety and trust.

**[0:01:05] Chris:** Exactly, Olivia. How we ensure the entire pipeline from discovery to testing to approval to implementation occurs efficiently. And Amy is no stranger to fostering new technology. During her time as commissioner, she oversaw the FDA's day-to-day functioning, rethinking how innovations flow through the administration.

**[0:01:23] Amy Abernethy:** Help the FDA develop an internal way of thinking about how to look at new innovation and develop a set of expectations and questions to get to innovators and say, here's what we're looking for. Help us figure out what is credible.

**[0:01:45] Olivia:** Now as president at Verily, Amy is focused on developing new algorithms that leverage AI to deliver the promise of precision health.

**[0:01:52] Amy Abernethy:** So it's obviously a part of our DNA to think about algorithm development and how do we leverage AI in all directions of what we're doing.

**[0:02:01] Olivia:** This balance between innovation and regulation is vital. And having leaders like Amy Abernethy is more important than ever. I'm Olivia.

**[0:02:08] Chris:** And I'm Chris. And you're listening to BioEats World from A16Z.

**[0:02:14] Vijay Pande:** Amy, thank you so much for joining BioEats World. It's terrific to be here with you again. That was great to have you, especially from the voice of someone who's been inside and outside regulation and thought about innovation. Where do you think we should start to try to innovate?

**[0:02:30] Amy Abernethy:** If you only innovate in one sector or part of the healthcare ecosystem, and you don't in a macro way say, well, it has to keep a pace in order for progress to continue, then essentially clog things up, right? So we can have an incredible discovery engine like what we saw in the context of COVID, right? I mean, the mRNA vaccines and their development of the vaccine itself so fast was because of all the science that had come before.

**[0:03:04] Amy Abernethy:** And that was just like literally miraculous. You know, essentially sequencing the genome, publishing it for scientists all around the world to use within 24 hours, and then moving from there. And then boom, we hit the wall of evidence generation where we went back to circa 1995 clinical trials to actually try and understand which of the mRNA vaccines worked for whom and when. And there was very little innovation truly in how clinical trials were designed and done.

**[0:03:35] Amy Abernethy:** We got smarter on the innovation side as the mRNA vaccines got out to real people. And then we leveraged, for example, data from the Israel health systems to understand duration of effectiveness and safety signals and do you need a booster. So some innovation there. But if you just look at that arc around COVID, what you see is that the innovation pace was different in different parts of the arc of developing and deploying mRNA vaccines. And if we don't actually stop to figure how do we push all of the sections forward?

**[0:04:13] Amy Abernethy:** We're going to continue to be in this zone. We have, for example, spectacular investments and output on the biologic side with slow pace on the clinical development side and even slower pace on the implementation side, right? So we actually have to have innovations on all of those sectors. What I hear you saying is basically,

**[0:04:35] Vijay Pande:** it's everyone's job to try to think about the full path of innovation, right? Otherwise, you will get at best clogged, right? That you'll have all these things sitting, waiting for regular approval. And that doesn't get this to patients. That doesn't actually achieve what any of us are trying to do. Right. And you just said something really

**[0:04:52] Amy Abernethy:** important, at best clogged. At worst, you ultimately either have no regulation at all, which means not only chaos, but real risk, right? Or you have such over-regulation that you actually totally take away any incentive to invest whatsoever and innovation just stops, right? Seeing how the whole system moves and then something else you said that's really important is it's all of our responsibility. And so if you're a life sciences company, you need the FDA to say, yes, this is okay, right?

**[0:05:31] Amy Abernethy:** If you're the FDA, you have to have some level of confidence that the results are going to be credible, at least in line with what you would have expected from formalized clinical trials. So as the innovator or the developer, I had the responsibility to figure out, like, how do I make it to be as credible and how am I going to demonstrate that? But of course, there wasn't a path to say, well, here's what demonstration looks like.

**[0:05:56] Amy Abernethy:** So it became my job to figure out how do I set the course? Like what kind of experiments are we going to need to do? How do we write up the results so people understand them in the right way? How do we actually even talk about this? And what's our lingua franca? And if I didn't actually stop to try and figure that out, then, you know, practically speaking, we wouldn't have been able to push things forward because we needed to get it to the FDA and have the FDA say, yes, no, this is what I trust.

**[0:06:21] Amy Abernethy:** This is what you can't do. In order to then have life sciences companies to say, ah, I see why it's credible. My core stakeholder, the FDA will be okay with this. And this is why I'll partner with this, you know, small tech company and buy their stuff. My next question became, how do you go to the FDA and start, you know, essentially spreading that kind of thinking across multiple areas. Help the FDA develop an internal way of thinking about how to look at new innovation and develop a set of expectations and questions to get to innovators and say, here's what we're looking for, help us figure out what is credible, but also build a crucible, so to speak, to bring

**[0:07:13] Amy Abernethy:** FDA and innovators together in close approximation so that innovators could say, here's what we're thinking about, here's what we're doing. We called this the evidence accelerator. And regulators and others could say, I trust this, I don't trust that, here's what I want to know more about. In quick succession, as you start to develop an understanding around the capabilities, like what we did at Flatiron, then what you can say is, all right, what if I now jump the line and go, for example, to the agency and figure out how do we take that process and scale it up, not just for real world data, but for many other areas of innovation too.

**[0:07:50] Vijay Pande:** Yeah. Well, it sounds like you're describing this ideally for all areas in innovation in principle, right?

**[0:07:56] Amy Abernethy:** Every innovation obviously has its unique features in terms of the science, the opportunity, the risks, and that practically is something that has to be applied within the context of a unique context. However, I do think that there's some core principles that show up that we can apply over and over again. So one of those core principles is that the innovators are the closest to their innovation. And provided there's healthy objectivity, right, the ability to say, here's what good in my space looks like.

**[0:08:36] Amy Abernethy:** Here's how I'm going to validate and pressure test in a transparent way that others can now cross-check my work and start to develop the framework for what good looks like in the communication thereof. That's one set of principles, right? And that starts to develop across a space. And oftentimes we'll talk about across industries, starting to think about that. Another set of the principles that is important in innovation, innovation is happening so fast. So if we don't have the chance to build what I call regulatory familiarity, right?

**[0:09:10] Amy Abernethy:** The understanding of what's potentially coming down the pipe and a set of questions of, and what am I going to be worried about when this starts to hit, right? Then that can take a really long time for those things to develop unless you build an instance for those conversations to happen really quickly. And so that's the second core principle is short circuit the time for regulatory familiarity to build the questions around the innovations to build and also that sort of comfort and understanding, including then a set of expectations of what good will look like, ideally informed by prior experiments that have said that this is what it looks like.

**[0:09:56] Amy Abernethy:** So that's a second. And then the third is ultimately step back and let the regulators or others who are responsible for now translating that into common guidance that everybody follows, let them do that part of the work, but then comment on it as an industry to say, this is what's working and this is what's not. Don't just let, for example, FDA put out draft guidance and act like it doesn't matter to you or your commentary doesn't matter.

**[0:10:27] Amy Abernethy:** Because in fact, that guidance now is the signals or the guideposts to industry of what to do. It should have been informed by those two processes coming beforehand. and as industry, take the time to care and comment and really make sure that you understand what this information means for your company.

**[0:10:50] Vijay Pande:** That is such a beautiful framework, but it feels like we haven't gotten there quite yet.

**[0:10:56] Amy Abernethy:** What parts are keeping us from that? Yeah, I think it looks different for different parts of industry. I certainly think about this in clinical trials and algorithm development in other places. So, the work we were doing at Flatiron and others really developed the basis for which we understood what good looked like and how to communicate about it. Friends of Cancer Research promulgated that thinking across the industry so that it wasn't just any one company, but rather all innovators having the chance to say what they thought.

**[0:11:30] Amy Abernethy:** And that work at the time, forensic cancer research actually acted as the crucible. They acted as the organization that got the regulators and the innovators together working very quickly. That then informed the guidance that came out from FDA at the end of 2021. There were three core guidances around robot data, all informed by that process. and then it became again incumbent on us to go back and cross check and look at those guidance in great detail and say you know you might want to rethink this particular section or even that sentence it may seem like minutiae but getting that sentence right is going to be really important for progress going forward and so I do think we've seen a play

**[0:12:20] Amy Abernethy:** out and it's what's interesting to me is that we're seeing this play out right now in AI, right?

**[0:12:27] Vijay Pande:** Well, let's take the AI area. It's very dear to my heart. Yes. AI can come into this in a couple different ways, right? There could be AI for discovering new targets or AI for getting to an IND faster or AI in principle for improving clinical trials and also AI for helping regulators. And I think we'll change healthcare and life sciences for Sure. But I actually, I bet a lot of people listening to this haven't thought about using AI to improve the regulatory aspect or to have it go hand in hand with how we're thinking about innovating the product.

**[0:13:01] Amy Abernethy:** Absolutely. I have so many different ideas in this space. One thing I'll say, Vijay, I kind of have this kind of crazy thought that if we can have AI help write code, right? So AI helping to write software, then practically speaking, you know, the idea of AI helping us be more efficient and scale as we regulate doesn't seem all that crazy either, right? First, let me kind of put a couple of core thoughts on the table to make sure that it informs this part of the conversation.

**[0:13:36] Amy Abernethy:** So there's some parts of the application of AI in the healthcare and life sciences landscape that don't really create a bunch of angst. Identification of targets is a good example of that. So on the scientific discovery side, we haven't gotten to the point of now taking a molecule and putting it into people. And so a lot of our work is safe from the perspective of not necessarily regulated yet and where AI can go to town, right?

**[0:14:10] Amy Abernethy:** Similarly, we don't spend a lot of time angsty around AI helping us have easier conversations, at least right now. Although if those conversations start providing medical advice and changing medical care, you start to shift into a very different space. right? But if we now talk about the role of artificial intelligence in the middle zone, so to speak, right, which is the zone from, you know, the early understanding around a molecule and getting into people and the clinical development as it moves through the process of understanding the performance of a molecule in people and then ultimately getting it to, for example, be prescribed, there's lots of places where there's nuance around how we think about artificial intelligence and want to cross

**[0:15:11] Amy Abernethy:** -check, for example, its performance. And the closer that we get to the AI being the product itself, where it's demonstrably intended to change health, to treat disease, diagnose disease, et cetera, get closer and closer to the landscape of being directly regulated as formally as a pro. And I mean, the fantasy,

**[0:15:32] Vijay Pande:** especially as a surrogate doctor, almost, right? Right. There's all these uses across the spectrum

**[0:15:39] Amy Abernethy:** and the intended use matters a lot for the complexity of the regulation. The second thing that I would say is that what's interesting, and this is kind of the next core principle. You and I both know, and I'm pretty sure most of these listeners, that the pace of development in artificial intelligence ultimately happened at the end of 2022, right? This has been, you know, incredible progress over decades. You know, I was in the artificial intelligence lab at NASA.

**[0:16:11] Amy Abernethy:** That was where my first real job was in 1982. And so that's a long time ago. And as I think about this, the innovations have been happening across time, but we've suddenly seen this massive visibility in our own lives in a way that's now suddenly making us think about the applications in every space within our lives. And that's as human beings and people who have kids that play soccer and, you know, who make dinner at night, as well as as biologists or clinical trialists or oncologists in a clinic.

**[0:16:47] Amy Abernethy:** So that's my second core point. So you need to think about regulation for intended use. Suddenly we're seeing AI everywhere. And so my third core point is that practically speaking, the landscape of availability in our minds of how AI is going to show up and be used within all the different contexts in healthcare has suddenly become a swarm of possibilities that's just like in every moment of the way we think about our day. And so you can imagine that because we have this rapid pace now of AI development and capability, rapidly expanding number of use cases, many of which where the intended use in healthcare is explicitly regulated, everything is expanding quickly and ramming up to a pipe that is

**[0:17:38] Amy Abernethy:** a pipe where there's going to be regulators sitting on one space with the responsibility of understanding. understanding which AI-based products one can trust and should be used at scale within the

**[0:17:51] Vijay Pande:** context of human health. How would regulators take something as complex as an LLM, and especially as complicated how it's trained, and get to the point where they can come up with a regulatory framework that they feel like will make sure that patients see the benefits of these technologies?

**[0:18:08] Amy Abernethy:** And I am not an FDA anymore. I'm not technically a regulator. I cannot tell you what's in the FDA's mind. I can sort of come at this from what have they put out there and also how might we think about this problem a bit. If we think about AI as a device, we have AI-based devices in our daily healthcare spectrum right this second. And there already is a framework for regulating these things. We can regulate them according to the traditional device framework, whether that's a diagnostic or a software as a medical device.

**[0:18:41] Amy Abernethy:** But that being said, very importantly, what's really coming at the agency is two things. One is that sort of speed, right? Like the massive number of AI-based products. And then the algorithms can learn, right? It's different when you think about getting to a fixed algorithm and a fixed output, and that's what gets regulated and only gets updated as there's a formal update process. But what if we talk about an algorithm that's now learning across time?

**[0:19:11] Amy Abernethy:** FDA has been working on, you know, how to think about regulating that. And I think it was like around January 2021, put out an action plan to specify, here are the details of what we think good's going to look like in this space. We need, you know, good software design. We need to make sure that there is an easy way to understand the output of an AI-based product and it's gone through appropriate user design, thinking we need appropriate validation testing and we need some mechanism for performance monitoring across time.

**[0:19:45] Amy Abernethy:** I've set off this, but that actually hasn't translated to a new rule set, as far as I know, beyond what we've previously had. The principles of what good's going to look like had been set forth, but the guidance of therefore this is what you do was still forthcoming. And so I think the interesting place where we are is sort of a speeding multiple trains all coming for the same little tunnel. And we have to come up with how we're going to make the tunnel bigger.

**[0:20:16] Amy Abernethy:** And that becomes our huge opportunity.

**[0:20:19] Vijay Pande:** And as we were talking about before, it's probably not just something where you throw more bodies at this on the regulatory side.

**[0:20:25] Amy Abernethy:** It can't be something where we just throw more bodies at it. We need people who have the brains to think through, okay, how are we going to do this? What does it look like? How are we going to keep up with and cross-check the work so that it keeps people safe and improves health? So we have to have the people who know how to do that. But we're going to have to build the efficiencies along the way.

**[0:20:47] Amy Abernethy:** And there are a few places we can build efficiencies. One is we probably need to build efficiencies in the regulatory review process. What parts of that can just be automated, right? Another efficiency that I tried to prepare for when I was at FDA was the efficiency of developing the validation output in the data that needs to be reviewed. Right now, if you're doing, for example, a clinical trial around a product, you ultimately package up the clinical trial, write hundreds of pages of reports, and ultimately ship that over the FDA and wait for questions to come back.

**[0:21:23] Amy Abernethy:** And so we were building the pipes at FDA to be ready for a landscape where really all you do is push your data across over FDA and FDA is ready to receive and do the data chunking and cross check on the FDA side. So that's another place where I think that there's like huge opportunity. Another place I think there's huge opportunity is to break apart the tasks. You know, one of the ways that you improve throughput, right, is you actually break apart the tasks.

**[0:21:52] Amy Abernethy:** And so in regulation, can we think about, for example, regulating foundational models and having that already be regulated? So, for example, a foundational model of reading a chest X-ray. And then that's already done, a check for the foundational model. And then the specialized model that sits on top of that is all that needs to be cross-checked. and that there's actually even a standardized package of what has to happen for the specialized model on top of the foundation model so that that can be handled much more efficiently.

**[0:22:27] Vijay Pande:** Well, this part seems especially important because if you have to start from scratch from a regulatory point of view every time, there's just no way we're going to get this done, right? On the other hand, there's a huge opportunity where you're just sort of testing

**[0:22:40] Amy Abernethy:** the parts that have changed in a sense, right? Yes. I think that there is a lot of opportunity here for thinking, and I'm just sort of giving you a sense of how I've been thinking through this, but you can set up a periodic cross-check of the entire system, right? So you can imagine setting up performance monitoring that cross-checks all the reading of chest x-rays to make sure that compared to some label data set, you're getting some consistency, and I've got a lot of different ways that I think that might look, so that you're not worried that if you've componentized all the different elements, you might miss something along the way or unintended consequences.

**[0:23:20] Amy Abernethy:** But you will be much faster if you are doing regulation around sort of base cases and then specialized regulation on top of that.

**[0:23:32] Vijay Pande:** If people are using AI to come up with drugs faster and even run trials better and come up with the documentation faster, you've got all this technology sort of leading that of clog you talked about, we will need to use something like AI to unclog it. Otherwise, we'll be stuck. And it's particularly interesting when you talk about doing data to data, because it does sound a little crazy that we're taking all this sophistication, putting it into hundreds to thousands of pages of English language that then gets re-read by AI to be able to have this direct connection where regulatory AI can sort of look at the data itself and ask its own questions and point to issues and so

**[0:24:11] Vijay Pande:** on that then presumably people then get involved and follow up. That feels like also just a huge leveling up of even what it means to be a regulator. It's in terms of sort of playing that much more, in a sense, in-depth role.

**[0:24:26] Amy Abernethy:** I 100 % agree. I always think about this concept of wick away the work. How do we leverage AI? How do we leverage computers and computation at scale to solve for the routine so that we put our brains to work on the most specialized tasks, right? And regulators are the same. I had this sort of aha in the clinical trial space where the people who are responsible for essentially going through and auditing sites and auditing data and auditing tasks and conduct of the clinical trial across, according to the clinical trial protocol.

**[0:25:06] Amy Abernethy:** So the monitors is often the way that they're called. When they're doing their work, they essentially do it in a very standardized process. so that you're going to audit a site on a certain schedule. You're going to look at a certain number of the data elements. The reason I'm telling you this story is, wait a second, don't we know where the highest risk places are? Can't we go ahead and leverage algorithms to tell us do more over there and do less over here?

**[0:25:37] Amy Abernethy:** And in fact, pattern recognition and statistical programs were written about 10, 15 years ago to start to do that and better target the work of monitors. And slowly but surely risk-based monitoring has become something that we talk about. And you can imagine now leveraging, for example, artificial intelligence to really do that at scale. And if we actually start to think about regulations similarly, because really it's, kind of another mechanism to cross-check, you could imagine that our AI helping our regulatory process for AI could help us get smarter about where to direct our attention.

**[0:26:20] Vijay Pande:** So Amy, I mean, that actually has a very natural question for how that connects to what you're doing today. So Verily is a part of an alphabet who's putting out all these really interesting LLMs. What does it mean for you today at Verily?

**[0:26:33] Amy Abernethy:** So it's obviously a part of our DNA to think about algorithm development and how do we leverage AI in all directions of what we're doing. But it shows up apparently, I think, in two ways. So we're very focused on precision health. And our goal is to unlock the promise of precision health for everyone every day. And precision health doesn't come around without having the development of more and more precise interventions and the better ability to match them to the people who need them.

**[0:27:01] Amy Abernethy:** So if we're going to pull that off, we need to start to build a data and technology bridge between research and care. And that's what we're very focused on is building products in the research ecosystem. So software for clinical research sites, software to conduct clinical trials, patient recruitment, as well as products in the healthcare delivery ecosystem, for example, software and services to help a person better manage their diabetes. and then having those start to snap together through data to be able to build this data bridge that's highly curated and also ready to both serve as a resource to build models of all types, but I think also equally important is serve as a resource to validate and continuously cross-check

**[0:27:51] Amy Abernethy:** the performance of models. And I personally believe that one way of unblocking the overall system is to make sure that you have underlying infrastructure that makes it ready to quickly make sure that high-quality models are getting developed and high-quality models are getting implemented and cross-checked.

**[0:28:13] Vijay Pande:** Fantastic. Well, Amy, thank you so much for joining us on BioReach World.

**[0:28:17] Amy Abernethy:** Thank you very much, and I look forward to seeing you soon.

**[0:28:26] Olivia:** Thank you for joining BioEats World. BioEats World is hosted and produced by me, Olivia Webb, with the help of the bio and health team at A16Z and edited by Phil Hegseth. BioEats World is part of the A16Z podcast network. If you have questions about the episode or want to suggest topics for a future episode, please email bioeatsworld at A16Z.com. Last but not least, if you're enjoying BioEats World, please leave us a rating and review wherever you listen to podcasts.

**[0:28:50] Olivia:** The content here is for informational purposes only, should not be taken as legal, business, tax, or investment advice, or be used to evaluate any investment or security, and is not directed at any investors or potential investors in any A16Z fund. Please note that A16Z and its affiliates may maintain investments in the companies discussed in this podcast. For more details, including a link to our investments, please see A16Z.com slash disclosures. Thank you.
