Inside Method

Generative AI in Healthcare

• Method • Season 6 • Episode 5

Share episode

Generative AI seems to be a hot topic in every field but what role should it play in Healthcare?

In this episode, we join Scott Pope, Method’s Global lead of healthcare and life sciences, Dr. Gareth Jones, a senior data designer here at method, as well as Andy Busam one of method’s Principal Strategists to discuss the current efforts to leverage Generative AI in the healthcare space as well as the hopes for where it could lead.

If you are interested in learning more about this topic, our guests have provided a white-paper which is available for download on our blog page here.

Chapters:

00;04;17;11
1. context setting and historical analogies

00;09;36;16
2. What we're seeing in the market

00;14;08;14
3. Transparency

00;17;36;15
4. Trust

00;20;52;00
5. Generalizations

00;25;04;20
6. Method's perspective

00;30;10;25
7. Validation, safety, and efficacy

00;00;00;00 - 00;00;25;18
Unknown
When we talk about machine learning and artificial intelligence, we primarily think about even descriptive or prescriptive. Where I feel like generative AI differentiates from that is that starting point is a little bit more opaque because we are creating something entirely new. You are listening to Inside Method, the podcast that takes you behind the scenes of a global consultancy.

00;00;25;20 - 00;00;50;13
Unknown
In this episode, we join Scott Pope Methods Global Lead of Health Care and Life Sciences. Dr. Gareth Jones, a senior data designer here at METHOD, as well as Andy Newsome, one of our principal strategists. The topic of discussion Generative A.I. and Health Care. If this topic is of interest to you, then you have come to the right place because our three talented teammates have a lot to share on the subject.

00;00;50;17 - 00;01;13;12
Unknown
And if you'd like to dig in even deeper than what this podcast provides, be sure to check out the white paper they are making available as well. We will link to the details for that in the show notes. Now, I don't want to take up any more time, so let's jump in here from them and learn about generative AI in health care.

00;01;13;14 - 00;01;46;25
Unknown
Thanks for joining us on the Inside Method podcast. Today we're talking about generative AI in health care. My name is Scott Pope. I'm the global lead for health care and life sciences here at Method. And we've got an outstanding line up panelists today to lead us through a conversation on generative AI in health care. Before we jump in and allow my guests to introduce Themselves and abuse segment and Gareth Jones, we're going to this topic of generally AI and health care is now we've really hit a tipping point here recently where it is absolutely the talk of the town, if you will.

00;01;46;28 - 00;02;08;27
Unknown
We're going to do a little bit of context setting, but there's kind of three main things that we're going to talk about as it relates to the topic today. One, we're talking about some some real life examples of things that we're seeing with AI and generative AI in health care. We're going to cover some specific nuances relative to health care above and beyond what we see in in other industries like retail and consumer and banking.

00;02;09;00 - 00;02;32;09
Unknown
And then specific to the white paper perspective paper that my panelists have have written, we're going to get into some of the perspectives and viewpoints that they've shared relative to how method is approaching AI in health care. So with that, it's my privilege to welcome any person here in the room with me and Gareth Jones all the way across the pond in the in the UK.

00;02;32;16 - 00;02;49;26
Unknown
Gentlemen, thank you so much for joining. Thanks, Scott. It's great to be here, Ioannis. So for our listeners, if you guys could please give it a little bit of introductions to yourself. Any since you're here in the room, I'll let you you kick us off. Sure thing. Thanks, Scott. As you mentioned, my name interview some. I'm a principal strategist here at Method.

00;02;49;29 - 00;03;25;04
Unknown
I've been with the firm for around four or five years. I've spent time working across our portfolio business across a range of clients, but most recently focused on our health care clients. And I do have a background working within health systems, but also in the pharmaceutical industry and most recently within the past year or so, really digging into our data practice with with Gareth as well as our colleagues at Global Logic around AI, as well as generative AI, and particularly excited about the application of those technologies within within health care.

00;03;25;06 - 00;03;53;18
Unknown
Yeah, excellent. Thanks so much for being here, Andy. Gareth, would you introduce yourself to our listeners, please? Yeah, thank you, Scott. My name is Gareth Jones. I'm a data designer based at Memphis London office, so I completed my Ph.D. in the application of machine learning classifiers to arterial disease detection. The focus of my research was really to look at if we could use machine learning to detect consistent and significant biomarkers of arterial disease within easy, obtainable peripheral measurements of blood pressure and flow rate.

00;03;53;20 - 00;04;12;22
Unknown
I then went on to help develop computer vision models that could detect early signs of coronary artery disease and CT scans. Since working for method, I spent a lot of time looking into how we balance information with comprehension when using data driven insights and tools to help clinical decision makers. So, Dr. Jones, thank you so much for that introduction.

00;04;12;22 - 00;04;37;00
Unknown
It's a pleasure to have you on here all the way across the pond there in the UK. So real quick context, setting the adoption of new technology and in health care isn't really a novel concept. We've been adopting new technologies, everything from penicillin to transplants to you name it. There's a ton of new technology that gets adopted in health care on a regular basis.

00;04;37;02 - 00;05;00;12
Unknown
Sit a little more context for us, Andy, in terms of how you see this really fitting into place, air adoption in health care. Sure, you're right, Scott. Technology progression within the health care space is certainly nothing new. I think we can go as far back to the 1950s to see some of the earlier applications of data, data science applied within the health care space.

00;05;00;19 - 00;05;27;20
Unknown
The progression of that has grown exponentially, especially since the 2000 and and within the last decade or so. I would actually love to hear Gary's perspective on this, just given the academic research, as well as the practical and professional research that you've done within the space and what you've seen grow technologically within health care. I think health care is, as always, kept up really well with progressions in machine learning and AI.

00;05;27;22 - 00;05;52;24
Unknown
Academically, there's been massive strides over the last, as you say, 20 years, 30 years ish. And generally health care has as utilize that. And my background is in creating computer vision models, right? So at the same time as the analog press elsewhere, like self-driving cars, health care is also utilizing it to its full potential. So I don't think is necessarily an industry that's like I think there's a lot more that could be done.

00;05;52;24 - 00;06;24;14
Unknown
However, it's not as if the saw behind it. Yeah, I think it's a great point and I think that what we're seeing now is an increasing number of challenges facing the health care industry at large and these technologies as a potential solution to help solve those. You know, a couple of the statistics that really stood out to me recently doing some research were that 50% of errors within the documentation space within health care are from administrative reasons.

00;06;24;16 - 00;06;48;12
Unknown
The cost from discovery to drug development within the pharmaceutical space average right now is about $1.8 billion. So if you think about some of these challenges that exist, the amount of money that's being spent, the amount of time that's being spent on delivering care to patients, the number of errors that that happen as a result of the of human manual process in the mix.

00;06;48;15 - 00;07;13;22
Unknown
I think that these technologies have a real opportunity to help us deliver better patient outcomes. You know, and it's a great lead in to some of the things that we're going to talk about in a little bit more detail here as we move forward. One of the one of the interesting pieces that you guys talk about a little bit in the in that perspective paper is the difference between just traditional artificial intelligence and generative artificial intelligence.

00;07;13;22 - 00;07;40;20
Unknown
You guys can choose who it wants to tackle that one, but help our listeners understand kind of that, that core concept. Yeah, as someone with with not the robust background as Gareth, my layman's term here or layman's definition for the shift that's happened is that artificial intelligence helps us synthesize and analyze data generative. I can actually create or generate as a result of analyzing that data.

00;07;40;23 - 00;08;04;11
Unknown
So I think that what we're seeing in the spaces is a new capability as a result of generative AI and not just the ability to analyze, synthesize and present insights. But Gareth, would you mind sort of unpacking that a bit more? Yeah, I think you covered it really well. And when we talk about machine learning and artificial intelligence, we primarily think about even descriptive or prescriptive.

00;08;04;12 - 00;08;26;12
Unknown
So descriptive being like we have data that what can we extract from it? Descriptive being, how can we draw out insights? It may not immediately be assessable from that data, but in both of these cases, users have this really clear is this data has the end insights. And although that middle piece might be a little bit missing, they can kind of see the two bookends on it.

00;08;26;14 - 00;08;48;00
Unknown
Where I feel that generative AI differentiates from that is that starting point is a little bit more opaque because we are creating something entirely new and so it's a little bit slower to build adoption and trust because we can't necessarily see, Hey look, here are some data and here's my outcomes, although we have the prompt. So we have something that actually triggers that generative AI.

00;08;48;02 - 00;09;18;18
Unknown
How that has come to the AI is a little bit different and it's a bit more abstract for clinicians, I think. So I think that's where we really start to differentiate, as you say, is we are now actually producing something new rather than just mining available data. So what might be in that growth? You touched on a really important term there with trust that we're going to unpack a little bit more greater detail here as we move forward in the conversation to the two of you wrote a perspective paper on gender of AI in health care that I had the privilege of getting a a pre read on.

00;09;18;21 - 00;09;41;20
Unknown
Our listeners can download that method dot com slash insights depending on when they're listening to this publication of the podcast. It may be out. It may still be a final final steps of production there but method dot com slash insights for that your perspective paper on gender of AI in health care. As you guys talk about in that paper there's already deployments of AI in health care.

00;09;41;20 - 00;10;05;18
Unknown
This is as you've kind of both alluded to, it's not necessarily anything new. I think it's certainly more in the spotlight recently than it has been. But give me a just a, you know, one or two examples of where you've kind of seen AI being utilized in in the current health care setting. Some of those companies that you see that are really good examples that jump out for our listeners to understand how it's being deployed.

00;10;05;20 - 00;10;32;19
Unknown
Sure. I think that one particular area that we've seen a couple of companies jump into because of its potential is around essentially ambient audio recording conversations between providers and patients. And so just to list a couple of companies here, a bridge is a startup in this space Nuance, which is a microsoft subsidiary around for a long time. Absolutely.

00;10;32;19 - 00;11;02;15
Unknown
And Amazon has just launched a new product as well in this space, all all sort of doing the same thing. But essentially what this what this use case or application does is it records the audio in a room between a provider and a patient. It's secure and encrypted, of course, and it transcribes that audio and it sends a copy of that into the EMR as well as to the patient via a mobile app.

00;11;02;15 - 00;11;26;04
Unknown
In a bridge's case. I can't speak to to nuance specifically, but a bridge helps generate a care plan as well as the treatment recommendations from the provider and delivers that in a user friendly way. And the goal there is to to shorten the space between the visit and the patient, sort of understanding what happened as a result of the visit.

00;11;26;04 - 00;12;02;28
Unknown
So they've got more visibility and transparency, a clearer record of what happened in that interaction, as well as giving them directly from the provider. This is the course of action for you. This is a treatment plan. This is this is what we expect. And the goal there is to, of course, increase adherence to what the providers prescribed. Yeah, and you touched on a point there that is a recurring theme for us here at Methot around health care, is that in a consumer experience and our listeners be able to enjoy more content from us on that front coming up in the very near future.

00;12;03;01 - 00;12;23;27
Unknown
Gareth There's another example that kind of prep for this episode that we pulled out. Talk a little bit about some some synthetic data for from a research perspective. You got to a deep background there and I think it really gets to the difference that we spoke about between artificial intelligence versus generative AI that is actually creating a net new product.

00;12;23;29 - 00;12;47;14
Unknown
Yeah. So as a major component of my PhD, I spent about two years generating synthetic data. This was an incredibly manual process. We had to get computational models that replicated, but so we had to get large amounts of statistical data on watch the distribution of material networks be across the population. And then we had to essentially sample citations of material networks.

00;12;47;14 - 00;13;13;08
Unknown
So these equations I see we had supercomputers running for about two months to generate this. So it's a massive list. It just makes it very infeasible outside of a case like a university where you can dedicate a year or two years to the pure pursuit of creating this synthetic data generative A.I. really allows that to become much more accessible, much more streamlined, allows the removal of so much of that manual process.

00;13;13;08 - 00;13;42;13
Unknown
And that was really high computational components of that process. I got pretty strong feelings about synthetic data, I think is a great tool for really early exploration. There's so many nuances and caveats and restrictions around health care. So if we can generate this data really quickly, really inexpensively, it's going to empower better research, more research, you know, to push the envelope by just getting our hands on things which normally aren't accessible, take a very long time to get to us.

00;13;42;15 - 00;14;10;04
Unknown
I think really powerful deployments on both those examples and I think a space that health care is in, in desperate need of advancement. Some of these domains that I really I'm excited about how this is going to be deployed and watching some of the the beneficial impacts again for our listeners, my my guest on the Inside Method podcast, their interview segment, Dr. Gareth Jones, you've written a great white paper on Gender VII in health care that's available method dot com slash insights specific to health care, right?

00;14;10;04 - 00;14;32;23
Unknown
We see aid being deployed across a number of different industries in France, but health care brings some really unique considerations and circumstances. We've talked about these kind of in prep around your transparency and trust generalize ability and then some of the regulatory components. What's going to maybe attack those in in order in a do you want to start maybe talk about some of the.

00;14;32;26 - 00;14;55;00
Unknown
Yeah. The transparency of how the air engines are arriving at some of the the content that they're creating and having transparency to that why it's relevant even more so in health care than anywhere else. Yeah, absolutely. So I think what's unique in the health care context is, is that we are dealing with high risk situations in many cases, especially when we're talking about patient care.

00;14;55;02 - 00;15;29;18
Unknown
And so if we have an AI model deployed in an instance that's going to affect patient care, I think we have to be especially careful. And that's where the growing field of explainable AI or AI you may see is really coming in to play a strong role. And that's essentially another AI model that runs alongside the primary model, that helps provide an explanation for how a recommendation or an output from the primary A.I. model came to be.

00;15;29;21 - 00;16;02;02
Unknown
And so there's a growing body of research into what are the best kinds of explainable AI models, what is the best user friendly way of presenting that information that is going to be adopted well within the health care system? And so I think it's especially important within, say, clinical decision support systems. Right? So so Microsoft's acquisition of open AI has has given them the ability to integrate with Epic, which is another use case that we're seeing out in the marketplace.

00;16;02;02 - 00;16;48;04
Unknown
And so Openai is being deployed within some instances of EPIC. They're running pilots currently primarily within clinical decision support systems as a way of looking at the all of the data that's coming into the h.r. And generating a prescription recommendation or treatment plan recommendation. And what i think is incredibly important for us as technologists, but certainly for health systems who are thinking about implementing these technologies is if a recommendation is shown or prescription is shown that the provider can see it and say, okay, but how did I get to that recommendation?

00;16;48;06 - 00;17;04;16
Unknown
And so we've got to be able to show those explanations and show our thinking. I think that we all we all probably heard that from a math teacher at some point, Right. It's not enough just to to circle the answer at the bottom of the page, but I need to see how you got there. And that's still true in this past year.

00;17;04;18 - 00;17;26;12
Unknown
Yeah. So as a pharmacist, you made a large number of clinical recommendations to physicians over the years. The number of times that I had to kind of back up those recommendations. I mean, it's very much a real world scenario and where you expect some of that in your human interaction, you're absolutely going to expect that in a in artificial intelligence interaction and maybe even more so.

00;17;26;15 - 00;18;03;02
Unknown
And we'll get into generalizability there, like kind of. All right, where did where did this information come from? What data set was loaded into that AI engine? Well, we'll get to that in a moment. Gareth, you talked earlier about the concept of trust. I saw an example of a deployment of AI earlier this week where an AI engine is recreating QR codes and instead of just kind of being that inkblot with, you know, square blocks on it, it's actually creating like beautiful pictures that has a QR code and well, that's a really kind of cute, neat deployment, artistic deployment of AI.

00;18;03;05 - 00;18;24;09
Unknown
Yeah. You know, the standards that we would accept there, you know, from a trust perspective or a cute tik-tok video versus what we expect in health care are wildly different. Can you talk about some of the trust components relative to AI and health care? Yeah, I think there's really two lenses of trust. We've got to think about which it clinician and decision maker and then also the patient trust.

00;18;24;11 - 00;18;52;27
Unknown
So from a patient they need to know that they're getting the best treatment they can end up being looked after properly and that involves an algorithm. They need to know that it is capable of making a good decision or recommended a good decision. And then from clinicians perspective, is balance in not control versus their accountability, right? Like if at the end of the day the buck stops at a decision, they need to know that anything that helps them get that decision can be relied on.

00;18;53;00 - 00;19;10;25
Unknown
I would say from that for me a very important element is we should always aim to keep clinicians with a hand on the wheel. They should always feel in the loop. We should never try and remove them from it. As soon as we start doing that, it makes it a lot harder for them to trust it. Leaning into and his previous point of transparency, right.

00;19;10;25 - 00;19;29;09
Unknown
If they're just getting an end of the line recommendations without any oversight into what's happening in new metal, trust becomes much, much harder. And so we really need to ensure people feel that they can rely on this. So we get good implementation and good utilization, right? If we have really good tools, but nobody uses it, we're not going to make an impact.

00;19;29;09 - 00;19;56;20
Unknown
So we need clinicians and we need patients to trust what we're building. Yeah, I think there's also some really interesting behavioral psychology research within this space as well. There was a study published a few months ago out of the UK that essentially looked at explainable AI models and different forms of explainable A.I. models. So ways of explaining, here's how the I got to its recommendation.

00;19;56;20 - 00;20;20;11
Unknown
Yeah, and one of the goals of the study was to see which way of explaining is most effective. And so there were, there were some results of that. But the thing that was most compelling and interesting to me was merely showing that there was an explanation over time almost lulled the providers into believing it, regardless of the accuracy of the recommendation.

00;20;20;11 - 00;20;45;16
Unknown
Wow. So even if there was an explanation there, even if it wasn't a good one, its presence alone, it's convincing. Exactly. Exactly. So I think that we have to consider a range of factors when we deploy these systems. We've got to think about, you know, what is the workload of a provider as well as training and education for those providers on actually critically thinking about what is the information I'm seeing here?

00;20;45;16 - 00;21;07;18
Unknown
Yeah, merely seeing an explanation doesn't mean that it's accurate. Well, it's a perfect segue into the next topic there on generalizability. Right. So when you think about AI engines, if if an AI engine in, you know, in a normal course of your life is pulling from Wikipedia, that may be really useful information. But we're talking about making like clinical decision support stuff for for patient care.

00;21;07;21 - 00;21;37;20
Unknown
Very much an unacceptable platform to pull that in. But these AI engines have to be leveraging some data assets to come to arrive at this conclusion. So talk more about the generalizable building aspect there for us. I may defer that one too. Yeah. All right. That's fair. Yeah, I think it's really difficult. I know there's been cases where, for example, AI models trained in one country are about to be retrained when moving borders just from basic differences in demographics, prolific health problems, etc..

00;21;37;23 - 00;22;05;15
Unknown
So when we get into gender today, I think that becomes even more nuanced and even more complex where we might be drawing a lot of different very heavily contextualized datasets, right? Because when we think about things like text or images, there is a lot of context baked into that. And so we really need to think about what is the full utilization, where are we anyone using this and is that actually captured within the datasets we can make available?

00;22;05;17 - 00;22;24;05
Unknown
And that can become very, very hard because something clinicians are very good at doing is applying prior knowledge to new cases and taking what they know and using it and, you know, waste. And that is something that is always going to struggle to do because it is replicating what it's previously seen. And so I think there's always going to be nuances and caveats.

00;22;24;07 - 00;22;44;15
Unknown
I think the best we can do is to make people aware of the limitations and and not be naive to the fact that, hey, look, it's not going to work every case of the time, right? This is great at doing those really repetitive stuff that it's seen before. But if you bring in a really actually envelope cases you've never seen before, it's going to struggle that alone yourself, right?

00;22;44;18 - 00;23;07;12
Unknown
So we'd probably be negligent if we didn't. You know, when we talk about specific considerations for health care with the AI, if we didn't talk about some of the regulatory and privacy and security components, you know, everything from IDA to HIPA and certainly international privacy laws as well. Gareth, you want to kind of touch on that one at the outset and talk about some specific things for health care.

00;23;07;12 - 00;23;32;15
Unknown
And I think a big one for me I've had concerns about before is this worry of essentially hallucinating, personal private data right? Like these models learn by looking at previous data sets, by learning how to replicate that, the chances of it ever are incredibly low. However, we can't exclude the fact that if I might introduce data that seemed previously that might be sensitive in the case of health care.

00;23;32;17 - 00;24;00;20
Unknown
So I think we have to be very nuanced. I think you've got this roots that too. We need to have a human mind set a great way of reducing workload. However, we can never slowly move humans. And I think when we start to think about regulation and I see by ensuring we always have a human in the loop who is accountable and is checking this, I think we can alleviate a lot of regulatory concerns in anything to add on to totally agree with with Gareth.

00;24;00;20 - 00;24;36;24
Unknown
And this space is evolving, right? So I think that will see privacy advocates as well as government agencies sort of step into this space and offer perspectives and maybe even change some of the rules and guidelines within the space. I think that's this is one of the areas for synthetic data to really make a difference. So Integra is is a company within the space that's that's generating these synthetic data sets because they are it's essentially generated a data based off of real patient data.

00;24;36;24 - 00;24;58;27
Unknown
But it's it's all synthetic. At the end of the day, there are no real privacy concerns. So if you have a pharmaceutical company who's spending up a clinical trial and they need a data set, being able to use a synthetic data set allows them to expedite through some of the regulatory and compliance considerations that they that they normally would be slowed down by using real patient data.

00;24;59;04 - 00;25;31;25
Unknown
So so I think that that's an area certainly that will see growth. And so in your whitepaper that's available at Match.com slash insights, there's a number of perspectives that you pull out. One of my favorites that it's kind of a theme throughout the course, the entire perspective paper is this notion of not being a hammer, looking for a nail when methods specifically talking about how we are engaged in in helping helping our customers deploy age, gender, VII in health care, maybe I need to come to you for this one first.

00;25;31;27 - 00;25;53;15
Unknown
Talk about some of that approach of how we when a customer comes to us and say we need an eye project, come fix this for us, how do we approach that here? Method Yeah, I think at the end of the day we are a human centered design firm and so we really start with people. One of our principles is to fall in love with a problem, not a solution.

00;25;53;17 - 00;26;27;18
Unknown
So we don't walk into a room with a predetermined notion of You need this new technology. We're not here to sell a specific set of softwares. And we frankly sort of pride ourselves on the fact that we do spend time getting to know end users, customers, employees, providers, front office folks who caregivers, all the people who are going to be impacted by a potential digital product or service that we might deploy into the real world.

00;26;27;21 - 00;26;48;06
Unknown
And so we put humans first. And I think that that doesn't change even with some of this new technology. Yeah. So in some of those scenarios where, you know, the problem calls for I as a solution, very eager and capable of being able to do that. But not deploying AI for sake is not the way that we go about kind of solving that.

00;26;48;08 - 00;27;16;12
Unknown
Gareth, you want to elaborate it all on the kind of the that human first human centric element? I just, I find that it's such a, a recurring theme throughout everything that you guys put in your perspective paper and in certain all of our ongoing conversations here method with in the client work that we're doing. Yeah, I think I tell you from a very academic background into a very user centric business like method, something I had to learn quite quickly is there is no inherent value in data, right?

00;27;16;12 - 00;27;39;11
Unknown
It's all about getting the right data to the right people at the right time in the right format. And sometimes generative. AI is a great tool for doing that. However, it is only one tool we have in our arsenal, right? So we really do want to talk to users and stakeholders and understand. We often describe it as a cognitive journey from you get some input or some data and you have to make a decision or do an action.

00;27;39;11 - 00;28;00;08
Unknown
And how do you go about that and what are the steps you take and understanding where we can interject new tools, where we could interject new insights into that to alleviate any pain points they might have is kind of the first approach we do. And then we figure out what are the best solutions to do that. And as I say, I mean, it's generative is the right way of doing that is an incredibly powerful tool.

00;28;00;08 - 00;28;19;09
Unknown
In a lot of cases, but it's not without its caveats, as we've already discussed. And so if we could get away with something, for example, that basic statistical models and maybe not as powerful, but also alleviates a lot of concerns we have. And so we are very, very keen to find use cases where it's the right tool to use it.

00;28;19;11 - 00;28;45;08
Unknown
But we're also not afraid to suggest other options if they are more suitable. Yeah, Gareth, you know, one, one other piece that I think that you might be able to speak to here is around the fact that generative AI is really a it's a continuation of a broader data strategy that that a company should have. Right. So there are so many other decisions and building blocks that need to be put into place before generative.

00;28;45;11 - 00;29;06;05
Unknown
I can be a real solution for for a company. And so so I think that we also think about data strategy and a data practice for an organization on a maturity scale. So would love your thoughts on on that as well. Yeah, I would say if your first step in a data strategy is generative AI, it's like learning to drive a Ferrari, right?

00;29;06;12 - 00;29;25;18
Unknown
Like you need that base, you need that groundwork in place before you can think by really trying to implement those big tools. And I would say that a lot of additional value can just be gained by pretty simple data strategy, data platforms, some really simple visualizations or basic business dashboards before you start thinking about trying to augment that.

00;29;25;18 - 00;29;48;10
Unknown
Right. And just really simple optimization, really simple centralization, all of those best practices and then build on to that the more complex AI and generative AI packages that can come on top throughout the course. That perspective paper, you guys both and you've alluded to here in our conversation today to allude to the concept of AI really being a collaborative tool.

00;29;48;13 - 00;30;10;04
Unknown
You know, we talk about some of the administrative opportunities for deployment. We can, you know, let I kind of run on do some of those things. But when it comes to clinical deployments, really having it be more of a recommendation versus the AI itself making a clinical decision for that's directly applied to patient care. And I think that's that's an important piece that's worth kind of calling out here.

00;30;10;04 - 00;30;32;20
Unknown
But the last piece that you guys had in your in your perspective was around in the work that we do in the way that method approaches. AI in health care is around validation and safety and efficacy, which is really, really close to home for health care. I'll kind of let you guys choose who wants to tackle that one First is kind of the last piece and your perspective paper.

00;30;32;26 - 00;31;15;03
Unknown
Sure. I mean, I'll I'll start just generally validation or user testing is a process for us regardless of the of the nature of of the project. Certainly within a health care context, I think it is potentially more complex because of the number of stakeholders and the potential risk or outcome of a product in the market. And so before we introduce any new technology, we certainly want to put that in the hands of real users and we do our best to not have bias entering those conversations or observations to watch how those products are being used.

00;31;15;06 - 00;31;38;23
Unknown
And that for us is incredibly important to make sure that the intent that we set out with making a solution is actually in line with what happens in the real world. Gareth, anything to add on that one? An app, an ambition, identify ambition is the right word, but I think we should be striving for clinicians spending 80% of their time doing those most impactful 20% of use cases or jobs, right?

00;31;38;26 - 00;31;59;22
Unknown
Really trying to take that mundane, low risk, tedious jobs away from them to allow them to do the most impactful, the most meaningful things they can do with their time. And I think I and generative AI are a great way of trying to really shift that balance. I think at the minute it's an emergent care. 75% of time is on administrative tasks, right?

00;31;59;24 - 00;32;20;13
Unknown
I think assessing we need to do and that is to actually understand, okay, well where are these low risk where these tedious jobs that we can remove them, we can automate and we can use AI intended to AI, which comes down to a lot of validation. It's actually going into into people and understanding. And then as we put them into small groups of people's work load, actually seeing are we having the intended impact.

00;32;20;15 - 00;32;42;00
Unknown
We don't want to then introduce extra workload because they're having to oversee or be involved with these new models, these new tools that go into place. And so really these small scale tests. So, Ali, trusting it, are they understanding it? Is it actually removing workload? Is it having the impact we want are incredibly important before we think about rolling out anything at scale.

00;32;42;02 - 00;32;59;28
Unknown
Excellent. Andy, Gareth, you guys have been excellent guests on this episode of The Inside Method Podcast. I hope you both will come back soon. Again, for our listeners, method dot com slash insights to download the White paper and and get to learn a lot more about what's in Gareth put in terms of the perspective of air and health care.

00;33;00;06 - 00;33;18;03
Unknown
I'm Scott Pope, Global lead for Health care and life sciences method. Thank you for joining this episode. I hope you enjoyed this episode and if you need more method in your life, you can always find us on social. And don't forget to check out any of our monthly tech talks. They're available both in-person and virtual. We would love for you to be a part of those.

00;33;18;03 - 00;33;38;22
Unknown
You can find out more information about them on social media as well as on our website. Keep your ears open. We'll be back with another episode soon. But until then, don't forget to stay nerdy.