Digital First Scientific Communications: Generative AI: Changing the Future of MedComms
Speaker: Steve Casey
Speaker: Matt Lewis
When we look at AI, and in particular natural language processing, we think about how we can use these new tools in the medical information space. How do we see the new generative AI models changing what medical information has been doing, especially in the context of chatbots and information dissemination? In this podcast, we interview Matt Lewis, Chief AI Officer at Inizio Medical, a well-regarded expert in data analytics and augmented intelligence.
Following is an automated transcription provided by otter.ai. Please excuse inaccuracies.
Welcome to this episode of the Medical Affairs Professional Society podcast: “Elevate”. The views expressed in this recording are those of the individuals and do not necessarily reflect on the opinions of MAPS are the companies with which they are affiliated. This presentation is for informational purposes only, and is not intended as legal or regulatory advice. And now for today’s elevate episode.
Jennifer Riggins 00:32
Welcome to elevate the Medical Affairs professional societies podcast. As a series within this podcast we focus on digital first scientific communications. How digital is transforming medical communications. In these podcasts, we speak with experts in the field of scientific and medical communications, and discuss how digital transformation is opening opportunities for medical communicators. I’m Jennifer Riggins, a co-host of this podcast. I currently serve as a member of the Digital Focus Area Working Group. I’ve worked in pharma for more than 30 years, and I’ve had a focus on medical information, scientific communications and medical digital, and I currently work for phactMI and nonprofit consortium of medical information leaders. I’m joined by my co-host Steve Casey of Omni Healthcare Communications, or Omni HC for short. Steve has been in pharma for over 35 years and has led Omni HC for the last nine years to become a leader in digital first medical communications. And in our last series of episodes, we focused on omni channel in medical communications. We discussed omni channel in medical how it’s defined, implemented and measured. In that series, I think it became clear that medical omni channel is a communication dissemination innovation that’s empowered by today’s digital environment. Now we would like to turn our attention to a more recent digital innovation. And I bet you can guess what that is. Generative AI clearly promises innovation in the development and publishing of medical communications.
Steve Casey 02:05
In this podcast, we interview Matt Lewis, Chief AI officer at in Inizio Medical. He is a well regarded expert in data analytics and artificial and augmented intelligence. He’s actively working with MAPS and other societies to guide Medical Affairs with the incoming AI onslaught. Matt, thank you for joining us today. I know that my introduction doesn’t do you any justice. So could you give us a brief overview of your background and what you’re doing with the different societies at this point regarding AI?
Matt Lewis 02:35
Sure, Stephen. First of all, thanks to you, Jen, and MAPS for having me on today. It’s a real honor to be able to share my perspective with your audience at this really critical time in our profession. And before we start, I should just state that the following comments are my own, they don’t necessarily reflect those of my employer and have not previously been approved by Inizio. In terms of how I got here, I’ve been in life sciences my entire career 25 years. My background is a mix of molecular and cellular biology and Health Services Research peppered him with advanced analytics. I started out in commercial analytics and a company that’s now part of AB the and I have a bit of grounding as such in the fundamentals of commercialization, but my first of all in medical was as a medical science liaison. After working in the field, I worked in a variety of headquarter roles such as grants, patient engagement publications, advisory boards, key opinion leader and global opinion leader management, and medical excellence and worked on the launch of about 10 molecules and indications before transitioning to the solution provider cyberspace about 14 years ago. Since then, my work is focused on strategic planning, advanced analytics and artificial intelligence. I currently serve as you said, as the global chief artificial events and intelligence officer with Visio medical. On the society side, I’ve been really fortunate enough to be able to share our thoughts with MAPS leadership earlier this summer, on the promise and potential of AI to help medical really reimagine its future, and have been assisting the MAPS executive staff and the production of a forthcoming white paper on the topic alongside other members of the partners circle team, if you will. Beyond MAPS I co chair of the artificial intelligence Task Force for his map of International Society for medical publication professionals with Keith Goldman have the serving as the other co chair and this taskforce has developed an AI position statement which will be published shortly, and will likely serve as guidance for some colleagues in the space to adopt policy positioning to advance the field. I’ve also co authored the United Kingdom’s healthcare communications Association’s AI roadmap, which has a bit of a broader scope, but a similar aim to catalyze action in the space.
Jennifer Riggins 04:41
So hi, Matt, as Steve said, thank you again for joining us today. And that is, you know, a really great resume of what we’re going to be talking about today. So, when I look at AI, and in particular natural language processing, I think about how we can use these new tools In the medical information space, you know, doing things like drafting documents helping HCPs find information more easily, maybe implementing chatbots for both HCPs and patients. So how do you see the new generative AI models changing what medical information has been doing, especially in the context of chatbots and information dissemination?
Matt Lewis 05:20
Yeah, Jen. That’s a great question. I mean, I actually had this exact conversation with the Vice President Medical Affairs quite recently, just at the beginning of the month. It he this, this particular individual works in a very specialized therapeutic area, and his company gets inquiries from clinicians and investigators and other folks out in the world on an ongoing basis. And in addition to needing to provide relevant information, it has to be timely has to be current, it has to be actionable. And staying up to date with the literature is always challenging. But getting content for the approval process sometimes is even more. So we talked about the way in which AI has been able to help really up until the introduction of generative, which like came about with the GPT and the rest late last year, such as through conversational AI, which can extend and enhance the ability of human colleagues to be accessible to clinicians both at hours and on days, when our staff may not be in the office such as late at night or on the weekends. And also by focusing on queries that don’t always require retrained expert or by following up with inquiries to ensure that appropriate considerations are made. And all this is just like the traditional or light or legacy AI that’s been stood up for the past few years. Now with generative, we’re starting to see real transformation and entire ways of working medical information, how we determine what needs exist, what queries will be put forward, how we craft responses, how we design develop and deliver dynamic content. That’s, that’s a lot of these. And also the type of media that content can take, for example, does it like you know, have to be written text every time or can it be a visual abstract can be an infographic could be a chart, maybe it’ll be a personalized video at some point in the future. But our conversation didn’t just end there. Because when you start remodeling the business with AI, it’s kind of like pulling on a long red thread. The challenges that live in medical information, like what the customer experience with improving outcomes, they often apply to other aspects of Medical Affairs and sometimes might be transferable
Steve Casey 07:17
Matt, before we get too far into the AI discussion on medical communications. We’ve been exposed to discussions of AI but you know, as you mentioned, we don’t have a real solid understanding of it just yet. A lot of us are reverting to like general AI as being Chad GPT, or things like that. But really, could you give us kind of an idea of what you think about AI and how we should think of AI? Where general AI fits into the broader AI discussion? And how you’re differentiating between AI is artificial intelligence or AI as augmented intelligence?
Matt Lewis 07:55
Yeah, sure. Sure, Steve, of course. So I mean, your audience might be really interested to know that the term artificial intelligence was created, it was is a kind of manufacture term, if you will, fashion architected, if you will, in a workshop, not unlike the ones that our teams broadly facilitate with their partners in support of things like scientific platforms, a very, very similar discussion. Before artificial intelligence existed as a term backfield was largely referred to as intelligence amplification as in amplifying or strengthening the innate intelligence of humans. But when researchers described it that way, the people in the field could never seem to attract enough private or public funding to the level of desire. So they held this little workshop at Stanford in the early 1950s. And it was decided that they would rebrand the field as artificial intelligence, indicating that it was computer derived methods of improving intelligence. And it worked. I think, only too well, because the AI term, this idea of artificial intelligence, computer drives intelligence stuck, but it also kind of like, manufactured this whole Hollywood industry, of robots and AI like taking over the world and stealing people’s jobs and the like, which, you know, is an unfortunate byproduct of the manufacturing of this idea of AI. Now, I think when people reference AI, they’re either describing this large class of technologies that use computers to provide alternative mechanisms of reasoning beyond what a human can provide, like language, vision, and other modalities, but within an NGO, and broadly when we refer to augmented intelligence, we’re specifically describing a human centric design pattern. This comes from Gartner, where the emphasis is on leveraging computer innovations to improve human cognition, decision making, efficiency, reasoning and effectiveness. Humans are both the Alpha and the Omega in augmented intelligence. It’s about humans leveraging AI being better than either humans or AI alone. That’s really what augmented intelligence stands for generative catch up Tina like are so meaningful because they really make this problem On this of augmentation, a possibility and a reality today, it’s really never been something that can actually be achieved in the real world until generative has first come into the world as it has now over the last year, year and a half.
Steve Casey 10:12
Man, that’s such a great answer. Thank you so much for giving us that background and information. You and I have had some conversation about how generative AI can change medical communications. And in the introduction, we mentioned that generative AI is opening up a whole new transformational opportunity in the development and publication of communications. Can you talk more about this and your thoughts of where this evolution is going to lead?
Matt Lewis 10:38
Yeah, I used to say, Steve, that I don’t have a crystal ball. But you know, I kind of tried to look around the corner wherever possible. But, you know, now now I just referenced the William Gibson quote that, you know, the future is here, it’s just not evenly distributed. And I think that is really true, especially nowadays, you know, there are some teams out there that are probably listening to this, that they’re already implementing and standing up generative models within their environment, others might be using and deploying GPT, or other applications, just kind of see what works experimenting, if you like, and others are still trying to figure out why they even have a kind of policy in place that prohibits the same. So it’s not that you know, the future is coming. It is already here, just it just not evenly distributed. You know, I was just added a generative AI conference in Boston, it was a cross vertical meeting, including discussions with folks in finance, and banking, and hospitals, and defense and telecom and retail, life sciences, everything. And I think what we’re witnessing here in Medical Affairs is also happening in every sector of the economy. And you know, it’s not just at work, if you will, and I’ve been in the space for 14 years, specifically in AI and advanced analytics. And there, there used to be a time when you could go and speak with a team, whether you were in house at a sponsor company, or on the consulting or agency side, and then go home and do your thing at home and kind of hang up your shingle for the day and be done. But now it’s not just at work, like when you go home, our sports are being analyzed and narrated by artificial intelligence, or movies are being recommended to us by Netflix, which is powered by machine learning. And a lot of music is either being generated or copied and deep fake by AI, which is also synthetic media, AI, generative AI, and we’re seeing reactions in the film, media and in politics. So what’s going to happen? I don’t know, I’m not certain, but I think many of us will look back and this time, the end of 22, the beginning of 23, as an inflection point, the end of the beginning perhaps, I’ve been working in AI, I’ve been working in it for 14 years. And for most of that time, AI has been like a point solution, a tactic. Over the next few years, I think it’s going to become more like an everything at everything we do, will kind of run off a master artificial intelligence driven operating system, behind the scenes, all our platforms, our services, our apps, our content will be powered by artificial intelligence, such to the point that it won’t even be mentioned, kind of like how we have electricity now, like we don’t really say like, oh, you know, we’re using electricity to use my computer or power my light or turn on my iPhone. It’s just there in the background. It’s like everything we do in med comes in all Medical Affairs, life sciences, will actually be powered by AI. Is it going to happen by Christmas? 23? No, I don’t think so. But it’s coming.
Jennifer Riggins 13:19
Yeah. So I, you know, I find all this pretty cool. And really interesting to think about and think about the impact that it’s going to continue to have over the course of time. So thinking in that kind of framework, can you give us some insight and how you think some of this AI may impact Medical Affairs and medical communications in particular?
Matt Lewis 13:42
Yeah, sure. Well, I mean, it’s already here, like, as I mentioned before, but but more is getting possible every day. I mean, I think the AI that you read about in today’s journals, reflects tech, that was current 10 to 12 months ago, as we know, because there’s such a long lag time between science and discovery and publication. So what is available now is almost an order of magnitude better. And the models are better, and are improving quite quickly. So the technologies that’s coming out from open AI, from Google and from Facebook and from all the other companies are so much better than they were last year that it’s almost unbelievable, and they’re getting better by the day. But in Medical Affairs, a relevant use cases somebody may be thinking about, like, understanding the landscape and a given therapeutic space. Like there used to be a time like back when I was an MSL 22 ish years ago, when the standard of care in an area didn’t change that frequently. And that highly relevant papers didn’t appear all that often which was quite helpful actually. But in many disease states, those days are long gone, like the I think the pandemic prove that in itself were upwards of 10,000 articles preprints and other scientific assets were emerging literally every week during the height of the crisis. And keeping up with that type of avalanche of data. Just it’s just not possible. No human can do that. Even if they spent Our whole career from this point forward to the time they retired, just looking at that 10,000 set of, of articles and that just in that corpus it just not possible, and it changes every week, so just you can’t keep up. But we actually did that we partnered with Maderna at the beginning of the pandemic, to stand up an AI powered literature monitoring solution, which augments the abilities of their internal staff to scour the landscape, surface relevant signals and suggest relevant actions for members of their executive team and their Medical Affairs group. And we support a number of similar platforms right now with groups like AstraZeneca, and others, recently won some awards and medical excellence. But I think these are just like initial examples of widening the aperture helping groups see what’s possible out in the world and thinking we don’t have to make sacrifices, we don’t have to make compromises. We don’t just say like, if we have a search frame of 100 keywords or MeSH terms in one include that we have to just limit it to, you know, the last two years or just these key terms are just you know, the ones that are full text versus the others because we don’t have the time or the bandwidth to look at everything. Now we can complement that with an artificial intelligence approach. And find all the content that’s relevant auto summarize it using AI highlight what’s relevant, stand up the metadata, make it accessible, and speed time to decision for clinicians and patients.
Steve Casey 16:14
Wow, for several years now, I’ve been saying that medical communications is undergoing a digital transformation. When I say this, I mean that Developing and disseminating communications in the old print manner, doesn’t really deliver the impact and medical communicators need to adopt a digital first mindset. This is really where we’re developing our communications to be accessed, consumed and shared digitally. I noticed that you recently posted a link to an article from the Harvard Business Review about where companies should start with generative AI. When I read that article, I found the author was talking pretty much about my mantra of digital first, and moving it into the generative AI world talks about a subset of that knowledge work called wins the and that’s the most susceptible, according to him in being replaced by generative AI. For the audience. What winds means is it’s work that is dependent on the manipulation and interpretation of words, images, numbers and sounds met, both of our organizations fall into this area, our focus it, my group on the HC currently surrounds using generative AI tools to augment the existing communication process, which we hope should lead to improved cost and efficiency. You’re undoubtedly a leader in the use of generative AI and medical communications. How are you using it today, both personally and through in Israel?
Matt Lewis 17:39
Yeah, it’s interesting. You know, that conference I mentioned in Boston that I’ve just added earlier this week is actually the the person who organized the conference is the is the lead author of that wins paper that you just mentioned in HBr. pallbearers who organized that that meeting that was just at, so there was a little workshop at the conference, I was just that where they took folks through the framework is really interesting, as in that article, which maybe you guys have a link to this, this the the session, the there are different ways of kind of thinking about where firms are. And I think if you’re in professional services, like we do consulting, for example, or on the agency side, you’re in the top right of the framework, which indicates that a lot of our work given you know that it’s language driven, it’s highly technical as expertise that’s required. It can be augmented with general AI. And then if you’re working, say, in life sciences, or biotech or medical device or digital therapeutics, you’re in the top left, which is also standing up generative or artificial intelligence to improve target discovery or other aspects of the business. So being in the top is really an area that’s ripe for augmentation and consideration. I think that’s it’s a really helpful way of kind of seeing the world for those that are looking to kind of get started talking about getting started. I mean, I think I’ve always been really interested and passionate about the space and how I’ve gotten really deep in over the last year year plus, is really just a mix of kind of curiosity and experimentation. And also, we’re seeing a lot of kind of practical implementation with an across to client teams, if you will. But you know, what am I using today? What is an SEO kind of doing? So let me maybe start with an SEO Addonizio work we’ve been implementing AI powered solutions for over eight years, we have more than 100 projects successfully completed across biopharma biotech and medical device that leverage the kind of traditional or light AI implementations and things like machine learning, deep learning, natural language processing for use cases across the whole Medical Affairs stack things like you know, scientific platforms, lexicon, publication, planning and writing patient lay summary, field, medical, Omni channel and beyond. And our team really does because of this have like really the broadest and deepest bench in industry and we have now over 200 active experiments internally across our industrial medical division to explore how can we work smarter to ensure that we’re really kind of staying at the leading edge on behalf of our clients as the ground really like moves beneath our V which is doing quite quickly. You know, personally, by my account, I’ve reviewed and tested over 500 unique applications that are generally AI powered in the past six months, I tried to test at least 10 new applications every single day, including weekends, which does get a little tiring. But it’s also quite interesting. And as that experience grows, the way that I approach many of the tasks in my own work in my life has shifted as well. I mean, like, one way to bring some color to this is by highlighting a little bit of a like practical example. And I travel a fair amount, right for work and for fun, but I don’t always know a lot about the places I go to before I get there. And I really haven’t worked with like a travel agent in quite some time, and never really found Google to be a great source of local information that’s both practical and helpful. So since generative came out, I tried Jaggi beauty and tried to ask pi the other those deep pines model. But since then, I’ve been really consistently relying on Bard, Google’s experiments a product for producing detailed itineraries for work trips, and for personal trips, both here in the States, as well as in Europe, including England and Ireland. And it’s surpassed my expectations really mean it’s really great. It has like really text detailed images and photos of areas for itineraries and might be worth considering. And I’ve suggested this as an option for folks that really haven’t had the best success with GPT. And I’ve heard from people in Medical Affairs, that this is a really nice complement to plan, you know, reunions or trips people are taking to hike, you know, the see some ruins in the UK that the Roman ruins, for example, where they wouldn’t have gotten information that otherwise is accessible, but just you know, can’t can’t be pulled down. So that’s a kind of practical way of thinking about it. Professionally, I think my, my approach is really, to developing conversational narratives, especially around panel presentations, or Keynote presentations. Like, for example, I’m doing a keynote at the MAPS digital innovation meeting in Chicago and a fortnight has completely changed as a result of generative, I’ve really found AI to be kind of like a welcome, partner, like a thought partner in ideating, in testing, and validating and really expanding initial concepts. And I now I’m running multiple pilots on the creative and digital learning transfer aspect of live experience interactions, to both strengthen the narrative, but also think about how it can be more efficient with my time so that it’s effective for my audience, but also consistent with other things that I have going on in my professional life. So I mean, that’s an n of one. But in my experience, I think I’m a better speaker, a better coach, and a better consultant because of AI. And I think it’s really invaluable.
Jennifer Riggins 22:37
So I love those examples, Matt, I think it’s really pertinent to how we’re using AI right now. And you know, how we might be using it in in the future. And, you know, I just love how you’ve been able to incorporate it into what you’re doing on a day to day basis. It’s just a really good practical examples. But you know, I think that there are still a lot of challenges that we need to overcome. And one of those is really around data security, and possibly the proprietary nature of it. And and really, until we can answer those issues clearly and overcome the open source fears that chat GPT has spawned within pharma, I’m not sure we can truly harness large language models to improve the efficiency of communication development. So in your opinion, Matt, are there things we’re doing from all the different societies that you’re working with or within your organization that can help mitigate farmer’s fears? You know, regarding security and confidentiality?
Matt Lewis 23:38
Yeah, I Well, first of all, a lot of these fears stem from direct experience that people have with the consumer version of these apps, or people, you know, using apps on personal devices, in organizations that have a policy that banned their use professionally, you know, having that prohibition or the exclusion of the use of gender in the in the enterprise, really kind of stifles innovation. And we do not condone such exclusions with you don’t have one, say within an ACO and the society’s broadly mentioned before, don’t broadly prohibit them, either or the we’ll talk about some of that in the policy statements that are forthcoming. And when the consumer applications though, first came out, there was a lot of risk, you know, a lot of concern, but many of the more broader tools that exist now like for example, you know, the models that underpin the chat, GBT, for example, now have enterprise licenses that are directly available to organizations like Lilly, Pfizer, and larger consultancies, if you will. Any company that wants to transact directly can do so through an enterprise license. Or they can work with a consultancy like ours to stand up a bespoke model directly within their cloud environment, or access it through an API that limits and mitigates the vast majority of the risk. So, a lot of that initial concern that existed with the consumer apps, the consumer technologies, if you will, is not a concern for enterprises likewise are the hallucinations that folks experience initially on with chatty beauty So for the consumer population or not real risks, if you will, in the life sciences enterprise, when working with an experienced team that has gone through security protocols, using specific datasets, etc, etc, within a data science environment. But definitely, it’s been something that folks have seen in a kind of superficial consumer environment, if you will. Addonizio we have a security governance and compliance team that ensures that no applications come into the hands of our teams, or our clients that are not safe, ethical, secure and appropriate, with respect to data privacy and other expectations. And this is an evolving space. I mean, it’s it’s something that we’ve paid a lot of attention to over the last eight years, but even as generative becomes more and more visible to teams into groups at large, and we’re looking at ways of kind of raising up the considerations around things like ethics and governance and data privacy and the different jurisdictions in which we play as well as things like data provenance, and you know, intellectual property or the right so it’s certainly not a fully satisfied topic. But it’s a lot safer now than it was, you know, even 10 months ago.
Steve Casey 26:04
Yeah, and I don’t know if you know, this, but Matt and I met when we were serving on a social media metrics committee. Matt’s expertise in data analytics was extremely helpful to the group in really trying to better understand metrics and how they can be used. Man, if you don’t mind, can we put our metrics hat back on and maybe you can give us some insight into how you see AI, assisting in what I call the elusive perfect metric, which shows how a specific article is generated or created, air quotes, impact or air quotes against sustainable value?
Matt Lewis 26:40
Yeah, Steve. Sure. First of all, I know that you’ve ever taken your metrics hat off, and I don’t think I’ve ever taken mine off either. But you know, thinking about things from that perspective, which is really a bad like, beginning with the end in mind, you know, most of the metrics that are currently in use across you know, pubs, but, you know, across the broader Medical Affairs environment, as well, were originally designed to measure outputs of a system that is, you know, the kind of Medical Affairs system, if you will, that is essentially free AI. And as we begin to enter the post AI era, the AI native era, if you will, I think your audience will begin to see a bunch of new key performance indicators that reflect the value drivers that medical is becoming responsible for some of which we’re going to say referenced the new McKinsey 2030 paper that was issued earlier this week, and which are now achievable through these types of innovations. Many of these things reflect things such as efficiency, effectiveness, engagement, as well as the time to decision. That is how long it takes to submit an article to a journal for example, or produce a patiently summary, that time matters, not just in terms of costs incurred for all involved, but also the opportunity cost to researchers in the field, what they could have been doing with that time, if they weren’t toiling away at the third submission, the third resubmission, etcetera, etcetera, or ultimately the health costs to patients suffering from the condition being discussed, if we can speed time to publication, and ultimately understanding hopefully, outcomes will improve for patients and the population at large.
Jennifer Riggins 28:11
So Matt, in our last series, we asked our interviewees about the impact of omni channel, and specifically, we delved into how omni channel impacted changes in clinical behavior. And although we geared it to be HCP centric, none of the interviewees responses really could pin point, or really point to evidence that omni channel did in fact, impact clinical behavior. So do you think that in the future, we can incorporate AI into the assessment of omni channel to show the impact? Or do you think that we’ll see some variation of omni channel that may be hyper geared towards being HCP centric due to AI?
Matt Lewis 28:48
Yeah, that’s, that’s a tough one. I mean, I think those responses really reflect the relatively early maturity curve of omni channel in medical. When you look at omni channel outside of Life Sciences, say in the big banks, or in FinTech, it is definitive that when executed well, Omni channel definitely shifts mindsets and behavior. And as a regulated industry, finance broadly is somewhat similar as a sector though not the same as our industry. I do think, though, that AI will more like catalyze the actual implementation of medical omni channel so that teams will both improve their confidence and being able to do more, have broader scopes, execute Bigger, Longer campaigns, and see more durable effects. And in doing so, the type of responses you get will shift, if not, because of the assessment itself, because of the fact that the interventions themselves are just so much more robust by like 10x sub and maybe even 100x, you’ll see just really profound effects where people are starting to kind of tease out what might be possible in a short amount of time. 612 months, you’re going to really see profound impact.
Steve Casey 29:59
Matt As you might know, I’m one of those people that has to know how something works so I can figure out how to use it. Generative AI is still based on algorithms, like the old search engines had secret and had their secret algorithms. Do you think that as generative AI continues to improve, we’ll see more Explainable AI where the user can investigate the AI model, its expected impact and potential biases to help characterize the really the accuracy, fairness, transparency and outcomes of what that AI decision making processes doing.
Matt Lewis 30:35
Yeah, yeah, I mean, we’re already starting to see a lot of this in the space, both in terms of the lack of kind of explain ability initially, as well as kind of like, what, what kind of underpins hallucinations? And like, why those happen and what they are? And, you know, are they features are they flaws, like why, why they exist and how we can learn from them. Like, for example, I know, there’s a lot of work going on right now in the higher education space, like in universities in the US, for example, where when a hallucination occurs from a large language model, which is essentially a prediction that ends up not being true that the teacher in the space professor can help students recognize why that may have been a suggestion and then help them think critically about that alternative hypothesis and get people to think about a world in which that hypothesis could have come through and think about all the interconnections that might be possible. So they’re using kind of the hallucination as a as a teaching tool, as opposed to something that went wrong or that, you know, happens to be factually incorrect. Of course, that’s not something that can be applied in all settings may not be appropriate, say in life sciences. But it’s an interesting kind of consideration as to ways of making the model work for us. And when we don’t really know what’s going on. Within our environment, we work directly with Microsoft and Bill Binney AI team, and we talked to them at length about, you know, why, you know, certain hallucinations are the way they are and why they they don’t seem to make sense to us. And, you know, they don’t know any more than we do in terms of like, why they produce certain outputs, and why they can’t explain why one output is the response versus a different output. But there’s still interrogating those underlying models and trying to get them to kind of show their work, so to speak to, to use an old kind of, you know, kind of grade school kind of analogy. And you can get through a lot of this now, in when you in some of the more advanced models will the explicate or show in Python, for example, or other code sets, what’s actually happening when they produce an outcome, if you will. So you can see a little bit more of why things are happening the way they are. But it doesn’t necessarily offer more of an explanation as to how that came to be or why they went down that path. So it doesn’t necessarily give you a lot of comfort necessarily that it’s doing the right thing is still requires some prompting or some training or some fine tuning to get to a place where you feel a little bit more comfortable about both the desired response as well as the one you actually get, you know, in the conversations we’ve had with medical teams with leadership at medical organizations and folks that are working alongside them. We haven’t really heard of explainability, or lack thereof as something that teams offer as a reason why they don’t want to engage with generative, they probably do feel it, you know, truthfully, but they don’t really express it, they I think we definitely shouldn’t be working towards making that kind of black box, if you will, of explainability a bit more transparent. But there are so many human factors with regards to AI, that if this is like the only one that’s addressed at all the other ones ignore it, I have like a laundry list of human factors, if you will, that are relevant for the space, I’m actually doing a talk on this next week, we’ll have failed our colleagues who say if you don’t get the human factors, right, and you just focus on the technical or technological factors than AI doesn’t work, because ultimately AI is about amplifying or augmenting human capacity that in the human pieces is imperative.
Jennifer Riggins 33:47
That’s interesting. So Matt, we all know that healthcare providers cannot keep up with the amount of information that’s coming to them daily. You know, it seems to me that with generative AI, medical communicators could produce a lot more publications and communications, you know, perhaps even creating a bigger firehose of information. So do you think that there are ways that publishers and readers can use generative AI to manage this new tidal wave of information? Or do you think there are ways that pharma might better police itself to ensure that they’re retaining the integrity of the medical communications?
Matt Lewis 34:23
Yeah, this is this is definitely a hot topic right now, Jen, I think, you know, it’s it. I have not seen a real clear approach to this within our space, in the broader environment, like in the broader economy, if you will, the big focus is on what’s called provenance, which is trying to essentially raise up the path of content creation for folks that are consuming media out in the world at large. So you know, think people that are looking at like an ad on TV or like looking at a political statement from a candidate and trying to understand what parts of that ad worked derived by human contribution versus what parts were derived by artificial intelligence right now, when you see such a thing, there may not be a watermark, they may not be an element that you can scan with a camera and look at a QR code and see, you know, this is 40%, human 60% Ai. But that provenance consideration is looking to make that transparent, so that at least the the viewer of the content can quickly understand and then bill to make a determination for transparency and trust. Is this something that I want to interact with and how I make a decision around it. And equivalent consideration around provenance has not yet kind of really made its way over into our world. But I have to imagine that it’s forthcoming. You know, I’ve seen some of the publishers like Springer, for example, and others have been taking a very strict approach to this, where they’re prohibiting the use of generative images and texts to some degree in any of their submissions, so as to kind of like tamp down the total amount of content they receive ICMJE e waim. And other groups have weighed in saying that like, you know, AI can’t serve as an author and papers and the like, I think that’s a little odd. Honestly, it’s almost like saying like, Oh, would you use Google as an author? Have you ever quote Microsoft Word as an author? It can’t be an author only humans are authors. But I don’t know, the bigger question kind of remains. General vi does lower the costs of creating content. So as a result, the increase in overall content is to be expected. I think it’s still early days, and it to consider how this will be moderated screen how the Providence has stood up? How do we consider it but it is definitely an area of active discussion. I can’t say I have an answer.
Steve Casey 36:32
I think you may have an answer to this last question that I have. But it’s just one last question. Where do you see generative AI headed? What do you think in over the next five years? A generative AI looks like and Medical Affairs to you?
Matt Lewis 36:50
Yeah, Steve, I like to say and my team is probably sick of me saying this now because I’ve said it like a million times. But I like to say that a week in generative is like a quarter in the rest of the world. Because things move so fast. And making a quarter is like three years. So as such, I’ll give you a sense of where I think we’re going to be when we chat live in Puerto Rico with MAPS 24, which is like six years in general of AI time. So you asked me about five years. But you know, Bloomberg together in Puerto Rico in MAPS at the 2024 meeting, that’s like six years in general AI time, the world will have changed, turned on its axis by then, by then the majority of big pharma companies that is like about six months from now, most of the major pharma companies will have an artificial intelligence policy. And we’ll be working on the shape of or have already set up an artificial intelligence strategy, if they haven’t already stood up. Now. The specialty pharma biotech and medical device companies, especially the more innovative and progressive ones, amongst them, will also have done the same. I also predict that within six months, I will not be as lonely as I currently am. And that a number of other chief AI officer peers will join me in this role across the industry as most of the other industries in the broader society and an economy have chief AI officers already. And the full kind of continuum of Medical Affairs will really be starting to think about what programs from an artificial intelligence perspective need to be stood up, which ones from an experimental perspective that are currently active need to be killed, which ones get tweaked and kind of grown, what gets scaled. As we head into the summer of 2014. I think you know, the space at large to Jen’s point is going to be facing an abundance of content, both medical content and generative content. And many of us, including me are going to be working out how to ensure that it remains relevant, that it’s value added that is meaningful for our partners and ultimately for the patients we serve.
Jennifer Riggins 38:44
So I really look forward to MAPS 2024, to see if your predictions come true. Matt, Stephen, I want to thank you and Inizio for giving your thoughts on generative AI what it is, where it stands today and where we can expect it to go in the future. To our listeners, we hope you’ve enjoyed our discussion on generative AI. If you enjoyed the podcast, please make sure to like it and feel free to comment to us, you can find our contact information on LinkedIn. Thank you for joining us today and listening to our podcast series digital first scientific communications a podcast production of the digital Focus Area Working Group at the Medical Affairs Professional Society.
Steve Casey 39:25
And remember, if you’re a MAPS member, thank you for your support. If you’re not yet a MAPS member, I want to encourage you to join us so you can access additional resources. Visit the MAPS website today at MedicalAffairs.org/membership.