Zohaib Sheikh: Hi, I'm Zohaib Sheikh, head of content at Fierce Life Sciences, and today I'm excited to speak with Srini Shankar, Chief Commercial Officer, Americas and Global Head, Life Sciences at Cognizant. Welcome, Srini.
Srini Shankar: Thank you for having me. It's really exciting to be here to talk about this topic.
Zohaib Sheikh: Great. So before we begin, could you tell me a bit about yourself and your role at Cognizant?
Srini Shankar: I wear a dual hat. I'm the Global Head of our Life Sciences business, which for us is all of the work that we do for our biopharma clients and our medical devices clients. I'm also the Chief Commercial Officer for the Americas on an interim basis, wherein I run the entire sales teams for our new logo business development as well as our technology sales organization for the Americas.
Zohaib Sheikh: AI has been around for a while with several applications that are already known for image recognition, autonomous vehicles, social media content, et cetera. What is unique about generative AI and why is there so much buzz around it?
Srini Shankar: Generative AI has really taken off with the advent of ChatGPT, I think, which all of us know. So there's nothing new about that. I'm going to basically characterize AI before generative AI as quote unquote "conventional AI." So let's talk about conventional AI versus generative AI. If you think about conventional AI, which has been around for a while, but it has become more powerful with the advent of computing infrastructure and the evolution of computing infrastructure.
Conventional AI is using computer algorithms that are based on concepts like deep learning and neural networks to be able to mimic human intelligence. So that's a very basic definition of say, conventional AI. And the manifestation of that is essentially the ability to do things like image recognition. Looking at a whole bunch of images, let's say, radiology images and predicting potentially what may be a cancerous image, if you will. So areas like image recognition, predictive analytics, automation of human tasks, I think a lot of that is really conventional AI.
When you look at generative AI, it's a form of AI. So not all AI is generative AI, but generative AI is a subset of AI. And generative AI, essentially, is using certain forms of neural networks that is characterized by large language modeling, to be able to actually generate content, which could be either text, images or music in a way and a level of originality with which humans can actually generate content. And that's what really makes it a game changer.
What it essentially does is that it looks at language. For example, when it comes to generating text, it looks at language, grammar, semantics, syntax, and it then is able to generate text like a human would. Looking at things like word associations in a sentence, sentence associations in a paragraph, and so on and so forth. So that's really the difference between conventional AI, which has been around for a while, it's got more powerful in the recent past, and generative AI.
Zohaib Sheikh: Great. What could be the most compelling use cases for generative AI in life sciences?
Srini Shankar: That's a great question, and the way I would start by saying is, where do we begin? Just given that, I think, just in the last few months, in speaking to a lot of our clients and reading some of the literature around this, it's almost like we're just at the tip of the iceberg. I truly believe that generative AI is at a point where it's almost a level of general-purpose technology that's going to be all pervasive, both in the consumer context as well as the enterprise context. So, it's just something where we're at the tip of the iceberg. Probably what I'm going to talk about is just going to be a subset of what this will evolve to, and we're still very much in the learning stages, but I can tell you the excitement with our clients, the industry, with our partner ecosystem, it is just incredible.
So, I'm going to give you a few examples to just bring it home in life sciences. So, there are four or five broad categories, and I'll try and drain some of them out with examples. I mean, the first one is really semantic search. If you think about our industry, a large part of what people do in our industry, is they access medical and scientific literature that in many cases is publicly available. So most of researchers and the scientific community and our clients, they actually go out to places like PubMed or clinicaltrials.gov or many other sites, both paid and public, and they go through, and in some sense you can say scrounge through, medical literature to be able to do what they have to do.
So if you take an example of PubMed, which is a publicly available site, which is probably one of the largest sources of medical and scientific literature, it has about 35 million citations, which is kind of organized in the same way like any other repository, which is it's based on what is called medical subject headings. So you can search on say, neoplasms, which is another form of say, cancer, and then next level breast neoplasms, which is essentially breast cancer and so on and so forth. And then you do keyword searches just like what you would do traditionally using any advanced search engine. And then it would give you a set of results. It may give you page rankings, about a hundred articles, and then you would go through a hundred articles and then read a hundred articles and figure out what's relevant. And then, that's how you progress with your medical research.
In the advent of generative AI, you can just fire off a query. Let's say you're a researcher in the cardiovascular space and you're researching about arrhythmia. And you want to, just say, compare with arrhythmia, which is essentially, irregular heartbeat. You want to really compare the use and the efficacy of anti-arrhythmia drugs, which are used to treat arrhythmia vis-a-vis more invasive techniques like for example, ablation. You could just say to a generative AI engine, go out to PubMed, look at all of the literature there and come back and compile to me a comparative study of the efficacy of both of these techniques. Can you believe the amount of time that would actually save when it actually does that for you, as opposed to you going and reading thousand or a hundred articles and then coming up with your own paper? That is how powerful it is when it comes to semantic search. So, that's one use case that we've talked about with a lot of our clients.
The other one is really content generation. Look, I mean, there's just so many places where content generation is needed in life sciences. The way drug discovery is done, you actually have early discovery and then you start and then you progress through a clinical trial. And one of the first things you do in a clinical trial is what's called “protocol authoring.” You author a protocol around, how do you design a clinical trial on a clinical study? What are the objectives? What's the design? What's the dosage? What's the patient population? What's the indication? All of that stuff. And all that is authored by somebody or a group of individuals.
I could foresee a future where you could have generative AI, essentially, author these protocols and then somebody validates that and chooses, okay, here is the best protocol from a set of options. I actually did this myself just for the kicks using ChatGPT. So I said, you know what? Non-small cell lung cancer is a form of lung cancer. So, I just asked ChatGPT, "For non-small cell lung cancer, tell me what the biomarkers are, what the clinical endpoints are," and so many other elements that characterize the protocol. And then it just gave me all the options. It says, "The biomarker is EGFR mutation." I have no clue what it stands for. It says, “The clinical endpoints are progression-free survival,” which means the number of years in which a subject actually survives without the progression of the tumor, because that's what you want to assess when you actually do a clinical trial for that particular indication. So, it's just fascinating that it just gave me the entire elements that perhaps comprise a significant portion of a protocol, which I can either use to write it or maybe I can have it write it itself. So content generation, content authoring is huge.
The other area is medical writing. I mean in our client community we have so many medical writers who actually create content. I mean, when you watch TV, you see ads for drugs. A lot of that content is actually written by medical writers and actually in some cases contextualized for geography. So, when you see an ad on CNN here in the US for psoriatic arthritis and somebody in the creative agency that's working on this wants to adapt that for, let's say, the Asia-Pacific market, you can actually tell generative AI to actually pick an image and completely rewrite that ad for that market. It'll go and actually pick an image that's perhaps more apt for that market. Because it's not just about generating text, but also generating images, which we've seen with the advent of things like DALL·E and others.
So, there's just so many different use cases. I mean things like customer experience. When I want product information about a drug, I can actually ask a series of questions to a generative AI based chatbot that can actually provide better curated answers than perhaps a human being. So, the opportunities are just tremendous. And we're probably just going to scratch the surface in this conversation.
Zohaib Sheikh: Right. Do you see the potential of generative AI to meaningfully accelerate the development of new therapies? If so, how?
Srini Shankar: Look, I gave you a few examples in areas such as semantic search or protocol authoring. Think about this. Sometimes for an indication, it takes many months to sometimes more than a year to actually author a protocol. We can't afford that, especially in situations like we had with the pandemic where we have to really accelerate therapies. And we're talking about patients who are struggling with survival in areas like oncology. So, the ability to collapse the cycle time to bring a therapy to market, which today is as high as 10 to 12 years on an average, by looking at different aspects of that value chain, whether it's designing a study, whether it's summarizing a study for submission to regulatory authorities, there's so many different points in that progression of discovery and clinical trials and then regulatory approvals where you can actually leverage generative AI to actually accelerate things.
And hopefully, I just give you a few examples earlier around areas like protocol authoring or medical literature research, or even in summarizing the output of clinical trials in what's called a “clinical study report.” I mean, once you conduct a clinical trial, you take all that data and you got to summarize that in a clinical study report where you demonstrate statistical significance both from an efficacy and a safety standpoint. And then you tell the regulatory authorities, "You know what? Take a look at this. I mean, my drug hopefully satisfies the requirements of efficacy and safety." That compilation itself takes an enormous amount of time, and just the potential of generative AI to actually accelerate the compilation for submission is just huge. So, at every stage of bringing a therapy to market, whether that's early stage in terms of looking at research, designing a clinical trial, taking data and making it submission ready for regulatory approvals, at every stage there's opportunity.
Zohaib Sheikh: Right. So, while the development of generative AI technologies will likely create new job opportunities, where do you see jobs being vulnerable to generative AI?
Srini Shankar: Look, there's no question that there's certain types of jobs that will be vulnerable to AI. So for example, we talked about medical writing. I mean, I would not characterize vulnerability as displacement. I would characterize it as an opportunity for automation and augmentation. So the way to think about this is let's not think about this as displacing jobs. Let's think about this as making humans smarter through augmentation and automation. Therefore, you can really see areas like medical writing, protocol authoring, medical literature review, even the people who actually interact with the scientific community, once a drug is approved, people who meet actually with physicians and educate physicians around the benefits of a particular drug or a particular therapy, all of that can be a lot more intelligent.
So, it's more augmentative as opposed to being seen as a displacement, if you will. And that's how we should look at it. I mean, that's how we should adapt it. I mean, it's not the right way to look at it from the standpoint of seeing it as a displacement. It's just going to make us all smarter.
Zohaib Sheikh: Of course.
Srini Shankar: And the net result is, if lifesaving therapies are going to get to patients faster, then anything is worth it.
Zohaib Sheikh: Right. All right. Well, Srini, we appreciate you sharing your thoughts and insights with us today. Thank you so much.
Srini Shankar: Thank you very much. I'm very excited about the opportunity that this technology beholds, and I'm sure we'll learn more as we progress.
The editorial staff had no role in this post's creation.