“There’s a lot of resentment towards the hype,” notes Christopher M. McSpiritt, head of life sciences at Domino Data Lab. “There’s a healthy skepticism around it. They’re wondering if this is just the next shiny object we’ve been chasing.”
It’s like the pharma industry got a shiny new toy, but half the kids on the playground are scared to touch it, while the other half can’t stop playing with it.
How pharma firms approach AI could shape their destiny
Some pharma companies have backed away from off-the-shelf GenAI tools, citing security concerns. This split reaction underscores the deep tension between innovation and caution in an industry where data breaches can cost several million dollars. GenAI is like that tireless, always-available intern, willing to work for a pittance and capable of superhuman feats of data analysis. “It never tires and can do as many tasks as you want, but it still requires supervision and guidance because it’s not an expert,” McSpiritt says. In pharma, GenAI is like that intern who never sleeps, never complains, and can skim a 500-page document in under a minute. In pharma, the tech can also flag adverse events and even help write a first draft of reports that might take days to complete otherwise.
But, for the most part, the early off-the-shelf models were more effective at straightforward tasks than complex ones. Without careful guidance and the right tools—like Retrieval Augmented Generation (RAG)—its superpowers can quickly devolve into a chaotic whirlwind. While reasoning models are emerging that are better at STEM-related tasks, the technology remains immature.
An always available intern willing to work for a pittance
The pharma industry is walking a tightrope between embracing GenAI’s potential and avoiding a face-plant into regulatory non-compliance. It’s all about finding that sweet spot where AI’s tireless capabilities meet human expertise. In the pharma sector, GenAI is emerging as a tireless digital assistant, akin to an “always available intern,” as McSpiritt put it. The tech can summarize documents hundreds of pages long in seconds, doesn’t clock out, take breaks, or suffer from fatigue. For instance, in drug discovery, GenAI can rapidly scan through millions of scientific papers, identify potential molecular targets or drug interactions that might take human researchers weeks or months to uncover.
A growing number of gen AI use cases
Current applications of GenAI in pharma are diverse and growing. Beyond literature reviews and report drafting, these systems are being employed for tasks such as predicting protein structures, optimizing chemical synthesis routes, and even assisting in regulatory compliance by flagging potential issues in documentation. A notable example comes from a mid-sized biotech firm that used GenAI to analyze years of historical trial data, identifying patterns in patient responses that led to a more targeted approach in their Phase II oncology trial design.
Despite these advancements, limitations persist. “It still requires supervision and guidance because it’s not an expert,” McSpiritt said. This limitation is particularly evident in complex decision-making scenarios. For instance, while GenAI can suggest potential biomarkers based on data analysis, it lacks the nuanced understanding of biological systems that experienced researchers possess. Similarly, in drug safety assessments, while AI can flag potential issues, human experts are essential for interpreting these flags within the broader context of patient health and regulatory requirements.
The successful integration of GenAI hinges on several factors, including data quality and cross-functional collaboration. “I think that’s one of the big things where GenAI or AI projects in general fail: lack of alignment on the data and ensuring that the data is both complete and fit for purpose,” McSpiritt points out. He also stresses the need for process-aligned thinking over functional silos.
Approaching data as a product — rather than a byproduct
McSpiritt emphasizes the need for a more holistic approach to data and processes to make the most of both genAI and AI at large: “I think it all boils down to needing more process-aligned thinking versus functional lines thinking.” He explains, “To me, that’s a big thing because if I’m running a clinical trial, I have data management, clinical operations, pharmacovigilance, and so on. They all have their own dashboards and their own transactional tools that generate information. But ultimately, as an organization, I want to have a holistic view of that data.”
Looking ahead, McSpiritt sees AI as a catalyst for reimagining roles and processes in the industry. “I think there is an opportunity for pharma to rethink processes and rethink roles in light of these new and enhanced support tools that we have,” he says. “There’s an opportunity for everyone to be a lot more efficient and for our really smart people to work on things that require creativity, which we know a lot of machine learning and AI aren’t as great at. They’re great at answering questions, but they’re not as great at coming up with those questions themselves.”
Filed Under: clinical trials, Data science, Drug Discovery, machine learning and AI, Uncategorized