[Adobe Stock]
But deploying such tools at scale requires a mix of strategic thinking, curiosity and new approaches to cultivating talent. As we enter 2024, life science organizations must rethink their approach to addressing the skills gap. As Bain has noted, almost three-quarters of engineering and R&D-focused companies struggle to find skilled talent. Meanwhile, some companies fall into binary thinking that prevents them from tapping the full potential of AI tools. “I’ve seen cases where you give them a technology and they think, ‘Oh, the technology didn’t do all of this stuff perfectly, therefore it’s a failure and I have to do the whole thing,’” said Michael Connell, chief operating officer at Enthought. In reality, the most effective strategy is to strike a balance between total reliance on technology and complete human execution.
Between two poles: Toward an AI-human synergy
Michael Connell
The journey of revamping traditional drug discovery and development workflows involves exploring how humans and the AI are working together in “different configurations,” Connell noted. “Humans, right now, certainly need to be the executive,” he said. Scientists need to thus help foster a symbiotic relationship where they guide and supervise AI tools, ensuring that they align with scientific rigor and real-world needs.
Ultimately, generative AI tools can help researchers raise the “floor” of efficiency while also lifting the “ceiling” of innovation, enabling more rapid response to emerging health challenges. In addition to the use cases outlined earlier, generative AI can assist with clinical site recruitment and selection, monitoring drug reactions, and assembling regulatory information. While positioning generative AI at the peak of its Hype Cycle model, Gartner projects a relatively swift embrace of the technology, which could reach a plateau of mainstream adoption in two to five years. The analyst firm highlights emerging global regulations and the inherent risks of incorrect scientific assumptions or recommendations generated by AI as potential headwinds.
The alien artifact of AI
Complicating matters is the enigmatic reality of generative AI systems, which calls for a new approach to system building and integration. “And I think what needs to happen is we have to rediscover how to build systems around this,” Connell said.
On one end of the spectrum is a reality where scientists have “figured things out,” Connell added. “We know how to do certain biological protocols, certain chemical kinds of formulation processes,” he noted. “And then you introduce this alien artifact that doesn’t snap in like a computer program that you wrote would.” Therefore, devising a strategy for how to use generative AI involves some poking, prodding and studying its capabilities, understanding that they may evolve over time.
Overcoming the generative AI rabbit hole
But because of the elusive quality of generative AI systems, an inherent challenge exists in understanding and predicting their behavior. Unlike traditional technologies that follow a linear and predictable path of development, generative AI models can behave in ways that can be counterintuitive or unexpected. Most notably, they can hallucinate, generating incorrect or nonsensical information despite receiving a correct prompt.
There’s a necessary period of exploration to understand how to effectively integrate these tools into systems which can involve experimenting with prompt engineering. But even prompt engineering can be something of a rabbit hole given the lack of clarity about what a generative AI system knows and doesn’t know, Connell said. “You’re trying to figure out, ‘Does it know this thing, but I’m not asking the question correctly?’” Connell said. “You can even go and research prompt engineering as a rabbit hole.”
Such complexities are note holding back Big Pharma firms such as AstraZeneca, GlaxoSmithKline, Sanofi, Bayer, Novartis, and Roche, which are actively forging partnerships with generative AI-focused biotech firms such as Exscientia, Insilico Medicine, and BenevolentAI. The aim is to integrate emerging technologies into traditional drug discovery and development workflows. These collaborations, along with in-house AI investments, aim to tap tools like machine learning and generative models to speed target identification, molecule design, preclinical testing, and other processes to chip away at the time and expense of traditional drug development. In August, Insilico Medicine announced that it had made significant progress in predicting clinical trial outcomes with relatively high accuracy, which has largely eluded AI researchers in the past.
A rational approach to dealing with irrational emotional reactions
Another wrinkle that can complicate deployment of generative AI technologies is the considerable hype and fear surrounding the technology. On the one hand, some professionals have a fear of missing out or falling behind competitors while on the other, others worry about the impact the technology can have on jobs, workflows, and the meaning of being human. Navigating such emotionality involves acknowledging the complexities of responses and organizational dynamics while developing a rational strategy. The ultimate goal is to merge the human ingenuity of scientific research with the capabilities of generative AI. The technology has the potential accelerate drug screening, streamline workflows, automate routine tasks, and support data analysis while ensuring systems are in place to ensure accuracy.
“There’s a lot of resistance to change, partly because it’s hard to find processes in R&D that work,” Connell said. Some researchers might think, “This is science. It’s hard enough to do what we do, and now you’re going to throw this thing at me?” While some scientists are excited by the prospect of automating mundane tasks such as using stopwatches to time mouse swimming in an experiment, others feel anxiety that such capabilities threaten their standing and livelihood.
Reactions differ greatly based on factors like career stage, hierarchy level, and personality. “I think the leaders generally are thinking, ‘We need to figure this out,’” Connell said. While rank-and-file employees hold a variety of opinions on generative AI, they are less likely to think deeply about its impact on their respective industry and the future of their career.
Unfreezing middle-management inertia
In discussing the challenges of implementing new technologies such as generative AI, Connell points to the “clay layer” as a significant barrier. A metaphorical term describing the tendency of innovative ideas and progress to get stuck in middle management. “We call this the clay layer, because things get stuck there,” Connell said. Because middle managers tend to be mid- or later-stage of their career, “they’re not incentivized to figure this out.” Instead, they are incentivized to maintain the status quo until they retire or get a promotion rather than embracing new, potentially disruptive technologies.
If life science researchers need some degree of software engineering proficiency or data analysis in Python, many organizations will offer them a comprehensive overview of both of those domains. Connell argues that such an exhaustive approach is rarely necessary. It may be more effective and efficient to focus on teaching them fundamental, practical skills, such as how to name variables in their code or basic approaches for data manipulation and analysis in Python. Connell stressed the value of the Pareto Principle here, suggesting that teaching the crucial 20% of skills that are directly applicable to their roles can enable researchers to perform 80% of their tasks more effectively.
To help melt the ice stalling new technology deployments, Connell recommends that life science companies carefully choose tools tailored to the specific needs and workflows of individual roles. Additionally, they should resist the temptation of overwhelming employees with excessive training. By employing the Pareto principle, organizations can focus on training employees in the crucial 20% of tools and skills that will cover 80% of their operational needs. This tactic not only smooths out the learning curve involved in new technology adoption but also ensures that the workforce remains agile and adaptable in a swiftly-changing technological landscape.
The power of hands-on learning
Connell observes that while conventional training methods like classroom learning or online courses can be helpful for imparting theoretical knowledge, real-world experiential learning is invaluable. “In six months of mentorship, you can take someone who’s not even sure how they can do work away from the lab bench, because they’ve never thought about the data as something separate that could be analyzed and worked on scientifically, to somebody creating dozens of apps a year using digital technologies to solve actual scientific problems for themselves and their colleagues,” he said.
The key difference is hands-on application versus passive absorption of concepts. By guiding researchers to build artifacts and technologies that provide value in their day-to-day work, they rapidly acquire the subset of skills needed to continue creating apps and leveraging technology autonomously. Rather than aiming to comprehensively teach them data science or software engineering, the goal is imparting enough pragmatic knowledge to unlock new modes of working.
Investing in new systems
While it is understandable that strategically rethinking operations is often tough work, the stakes can be high for organizations that choose to bury their head in the sand. Connell points to the emergence of the Apple II, which decimated the business of Digital Equipment Corporation (DEC) in a matter of months in the late 1970s. “You have to think about the whole system that’s involved when you’re making choices about investment and priorities,” he said.
As Connell pointed out, the integration of new technologies like generative AI into life science organizations is not just about incremental improvements but about fundamentally transforming the organizational DNA. This echoes the transformative journey of companies like Amazon, which started out as an online bookstore in 1994 and evolved into a tech giant whose retail sales alone could top $746 billion in 2023. Contrast that with brick-and-mortar-based bookstores. Similarly, the emergence of tools such as generative AI offers life science organizations unprecedented opportunities to redefine their operational models, creating what Connell called “new DNA.” “You’re not just optimizing and replacing one process with a faster version,” he said.
Filed Under: Data science, machine learning and AI
