The inevitable impact of AI and machine learning
The impacts and limitations of AI in life sciences
2023 witnessed an unprecedented surge of activity in large language models (LLMs) as vendors pumped out new features and products in surprisingly accelerated style and fanfare. The life sciences industry is no exception for falling into a technology hype cycle with high—perhaps even unrealistic—expectations and disillusionment. AI-driven drug discovery and development is certainly no exception; recall IBM’s Moonshot to solve complex biological problems and deliver precise diagnoses and treatments at unimaginable time and scale? What the industry has learned since is we must all acknowledge that AI has limitations, integration must be crafted carefully, and expectations need to be set and reset. A recent Economist article highlights that the impact of AI to date isn’t significant and that many pharma CEOs worry that the tendency of GenAI to fabricate information could send researchers down blind alleys.
There’s little question that GenAI will transform how life science companies will develop, produce, and commercialize new products to market and make impactful differences to clinical care teams and the lives of patients. Already proven in other industries, it is increasingly evident that GenAI will make similar impacts on the healthcare ecosystem. We are seeing impacts today in biopharma, digital health, and MedTech, with significant benefits and measurable outcomes.
GenAI use cases for enhancing efficiency and increasing value
To discuss areas where we see the most activity and the most value being unlocked through GenAI in biopharma and MedTech, it is critical to first frame the identification of use cases into two general categories:
1. Foundational use cases, which we also refer to as “table stakes”
These are areas where you can deploy GenAI into routine everyday tasks to improve efficiency and enhance operations that are often cost-focused. Typically, the benefits come from leveraging some off-the-shelf solutions here (e.g., enterprise and mature LLM subscriptions). These commonly range from mundane tasks to more complex operational sides of the drug and device business, but it is critical to embrace these simple use cases because failing to do so will result in a loss of competitiveness.
2. Impactful enterprise use cases
These are critical, highly visible, and build long-term advantages. They often leverage data assets of different companies and may be internally owned and curated data sets. Commonly, they are unstructured data, thus heightening the criticality of executing your table stakes use cases to assure data governance and readiness for ingestion. We may see them deployed in areas where there’s high potential for disruption (e.g., clinical trials startup, monitoring, filings). The opportunity loss of inaction would run the risk of becoming irrelevant and fundamentally being disrupted by other players or new entrants as they may have better, smarter, and superior methods of leveraging data sources.
Here are a few impactful enterprise use cases where we see the most value across the biopharma and device lifecycle:
1. Discovery research
Advanced AI methods like GenAI have been actively deployed for discovery research for many years now. Example applications include high throughput target validation (e.g., molecules and compounds of interest) and their interactions at the molecular level (e.g., 3D models, protein folding structural analysis, etc.).
2. Clinical development and clinical operations
Using AI to generate documentation that is essential for clinical trials and medical development and operations can readily decrease study start-up time and complexity, patient recruitment, study conduct, and close out key activities and documentation, translating into faster trials and getting drugs to patients faster.
3. Quality operations management
The quality assurance function runs across the entire value chain. AI is used to write quality reports, evaluate deviations, and define mitigation strategies, including investigations and continuous improvement feedback loops. AI use in pharmacovigilance, or drug safety reporting and identification of adverse events, continues to increase.
4. Commercial content generation, assembly, and processing
AI can generate content on the commercial and clinical care team side by helping teams focus on effective methods of communication and leveraging and enabling accelerated localization of the identified content to the right channel to scale (e.g., social media, direct personalized communication to physicians and patients).
5. Medical, legal, regulatory review (MLR)
Due to the vast amount of data, MLR processes are overburdened and remain bottlenecks. GenAI is actively used in this process via onboarding and pre-screening bots, aids, and co-pilots; appending, editing, and updating content; and posting MLR reviews with refined suggestions on how to improve, increase compliance, stratify risk, and triage content (depending on the risk inputs).
Integrating responsible AI principles for human empowerment
In 2024 and beyond, we’re seeing a renewed focus on a human-centered approach and responsible AI principles where AI-based technologies accelerate and empower the role of clinical scientists, clinical care teams, and their support staff. The view of AI as a complement, not a substitute, means a renewed focus on integration.
Key points to think through in your integration planning:
AI is not replacing humans; instead, it’s creating a situation where people are working on smarter things, while AI takes on basic routine things and gives more insight that people can work from.
Data is the core of everything in AI—both the first-party data as a biopharma and device company and the third-party data you leverage (e.g., LLM procurement). It is the data, along with the data in the applications and models, that creates the proprietary advantage for your organization; therefore, data quality and governance are imperative to apply GenAI effectively.
People and the associated processes and policies are at the core; thus, talent upskilling, training, and rethinking the processes of AI in our day-to-day workflow are critical. It needs a cultural change that is deeply rooted in the operational ways of working.
Responsible and ethical AI must be adhered to. Good Machine Learning Practice must be part of the core policies as you scale, grow, and continue to increase and leverage AI in your organization.
We recommend jumping in and actively applying the agile principle of experimentation, with concurrent and significant progress towards the foundational table stakes use cases while assertively actioning the impactful enterprise use cases to ensure long-lasting value in your journey.
Slalom contributor: Alvin Lee
