For years, conversations about artificial intelligence (AI) in pharmacovigilance (PV) have revolved around a single question: will AI replace humans?
Today, that question is no longer the right one.
The real shift happening across the industry is not about replacement, but about collaboration. As highlighted in our recent whitepaper, PV is moving toward a model where automation, AI, and human expertise are seamlessly integrated into intelligent, adaptive workflows. Evidence shows that professionals are not all using AI in the same way. Instead, they tend to fall into three categories: self‑automators, cyborgs, and centaurs.
Understanding these three modes is key to understanding why pharmacovigilance is converging on one model in particular: the centaur.
Not All Human/AI Collaboration Is Created Equal
The Harvard framework provides a useful lens for understanding how AI is actually used in complex, high‑stakes work environments.
Some professionals operate as self‑automators, delegating entire tasks to AI with minimal oversight. This approach can deliver speed, but it risks sacrificing depth and critical thinking, especially in domains where nuance mattersor with high compliance needs.
Others work as cyborgs, engaging in continuous back‑and‑forth interaction with AI. In this model, human and machine co-create outputs through iterative dialogue, refining and challenging each other throughout the process.
But the model that stands out, particularly in regulated environments, is the centaur. Here, humans remain firmly in control, defining both the problem and the approach, while using AI selectively to enhance efficiency and improve outputs.
This distinction may seem subtle, but it has profound implications. It shifts AI from being a decision maker to being a decision support system; a powerful one, but one that operates under human direction.
And this is precisely the model that PV is gravitating toward.
Why Pharmacovigilance Naturally Favors the Centaur Model
Pharmacovigilance is not just another data-driven department. It is a highly regulated discipline where decisions can directly impact patient safety. As a result, the margin for error is vanishingly small, and the need for accountability is absolute.
The analysis in our whitepaper reflects this reality clearly. While AI is already being integrated into areas such as case intake, narrative generation, and regulatory monitoring, it is not positioned as a replacement for human expertise. Instead, it is designed to augment it.
This is why the centaur model fits so well.
In pharmacovigilance workflows, tasks tend to fall into three broad categories. At one end, there are highly repetitive, rules-based activities such as data entry, coding, and structured updates. These are increasingly handled by automation, reducing manual effort and minimizing the risk of human error.
In the middle, there are tasks involving pattern recognition and content generation, where AI performs particularly well. This includes drafting narratives, integrating structured data into reports, or screening large volumes of regulatory information.
At the other end, there are activities that require clinical judgment, ethical reasoning, and contextual interpretation. These include causality assessment, signal evaluation, and final safety decision-making, areas where human expertise remains indispensable.
The centaur model works because it aligns perfectly with this structure. It does not attempt to automate everything. Instead, it assigns each layer of work to the system best equipped to handle it.
From Use Cases to Operating Model
When you look at specific pharmacovigilance activities, this partnership becomes tangible.
In periodic safety reporting, AI can already support the integration of structured data and the drafting of text, significantly reducing the time spent on repetitive updates. This allows safety writers to focus more on interpreting results and ensuring that the narrative accurately reflects the safety profile.
In regulatory intelligence, AI can scan health authority websites, identify updates, and surface relevant changes. However, interpreting these changes, especially across different regulatory frameworks, still requires human expertise.
In case processing, AI can support the generation of case narratives, assist with data extraction from source documents, and enhance consistency in coding and data structuring. Yet the responsibility for assessing data quality, interpreting clinical context, and making reportability and causality decisions remains firmly with human contributors.
Across all these examples, the pattern is consistent. AI accelerates execution and enhances consistency, while humans retain control over meaning, interpretation, and accountability.
A Shift in What It Means to Be a Safety Scientist
Adopting the centaur model does not diminish the role of pharmacovigilance professionals; it transforms it.
As automation takes over routine tasks and AI supports content generation and analysis, safety scientists spend less time on mechanical activities and more time on high-value work. Their role increasingly centers on interpreting safety signals, validating outputs, and ensuring that decisions are clinically sound and ethically justified.
Importantly, this shift also requires new skills. Professionals must not only deepen their domain expertise but also how to effectively guide, challenge, and validate AI outputs. The question is no longer “Do you use AI?” but “How do you work with it?” as we discussed in our previous blog.
The Real Takeaway: It’s Not About Using AI, It’s About How
The most important insight from both pharmacovigilance practice and broader AI research is this: AI adoption alone is not enough.
What matters is how it is embedded into workflows and validation.
Professionals who fully delegate work to AI may gain speed, but risk losing control. Those who rely purely on iterative interaction may blur the boundary between human reasoning and AI output.
The centaur model offers a more sustainable path. It preserves human authority while leveraging AI where it adds the most value. It enhances performance without eroding expertise. In a field like pharmacovigilance, where accuracy, accountability, and patient safety are non-negotiable, that balance is not just desirable, it is essential.
About UBC
United BioSource LLC (UBC) is the leading provider of evidence development solutions with expertise in uniting evidence and access. UBC helps biopharma mitigate risk, address product hurdles, and demonstrate safety, efficacy, and value under real-world conditions. UBC leads the market in providing integrated, comprehensive clinical, safety, and commercialization services and is uniquely positioned to seamlessly integrate best-in-class services throughout the lifecycle of a product.
About the Authors
Dobrochna Dolicka, Safety Scientist, Global Safety Writing
Dobrochna Dolicka, PhD, serves as a Safety Scientist on UBC’s Global Safety Writing Team. For the last 3 years, she has been responsible for authoring periodic safety reports such as DSURs, PBRERs, and PADERs, as well as conducting other signal management activities. Ms. Dolicka is also working on the development and implementation of automation and artificial intelligence across UBC’s comprehensive pharmacovigilance services. She holds a PhD in Biomedical Sciences as well as a master’s in molecular biology.
Christopher Henry, Digital Initiatives Manager, Pharmacovigilance
Christopher Henry, serves as Digital Initiatives Manager for Pharmacovigilance at UBC. Passionate about the intersection of life sciences and technology, he leads initiatives focused on artificial intelligence, automation, digital innovation, and process transformation within pharmacovigilance. His work bridges scientific, operational, and technical teams to accelerate the adoption of modern solutions that improve patient safety and operational excellence. Before transitioning into digital leadership, Dr. Henry worked as a Safety Scientist, authoring global aggregate reports and supporting signal management activities across a diverse portfolio of pharmaceutical products. He holds a PhD in Cell Physiology and a Master’s degree in Genetics/Epigenetics, and has been contributing to UBC’s growth and innovation for more than four and a half years.
