Generative AI can process information at incredible speed, but it cannot yet think like a scientist. In biopharma R&D, its real value lies in assisting scientists, automating routine tasks, interpreting data in context, and moving research faster.
AI can add value to science in certain context. This need has led to the emergence of third-generation Electronic Lab Notebooks (ELNs), sometimes referred to as Artificially Intelligent Lab Notebooks (AILNs) – AI-native platforms that go beyond documenting experiments.
Yet there are limitations: Many models often fail to distinguish between a sample and a reagent, and cannot interpret assay results in context or anticipate whether a protocol step is valid or flawed. They know a lot, but do not think like scientists.
This author has demonstrated this when assessing microbiological data in the context where increasing numbers of bacteria is a bad thing, AI has indicated that the growth is good – a conclusion unsuited to the task.
To understand these mixed results in a scientific context, Digital Journal has heard from Andrew Wyatt, who is the chief growth officer at Sapio Sciences.
AI complements, it doesn’t replace
Wyatt says: “In biopharma R&D, however, the question is not simply what AI can do, it is how it should help. Since the release of modern generative AI tools, there has been speculation about whether these systems could one day replace scientists, with suggested use cases ranging from accelerating literature reviews to protocol drafting. While these models are capable of impressive analysis and pattern recognition, they struggle to apply true scientific reasoning, understand experimental intent, interpret results in context, and link data to hypotheses.”
So, what can be done with AI as it currently stands. Wyatt thinks “the real opportunity for AI today is not as a replacement; it is as a complement to the tools and scientists already driving innovation. The issue is not that generative AI models aren’t powerful, it’s that they are designed to be broadly useful across many domains. They are trained using public content and generalised data, not the proprietary, structured, and experimental data that drives biopharma R&D.”
This is because, Wyatt points out: “Generative AI may excel at handling language, but it still lacks scientific fluency. These models often fail to distinguish between a sample and a reagent, and cannot interpret assay results in context or anticipate whether a protocol step is valid or flawed. They know a lot, but do not think like scientists.”
From assistance to agency
On the plus side, Wyatt says: “AI can help streamline repetitive or administrative lab tasks, assist in drafting workflows, and suggest possible interpretations of structured data. For most scientists, decision-making and creative problem-solving are core to their work. There is little appetite and no current need to give that up.”
In terms of examples: “Some of the most impactful AI applications today are focused on accelerating the gap between idea and execution. Rather than automating science itself, AI can reduce the manual burden of tasks that pull scientists away from research. This helps scientists stay focused on the science itself, rather than getting pulled into administrative or technical detail.”
AI can:
- Translate high-level experiment descriptions into structured steps and protocol templates
- Retrieve data based on contextual, natural-language queries, rather than requiring complex filters or forms
- Track materials and consumables based on protocol logic
- Guide scientists through unfamiliar lab software, reducing the learning curve for new tools
However, Wyatt observes: “To do this well, the AI must not only understand language. It must understand science, and it must be embedded in scientific software that reflects real-world research environments.”
