Where Life Science Should Focus its AI Investments

 Where Life Science Should Focus its AI Investments

by Beate Mueller-Tiemann, PhD., Chief Technology Officer, Cytiva

Artificial intelligence is now part of our daily work in the life sciences industry. I am a believer in its potential and I see opportunities for it to help scientists work more efficiently, discover patterns in complex information, and make better-informed decisions.

But belief in AI does not mean that every AI application deserves investment.

As interest grows, companies risk approaching AI as a technology in search of a problem. They acquire new tools, launch pilots, and measure progress by the number of applications deployed. Yet that activity does not necessarily translate into better science, faster development, or more capable manufacturing. Similarly, metrics such as token consumption may measure AI activity, but they say little about its actual business impact, whether in R&D or any other function.

Rather than asking: “Where can we use AI?” the more important question is, “Where can AI help us solve a consequential scientific or operational problem?”

Life sciences companies must invest in the foundation and capabilities that make AI valuable to their business. The organizations that gain the greatest advantage will not necessarily be those with the most AI tools. They will be those with contextualized, FAIR (F – Findable, A – Accessible, I – Interoperable, R – Reusable) data, data analytical capabilities, connected and ideally autonomous workflows, and the scientific expertise required to turn the insights generated by AI into faster learning and better decisions.

Make Scientific Data Usable, Not Simply Digital

The first investment should be in data. AI can only learn from the information available to it. In life sciences, that information may be spread across instruments, laboratories, electronic notebooks, manufacturing systems, and  external partners. It may also be recorded in different formats and without the scientific context needed to interpret it properly.

Many organizations have spent years digitizing their operations, including R&D. But digitized information is not automatically AI-ready information. Data should be captured directly from connected instruments with experimental context. A company may have electronic laboratory notebooks, also dubbed as “paper on glass”, and still struggle to locate or compare experimental knowledge. Different teams may record similar information in different ways, which is viewed as lack of standardization. Results may be captured without the experimental or process context needed to reuse them. Data generated during early development may not connect effectively with the information required later in the product lifecycle.

Before investing in another AI application, companies should ask whether their most valuable scientific knowledge is accessible to the intelligence they are buying.

Can scientists find and reuse previous experimental knowledge? Can laboratory data be connected across teams and workflows? Can development data inform later manufacturing decisions? Is the organization capturing failed experiments and unexpected findings, as well as successful results?

If the answer is no, the first investment should not be another algorithm. It should be a better data foundation.

Put AI Close to the Scientific Learning Cycle

The second priority should be to invest where AI can strengthen the scientific learning cycle.

Science advances through iteration. We design an experiment, analyze the results, develop new understandings, and use that knowledge to make the next decision. The potential of AI is not simply to automate one step in this process. It is to help the entire cycle operate more effectively.

Consider the amount of knowledge generated during process development. Scientists may investigate multiple parameters, raw materials, analytical methods, and operating conditions. Yet the value of those experiments is limited if the resulting knowledge remains isolated within a particular team, program, or system.

AI could help organizations recognize relationships across a broader body of information, identify the next most informative experiment, or reveal patterns that may not be apparent through traditional analysis alone. This could allow each experiment to contribute more effectively to the next scientific decision.

The objective is not to remove scientists from the process, but to remove the muda. Scientific knowledge becomes even more important as the technology becomes more powerful. Scientists must determine whether that output makes biological and process sense. They must understand the quality of the underlying data, consider competing explanations, and recognize when an answer needs to be challenged. This is comparable to us challenging the results on the AI models we use in our everyday life.

AI should extend scientific judgment, not attempt to replace it. The most meaningful measure of value, therefore, is not how many activities an AI system performs. It is whether the system helps the organization learn faster and make better decisions.

Invest in Measurement and Analytics Alongside AI

Companies should also avoid viewing AI investment separately from investment in data analytical science.

An AI model cannot extract meaning from information that was never adequately measured. If we want more intelligent processes, we first need process insights, understanding relevant process parameters.

This is especially important as the biopharma industry develops increasingly complex therapies and modalities. With many of these new modalities, even now more than ever, the longstanding adage of a Biologic drug holds true: “the process is the product”. Better measurements of the process produce richer information. Richer information can support better process models. Better process models can deepen process understanding and support better decisions and eventually more robust and reliable biomanufacturing processes.

Companies should consider investments in data analytical technologies, data contextualization in combination with process monitoring and in AI as parts of the same scientific learning loop system.

The goal is not simply to generate more data. It is to generate the right information to gain insight so that a human in the loop can act on it.

Target Bottlenecks, Not Convenient Use Cases

Organizations will naturally gravitate toward AI applications that are easily piloted and a “quick win” can build confidence and demonstrate that a technology works. A successful pilot is important to drive technology adoption powering change management in the workforce, but it might not necessarily drive maximal business impact.

The value of AI should be measured by the importance of the bottlenecks it can remove. Companies should begin by identifying what is actually slowing down scientific learning, process development, or manufacturing. Where are scientists spending time searching for information that already exists? Where are teams repeating work because previous knowledge is difficult to find or reuse? Where are repetitive tasks which are consuming resources? Where does knowledge move slowly between research, development, analytical and manufacturing groups? Where are decisions delayed because relevant information is not available in a usable form?

These questions direct AI investment toward consequential problems.

They also help organizations avoid automating inefficiency. Applying AI to a poorly designed workflow may make parts of the workflow faster without improving the broader system. In some cases, it may add another layer of technology without addressing the reason the process is slow. In these situations, the next bottleneck will expose itself more prominently, ultimately impeding overall productivity gains.

AI investment should begin with the constraint, not the tool.

This requires companies to look across organizational boundaries. A system that saves time in one laboratory may create little overall value if it does not connect with subsequent development or manufacturing decisions. By contrast, an application that improves the transfer of process knowledge across the product lifecycle could have a much broader impact.

Keep Scientific Judgment and Accountability in the System

AI does not eliminate the need for clear decision-making. It makes clarity even more important.

Organizations need to establish who is responsible for evaluating an AI-generated output, what evidence is required before acting on it, and how the basis for an important decision will be documented. Scientists and leaders should understand how AI informed a decision without allowing the technology to obscure accountability for that decision.

This will require investment in people as well as platforms. The greatest value will come from connecting scientific, engineering, analytical, data, and digital expertise. Scientists need enough understanding of AI to question its outputs effectively, while data and technology specialists need sufficient context to understand the scientific problem being addressed.

The competitive advantage will come from the interaction between technology and expertise and will be created at the interface of disciplines.

Invest for the AI We Do Not Know Yet

No one can say with certainty what AI will look like five years from now. Models will evolve, tools will change and new applications will emerge.

That uncertainty should not prevent investment. It should shape it. Life sciences companies should invest in capabilities that will remain valuable as AI changes: high-quality data, strong analytics, connected systems, scientific expertise and rigor and an organizational ability to make timely, informed decisions.

The goal should not simply be to become an AI-enabled life sciences company. It should be to become a learning organization in which every experiment, dataset, and process can contribute more effectively to the next scientific decision.

That is where AI begins to become transformative. Not when it replaces how we work today, but when it allows us to learn in ways that were not previously possible.

About the author

Dr. Beate Mueller-Tiemann leads Cytiva’s efforts to advance innovative R&D and manufacturing technologies. She brings extensive experience across the pharmaceutical value chain, spanning drug discovery, development and commercial manufacturing. She holds a PhD in Molecular Oncology from the Leibniz Institute of Virology in Hamburg, Germany.

Subscribe to our e-Newsletters!
Stay up to date with the latest news, articles, and events. Plus, get special offers from Labcompare – all delivered right to your inbox! Sign up now!

References

  1. 000
  • <<
  • >>