If the data is new oil, AI is a refinery

“Data is a new oil,” we are beginning to hear more and more of the statement. and panelists 2025 Bioprocessing Summit Boston City Hall agreed. Irene RombelCEO of BioCurie. Cenk Undeyleads Sanofi's global ICMC digital transformation program. and Colin ZickFoley Hoag's managing partner spoke to this idea that data is invaluable raw materials in today's technology-driven era.

However, process development and manufacturing data are small and sparse, especially for more advanced treatments. This means it is not very suitable for traditional artificial intelligence/machine learning (AI/ml). Contrary to popular belief, she said, “There is no AI/ML Data Goldmine to utilize in manufacturing.”

However, when discussing AI, data/oil analogy may be appropriate. Like oil, data is inherently dirty. Therefore, data must be invested in time and money to extract any value. This is where AI tools are entered as data refiners.

Data Oil Drip

The small and sparse nature of process development and manufacturing data is particularly important in understanding the use of AI/ML. If trained in a robust, carefully curated or clean industry, it is suitable for interpolation, but it cannot predict anything beyond that dataset.

A good example of this was quoted by Zick. a Bloomberg ArticleZick explained that of all the challenges facing Boston's self-driving cars, the biggest one is urban seagulls. Like a Western standoff, the two stare at each other until Segal decides to move. In other words, if the AI/ML model is not exposed to data, it is not possible to properly predict results.

That being said, Undey pointed out that the ML/DL model is a common myth that requires millions of data points. It depends on the quality of the data. There are many calculation methods developed to handle limited datasets.

The consequences of confusion

The power of AI as a whole has caused confusion. This was clear at the meeting. Zick “I think it's the most confusing thing about AI and manufacturing if you think it's special or different from other tools that use AI.” He further explained that this idea leads to a lot of sense that the usual rules for using new tools don't apply. This is extremely dangerous.

Undey supported the statement. “It's a common pitfall to jump straight to AI tools like machine learning algorithms when people have data without understanding basic statistics, data they're trying to solve, data understanding, problems,” he rebutted another common myth that AI is suitable for almost any situation. “AI/ML methods may not be suitable for tasks, but simple, ordinary processes can make the job even better.”

Rombel brought up what was called a particularly unsettling topic that was repeatedly discussed during the meeting. Composite data that uses large-scale language models (LLM) and generates “missing data.” Scientists such as Mark Mackey and Cressett's CSOhas been alarming about this potential use of AI, and we understand enough to make sure that the industry as a whole has no understanding of biology, that we have all the data that AI needs to do this. The panelists also raised concerns, with Lombel stressing that “it's not only a bad practice from a data modeling perspective, it can be dangerous.”

The other side of this coin is that some still question how digital conversion and AI can dramatically improve the time, quality and/or costs of process development. These topics were highlighted not only in the City Hall but also in the entire conference, so it is clear that parts of the industry need to be more persuasive while other parts of the industry are off and running.

During City Hall, Undey was surprised to focus on topics that he felt were already addressed in the past, such as explainable AI and GMP risk-based applications. While it's good to continue the ongoing interest and discussion, Undey said, “I was hoping to get more questions in AI's robotics and Genai Frontiers, as applied to bioprocessing.”

Don't take a walk

To stay competitive, both leaders and operators need to understand that digital transformation and AI are essential. Undey advised on how to minimize prediction errors by addressing aspects of learning as AI/ML models evolve and adapt. In his view, it is all summarised in the risk assessment, use points, and how to document the calculation method.

As a panel lawyer, Zick was briefly positive about his perspective. He emphasized that “in so many situations, people are applying AI critically inconceivably.” He further explained that AI is a tool for critical thinking and is not a substitute for critical thinking. He is a very emphatic leader and scientists should read the new FDA Draft Guidance He brought a paper copy of how AI can support regulatory decisions. “Read that.”

Final thoughts

Rombel suggested that the idea of ​​collecting as much data as possible, so switch to getting the right data, which can be costly. “This is where subject expertise and robust statistical analysis are equally important.” Undey stressed that “using AI is not the goal. It is more important to use it where it makes sense and ask the right questions to solve the right problems.

Zick put it together best with his approach to AI. “It can make everything better and nothing should make it worse.” To achieve this goal, Undey said, “The industry is supplying AI perfectly and therefore we need to continue or focus our data products.” He explained that the development and advancement of these models for decision-making is a critical step in the continuation of product development and biopharma manufacturing. “Our industry is not on this aspect yet, but there are many success stories in different segments of the value chain.”