Accessibility Tools

  • Content scaling 100%
  • Font size 100%
  • Line height 100%
  • Letter spacing 100%


AI in highly regulated settings

Example of pharmaceutical sector

Interview with Dr. Hans Klöcker, Manager at msg industry advisors


AI in highly regulated settings

Example of pharmaceutical sector

Interview with Dr. Hans Klöcker, Manager at msg industry advisors

AI validation in the pharmaceutical industry

The pharmaceutical industry holds high expectations for the potential of AI technologies. However, companies are facing challenges due to stringent regulatory demands that hinder smooth implementation. In the interview, Dr. Hans Klöcker, manager at msg industry advisors, sheds light on the regulatory requirements for utilizing AI in the pharmaceutical sector, current applications of AI solutions, and strategies for ensuring the functionality of AI-driven systems:

Dr. Klöcker, how are AI solutions utilized in the regulated setting of the pharmaceutical industry?

Hans Klöcker: AI and GenAI solutions are already being effectively utilized in various areas of the pharmaceutical industry. One notable area is drug development, where AI applications are employed to autonomously search vast libraries for potential new drug candidates. It is important to note that the pharmaceutical and medical technology sectors are subject to specific guidelines, such as "Good x Practice" (GxP), which outline stringent requirements for compliant working methods.

These guidelines necessitate the demonstration that all systems and processes utilized in the manufacturing process, which can impact product quality, data integrity, or patient safety, consistently produce the same outcome under identical conditions. The determination of whether a system is GxP-relevant and thus requires validation also depends on the level of autonomy of the system: If AI is supported by additional control mechanisms, such as human oversight or extensive approvals, its implementation is already feasible. However, the implementation of autonomous AI systems in the GxP setting has been challenging thus far due to the absence of binding regulatory requirements.


What challenges do companies face when validating AI solutions?

Hans Klöcker: Companies face significant challenges when validating AI solutions, particularly in the GxP setting. The requirement for transparency, comprehensibility, and reproducibility clashes with the nature of self-learning AI systems. Unlike traditional IT applications, AI systems do not rely on fixed, hard-coded sets of rules but instead adapt dynamically to new data and environmental conditions.

This continuous evolution of your own model can result in different outcomes for the same input data over time, and this non-deterministic behavior makes it difficult to follow a standard validation process. Consequently, companies may need to revalidate their AI systems after each learning, which is a time-consuming and resource-intensive task. As a result, the validation of self-learning AI systems for use in the GxP setting remains a significant challenge.


How should companies address the lack of regulatory requirements?

Hans Klöcker: Companies can address the lack of regulatory requirements by implementing AI/ML solutions that have already been approved and utilize closed or frozen systems. These systems have completed their learning process before being put into productive use, ensuring compliance. As AI solutions continue to mature, the industry anticipates the ability to implement true self-learning systems within the GxP setting. While awaiting the necessary regulatory framework, organizations must proactively prepare for the validation of AI systems: This involves developing standardized operating instructions (SOPs) for ensuring the safe operation of ML/AI systems.

These instructions form the foundation for future operations, outlining key framework conditions like roles for implementation and operation, monitoring procedures, risk classification of the ML/AI systems, and strategies for minimizing risks. On the other hand, the effectiveness and accuracy of the ML/AI model heavily rely on the quality and representativeness of the training and validation data. Developing procedures to systematically check and verify the balance of data sets is thus crucial for successful implementation.

hans kloecker msg 

Hans Klöcker


Weiterführende Themen

  • Getting your organization GenAI-ready

    Getting your organization GenAI-ready

    What prerequisites do organizations need to create to exploit the potential of GenAI solutions?
    msg has analyzed the status quo in German companies and identified the five most important fields of action.

    Read more

  • 1
  • Cloud Center of Excellence Part 3

    Cloud Center of
    Excellence Part 3

    How the Cloud Center of Excellence (CCoE) can contribute to a successful multicloud strategy and its implementation

    Read more

  • 1
  • Decentralization as a prerequisite for the heating transition

    Decentralization as a prerequisite for the heating transition

    The heating transition is a far-reaching transformation. Katrin Aytepe explains in an interview why the heating transition is only possible in a digital ecosystem.

    Read more

  • 1
© 2024 msg advisors