Addressing the AI Adoption Gap

At INTERPHEX 2026, which is being held in New York City from April 21-23, the use of artificial intelligence (AI) in pharmaceutical development and manufacturing was a popular topic. The Panel Discussion: “Closing the AI Adoption Gap in Pharmaceutical Development and Manufacturing”¹ featured Shane Grosser, executive director, Pharmaceutical Analysis and Digital Technologies at Merck; Nicholas Guros, director, Upstream and CLD Digital and AI Transformation at AstraZeneca; Gorgi Pavlov, principal scientist at Johnson & Johnson (J&J); and Joshua Sperry, associate director, Automation at Regeneron. The panel was moderated by Kiefer Eaton, head of Pharma & Chemicals at Basetwo AI. The discussion centered on the practical application of AI, machine learning (ML), and digital twins to optimize process development and manufacturing while addressing the systemic challenges of data integrity and organizational culture.

What Are AI’s Advanced Applications in Process Development?

A primary focus of the discussion was the move toward hybrid modeling. The experts detailed the use of neural ordinary differential equations (neural ODEs), which combine machine learning with mechanistic understanding to predict bioreactor performance. This approach allows scientists to forecast day 12 titers with high accuracy as early as the beginning of a run by analyzing glucose profiles and environmental conditions.

The use of modeling to bridge gaps between experimental and theoretical work, specifically aiming to predict critical quality attribute (CQA) failures, was highlighted. For high-value products, where batch failure is not an option, these models serve as a tool to translate laboratory insights to commercial-scale manufacturing. At AstraZeneca, the focus includes using digital twins to reduce the volume of analytical assays required, thereby streamlining the development lifecycle.

What Is the Most Significant Barrier to AI Adoption?

Despite the promise of AI, the panel unanimously identified data foundations as the most significant barrier to progress. While advanced ML models exist, the lack of structured metadata and the prevalence of data silos—spanning PDFs, Excel files, and SharePoint—hinder scalability.

To combat this, the panelists proposed two distinct strategies:

• Automated data structuring. Manual data structuring may create an undue burden on scientists. Intelligent controls and instruments can automatically generate machine-readable, structured data at the point of collection.
• Unified namespaces: There is a need for digital shadows and a unified namespace across sites to eliminate data transients and transcription errors, ultimately aiming for a vendor-agnostic digital ecosystem.

Cultural Transformation and Trust

The transition to AI-driven manufacturing requires a cultural shift, as well as a technical one. The panel discussed the higher bar of trust applied to AI systems compared to humans; while a human error might be overlooked, a single AI hallucination can trigger skepticism. Building trust requires the following:

• Incremental validation. Starting with low-risk applications, such as clone selection, and proving statistical confidence before scaling up.
• QA collaboration. Engaging quality assurance colleagues early through computer software assurance (CSA) frameworks to ensure that AI-driven, risk-based decisions meet regulatory standards.

What Is the Business Case for AI?

The return on investment for these technologies extends beyond simple cost-cutting. Panelists highlighted that AI could lower the cognitive load on scientists by automating bookkeeping and data parsing, allowing them to focus on the high-level research they were trained for.

Furthermore, with a reported 20% year-over-year increase in data integrity observations across the industry, AI is increasingly viewed as a defensive necessity. By “hardening” data through automated pipelines and AI-driven scripts, companies can support increasingly complex pipelines with existing staff levels while maintaining rigorous compliance.

The path forward for the adoption of AI in pharma processes may involve moving away from manual data management toward integrated, automated systems that empower human expertise with predictive, machine-led insights.

Visit PharmTech.com’s Conference Page for more from INTERPHEX 2026 and other conferences.

Reference

1. Eaton K, Sperry J, Grosser S, Guros N, Pavlov G. Panel Discussion: Closing the AI Adoption Gap in Pharmaceutical Development and Manufacturing. INTERPHEX 2026. New York, NY. April 21, 2026. https://www.interphex.com/en-us/education/education-schedule/session-details.4766.258092.panel-discussion-closing-the-ai-adoption-gap-in-pharmaceutical-development-and-manufacturing.html