On November 27, 2025, ERA Sciences joined industry leaders at the PDA Ireland Emerging Technologies conference to discuss one of the most fundamental challenges in regulated AI: Quality Asymmetry.
The presentation dove into the reality that models excel at what they were trained on, but provide no predictive confidence for what they were not trained on. This is the core difference between traditional software and statistical machines.
Now, you can download the slides presented by Andy O'Connor and explore the strategies for validating and managing AI models in GxP environments.
This presentation provides a critical framework for quality professionals to navigate the shift from deterministic software to fundamentally probabilistic machine learning.
The Turkey Problem: Understand the danger of relying solely on past data and how it can create false confidence, especially in regulated environments where patient safety is at risk.
Mild vs. Wild Risk Environments: Learn to differentiate between controlled environments (mild risk, Gaussian distributions) and complex, real-world scenarios (wild risk, power laws) to appropriately assess model applicability and vulnerability to 'black swan' events.
The Quality Asymmetry Principle: A deep dive into the four failure modes: False Positives, False Negatives, the insidious Quality Asymmetry (failing to detect unknown anomalies), and the dangerous Gaussian Assumptions.
Validation for Statistical Machines: Discover how validation is shifting from proving correctness to demonstrating controlled uncertainty, quantifying and managing model risk within acceptable limits.
Managing Model Risk: Explore the three-step framework for managing model risk: defining the Context of Use (COU) to acknowledge asymmetry, assessing Model Influence and Decision Consequence, and developing a Credibility Assessment Plan (CAP).
Model Cards for Critical Thinking: Learn how transparency about model uses and limitations, communicated via a standardized Model Card, enables quality teams to assess and manage the probabilistic impact on product quality.
These slides are essential for decision-makers and technical leaders building, deploying, or vetting AI solutions in the pharmaceutical and life sciences industries. Perfect for:
Quality and Validation Leaders
Data Science and Digital Innovation Teams
Regulatory Compliance Specialists
Anyone concerned with the real-world reliability of AI in GxP environments
Attendees walked away with a new perspective on how to responsibly deploy AI, focusing on the ultimate goal: demonstrable improvement over current practice. You will learn how to:
Manage the inherent quality asymmetry by rigidly defining what the model is and is NOT designed to do.
Evaluate the combined performance of the Human + AI team, rather than the model in isolation, to find real-world benefit.
Use risk assessment to map model limitations to potential quality and safety impacts.
Fill out the form below to download a copy of the PDA Ireland Emerging Technologies presentation slides on AI Quality Asymmetry and start applying these insights to your own quality assurance strategy.