How do mathematical models help predict the future? - with Erica Thompson

About this Video

• Video Title: How do mathematical models help predict the future? - with Erica Thompson
• Channel: The Royal Institution
• Speakers: Erica Thompson, Students (unidentified)
• Duration: 00:58:48

Introduction

This Royal Institution lecture by Erica Thompson explores the use of mathematical models in predicting the future, focusing on the distinction between data-driven and assumption-driven models. Thompson examines how model quality is assessed, the role of values and ethics in model creation, and the challenges of interpreting model outputs in relation to complex real-world scenarios.

Key Takeaways

• Data-driven models rely heavily on existing data for predictions within observed ranges (interpolation), offering higher confidence. Examples include short-term weather forecasting and mechanical testing.
• Assumption-driven models extrapolate beyond observed data, relying more on assumptions about future conditions. These models are crucial for predicting long-term trends like climate change impacts or pandemic evolution, but carry greater uncertainty.
• Model evaluation depends on the intended use. Data-driven models are evaluated against new data, while assumption-driven models require expert judgment on the validity of underlying assumptions. Different perspectives (statistical, dynamical, social) significantly influence model evaluation.
• Values are inherent in models. Choices about which data to include, how to simplify reality, and which assumptions to make reflect the modeler's values and priorities, impacting model outputs and their interpretation. Ethical considerations are paramount.
• Ensembles of models don't necessarily increase confidence. It's more valuable to demonstrate that no plausible model contradicts a specific prediction. Defining "plausible" requires expert judgment and raises questions about scientific gatekeeping.