65th ISI World Statistics Congress 2025

65th ISI World Statistics Congress 2025

When are "Black box" algorithms justified?

Author

JR
Jordan Rodu

Co-author

  • M
    Michael Baiocchi

Conference

65th ISI World Statistics Congress 2025

Format: IPS Abstract - WSC 2025

Keywords: ai, algorithmic-appropriateness, machine learning

Session: IPS 983 - Bias, Variation, Error, and Interpretability of Algorithms Used in Forensic Applications

Thursday 9 October 10:50 a.m. - 12:30 p.m. (Europe/Amsterdam)

Abstract

Statistical algorithms have a long history for prediction and/or classification (discriminant analysis, logistic regression), but recently have been overshadowed by "Black-box" algorithms in almost all disciplines. When are (or aren't) black-box algorithms justified? In this talk we describe Outcome Reasoning as the engine that powers the development of black-box algorithms, and contrast it with the more familiar Model-Based Reasoning in statistics. We discuss the strengths and limitations of each, and provide a framework and multiple examples across several domains, so both technical and non-technical people can discuss and identify key features of their prediction problem that will help them decide when to use one versus the other.