Hidden technical debt in Machine learning systems

TLDR

Machine learning enables rapid development of complex predictive systems. This paper argues that treating quick ML wins as cost‑free is dangerous and highlights the need to account for hidden technical debt. The authors identify ML‑specific risk factors—boundary erosion, entanglement, hidden feedback loops, undeclared consumers, data dependencies, configuration issues, external world changes, and system‑level anti‑patterns—that contribute to technical debt. Real‑world ML systems frequently accrue substantial maintenance costs due to technical debt.

Abstract

Machine learning offers a fantastically powerful toolkit for building useful complex prediction systems quickly. This paper argues it is dangerous to think of these quick wins as coming for free. Using the software engineering framework of technical debt, we find it is common to incur massive ongoing maintenance costs in real-world ML systems. We explore several ML-specific risk factors to account for in system design. These include boundary erosion, entanglement, hidden feedback loops, undeclared consumers, data dependencies, configuration issues, changes in the external world, and a variety of system-level anti-patterns.