A Classification of Feedback Loops and Their Relation to Biases in Automated Decision-Making Systems

TL;DR

This paper introduces a dynamical-systems framework to analyze how feedback loops in ML-based decision pipelines propagate and reshape biases over time. It formalizes five core loop types—sampling, individual, feature, ML model, and outcome—and adds adversarial variants, noting that multiple loops can coexist and interact, yielding complex dynamics. It further maps these loops to biases (representation, historical, measurement) and demonstrates distinct bias evolution in a recommender-system case study, illustrating loop-specific equilibria. The work argues for long-term bias mitigation guided by dynamical-systems and control-theory concepts, and suggests integrating tools from distribution shifts, adversarial ML, and optimal transport to design fair, stable decision pipelines with provable guarantees.

Abstract

Prediction-based decision-making systems are becoming increasingly prevalent in various domains. Previous studies have demonstrated that such systems are vulnerable to runaway feedback loops, e.g., when police are repeatedly sent back to the same neighborhoods regardless of the actual rate of criminal activity, which exacerbate existing biases. In practice, the automated decisions have dynamic feedback effects on the system itself that can perpetuate over time, making it difficult for short-sighted design choices to control the system's evolution. While researchers started proposing longer-term solutions to prevent adverse outcomes (such as bias towards certain groups), these interventions largely depend on ad hoc modeling assumptions and a rigorous theoretical understanding of the feedback dynamics in ML-based decision-making systems is currently missing. In this paper, we use the language of dynamical systems theory, a branch of applied mathematics that deals with the analysis of the interconnection of systems with dynamic behaviors, to rigorously classify the different types of feedback loops in the ML-based decision-making pipeline. By reviewing existing scholarly work, we show that this classification covers many examples discussed in the algorithmic fairness community, thereby providing a unifying and principled framework to study feedback loops. By qualitative analysis, and through a simulation example of recommender systems, we show which specific types of ML biases are affected by each type of feedback loop. We find that the existence of feedback loops in the ML-based decision-making pipeline can perpetuate, reinforce, or even reduce ML biases.

Figures (6)

• Figure 1: The ML-based decision-making pipeline as an open-loop system.
• Figure 2: The ML-based decision-making pipeline as a closed-loop system in which different feedback loops can emerge.
• Figure 3: Dynamic effects of different types of feedback loops (FL) acting on an RS pipeline for an online platform. Circles in the box plots denote outliers.
• Figure 4: Open- and closed-loop dynamical systems
• Figure 5: The detailed ML-based decision-making pipeline