Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Bridging Prediction and Intervention Problems in Social Systems

Lydia T. Liu, Inioluwa Deborah Raji, Angela Zhou, Luke Guerdan, Jessica Hullman, Daniel Malinsky, Bryan Wilder, Simone Zhang, Hammaad Adam, Amanda Coston, Ben Laufer, Ezinne Nwankwo, Michael Zanger-Tishler, Eli Ben-Michael, Solon Barocas, Avi Feller, Marissa Gerchick, Talia Gillis, Shion Guha, Daniel Ho, Lily Hu, Kosuke Imai, Sayash Kapoor, Joshua Loftus, Razieh Nabi, Arvind Narayanan, Ben Recht, Juan Carlos Perdomo, Matthew Salganik, Mark Sendak, Alexander Tolbert, Berk Ustun, Suresh Venkatasubramanian, Angelina Wang, Ashia Wilson

TL;DR

This paper reframes automated decision systems (ADS) from purely predictive tools to components of broader policy interventions within social systems. It develops an end-to-end framework that links predictions to decisions and outcomes (X→R→D→Y) and integrates causal, observational, and experimental methods to evaluate ADS in deployment. By emphasizing the institutional, organizational, and societal context, it highlights limitations of isolated predictive accuracy and advocates for intervention-aware design, evaluation, and implementation science. The work showcases how problem formulation, decision-theoretic reasoning, and rigorous interventional evaluation can guide responsible deployment of ADS across domains such as criminal justice, healthcare, housing, and social services. It also outlines prescriptive paths toward better governance, engineering, and stakeholder-inclusive deployment to maximize equitable social impact.

Abstract

Many automated decision systems (ADS) are designed to solve prediction problems -- where the goal is to learn patterns from a sample of the population and apply them to individuals from the same population. In reality, these prediction systems operationalize holistic policy interventions in deployment. Once deployed, ADS can shape impacted population outcomes through an effective policy change in how decision-makers operate, while also being defined by past and present interactions between stakeholders and the limitations of existing organizational, as well as societal, infrastructure and context. In this work, we consider the ways in which we must shift from a prediction-focused paradigm to an interventionist paradigm when considering the impact of ADS within social systems. We argue this requires a new default problem setup for ADS beyond prediction, to instead consider predictions as decision support, final decisions, and outcomes. We highlight how this perspective unifies modern statistical frameworks and other tools to study the design, implementation, and evaluation of ADS systems, and point to the research directions necessary to operationalize this paradigm shift. Using these tools, we characterize the limitations of focusing on isolated prediction tasks, and lay the foundation for a more intervention-oriented approach to developing and deploying ADS.

Bridging Prediction and Intervention Problems in Social Systems

TL;DR

This paper reframes automated decision systems (ADS) from purely predictive tools to components of broader policy interventions within social systems. It develops an end-to-end framework that links predictions to decisions and outcomes (X→R→D→Y) and integrates causal, observational, and experimental methods to evaluate ADS in deployment. By emphasizing the institutional, organizational, and societal context, it highlights limitations of isolated predictive accuracy and advocates for intervention-aware design, evaluation, and implementation science. The work showcases how problem formulation, decision-theoretic reasoning, and rigorous interventional evaluation can guide responsible deployment of ADS across domains such as criminal justice, healthcare, housing, and social services. It also outlines prescriptive paths toward better governance, engineering, and stakeholder-inclusive deployment to maximize equitable social impact.

Abstract

Many automated decision systems (ADS) are designed to solve prediction problems -- where the goal is to learn patterns from a sample of the population and apply them to individuals from the same population. In reality, these prediction systems operationalize holistic policy interventions in deployment. Once deployed, ADS can shape impacted population outcomes through an effective policy change in how decision-makers operate, while also being defined by past and present interactions between stakeholders and the limitations of existing organizational, as well as societal, infrastructure and context. In this work, we consider the ways in which we must shift from a prediction-focused paradigm to an interventionist paradigm when considering the impact of ADS within social systems. We argue this requires a new default problem setup for ADS beyond prediction, to instead consider predictions as decision support, final decisions, and outcomes. We highlight how this perspective unifies modern statistical frameworks and other tools to study the design, implementation, and evaluation of ADS systems, and point to the research directions necessary to operationalize this paradigm shift. Using these tools, we characterize the limitations of focusing on isolated prediction tasks, and lay the foundation for a more intervention-oriented approach to developing and deploying ADS.

Paper Structure

This paper contains 85 sections, 10 equations, 14 figures.

Table of Contents

  1. Introduction
  2. Scope
  3. Paper Organization
  4. Background
  5. Other surveys of ADS
  6. Details of the Institutional Context
  7. Goals, Resources, Activities
  8. Policy Context
  9. Bureaucratic Counterfactuals
  10. Institutional context as ADS deployment context.
  11. Model Design
  12. Key Takeaways
  13. Setup: Predictions and Interventions
  14. Interventional framework
  15. Predictions to decisions
  16. ...and 70 more sections

Figures (14)

  • Figure 1: Conceptual diagram of the ADS universe. ADS (right hand side) are deployed in institutional contexts (left-hand side). The blue "Prediction as Intervention" box zooms into the ADS and highlights the role of historical data on decisions and outcomes (orange database), developing new predictive models to inform future decisions mediated by human decision-makers, all in order to improve target outcomes. While most research focuses on the "prediction as intervention box", institutional context is key for understanding ADS as one alternative among others.
  • Figure 2: Program logic models (illustrative example)
  • Figure 3: Estimated treatment effects of supervised release by race and FTA risk. (a) Average treatment effect (with 90% confidence intervals) for different groups, stratified by FTA risk. (b) Distribution of heterogeneous treatment effects by group, stratified by FTA risk.
  • Figure 4: Figure from black-smith-berger-noel-03-UI-rdd. Percentiles of baseline risk $\mathbb{E}[Y(0)\mid X]$ on the x-axis; estimated heterogeneous treatment effects on the y-axis. A nonlinear relationship indicates prioritization on baseline risk alone is suboptimal.
  • Figure 5: Our interventional evaluation paradigm vs. the status quo of "blind men studying the elephant", highlighted. Blue: $D\to Y$ (ITR evaluation). Green: policy-level impacts of introducing an ADS (event studies). Red: $R\to D \to Y$ with micro-level data (human--algorithm interaction and encouragement designs).
  • ...and 9 more figures

Theorems & Definitions (10)

  • Example 1: Risk assessment for pretrial decision-making
  • Example 2: Medical cost prediction
  • Example 4: Supervised release in criminal justice
  • Example 5: Triaging social services, unemployment benefits
  • Example 6: Social prediction, Fragile Families Challenge
  • Example 7: Criminal justice - Further Policy Context for \ref{['ex:justice']}
  • Example 8: Wisconsin Public Schools Dropout Prediction
  • Example 9: Health - Liver transplantation
  • Example 10: Health - diabetes prediction; how to choose predictors
  • Example 11: Pneumonia/asthma "spurious correlation"