The Measure and Mismeasure of Fairness with Sharad Goel - #363

Topics covered
Popular Clips
Episode Highlights
Policing Practices
explores the intersection of machine learning and policing practices, particularly focusing on stop-and-frisk policies. He highlights a significant finding where 40% of individuals stopped had less than a 1% chance of possessing a weapon, questioning the constitutional threshold for reasonable suspicion 1. This analysis reveals potential biases in police searches, showing that black and Hispanic individuals face lower thresholds for searches compared to white individuals 2.
We estimated that something like 40% of people were being stopped with less than 1% chance of having a weapon on them.
---
By focusing on discretionary searches, Goel aims to refine the understanding of bias in policing, separating them from procedural searches that aren't based on specific suspicions 3.
Open Policing Project
The Stanford Open Policing Project, spearheaded by , aggregates data from over 200 million traffic stops to scrutinize police interactions across the U.S. 4. This extensive dataset, publicly available, supports various stakeholders in auditing local policing practices and has already influenced policy changes, such as the LAPD's reduction in stop practices following an analysis by the Los Angeles Times 5.
We've actually released over 200 million traffic stops from across the country.
---
The project underscores the challenges of data collection and standardization, requiring significant resources to compile and clean disparate data formats from numerous jurisdictions 6.
Threshold Test
discusses the development of the threshold test, a tool designed to detect discriminatory practices in police searches. This test addresses limitations of the traditional hit rate analysis, which sometimes misleadingly suggests discrimination against white drivers 7. By focusing on the success rate of searches rather than just their frequency, the threshold test offers a more nuanced understanding of discrimination in policing 8.
Let's not look at the rate at which we're searching people. Let's look at the rate at which those actions are successful.
---
This approach helps to infer the standard of evidence used in searches, aiming to provide a clearer picture of bias in law enforcement practices.
Related Episodes


Causal Conceptions of Fairness and their Consequences with Sharad Goel - #586
Answers 383 questions

Approaches to Fairness in Machine Learning with Richard Zemel - TWiML Talk #209
Answers 383 questions

Privacy vs Fairness in Computer Vision with Alice Xiang - 637
Answers 383 questions

Trends in Fairness and AI Ethics with Timnit Gebru - #336
Answers 383 questions

“Fairwashing” and the Folly of ML Solutionism with Zachary Lipton - TWIML Talk #285
Answers 383 questions

Algorithmic Injustices and Relational Ethics with Abeba Birhane - #348
Answers 383 questions

Beyond Accuracy: Behavioral Testing of NLP Models with Sameer Singh - #406
Answers 383 questions

Understanding AI’s Impact on Social Disparities with Vinodkumar Prabhakaran - 617
Answers 383 questions

Assessing the Risks of Open AI Models with Sayash Kapoor - 675
Answers 383 questions

AI’s Legal and Ethical Implications with Sandra Wachter - 521
Answers 383 questions

Trust and AI with Parinaz Sobhani - TWiML Talk #208
Answers 383 questions

The Fallacy of "Ground Truth" with Shayan Mohanty - #576
Answers 383 questions

Fairness and Robustness in Federated Learning with Virginia Smith - #504
Answers 383 questions














