Biased Programmers? Or Biased Data? A Field Experiment in Operationalizing AI Ethics

TL;DR

This study investigates the causes of biased AI predictions, emphasizing the role of training data and incentives, and evaluates interventions like improved data and incentives to reduce bias in algorithmic predictions.

Contribution

It provides experimental evidence that biased training data is a primary cause of bias and introduces a novel economic mechanism linking data quality to programmer effort and responsiveness.

Findings

Biased training data is the main cause of biased predictions.

Better training data increases programmer effort and responsiveness.

Demographic characteristics do not significantly affect bias in predictions.

Abstract

Why do biased predictions arise? What interventions can prevent them? We evaluate 8.2 million algorithmic predictions of math performance from $\approx$400 AI engineers, each of whom developed an algorithm under a randomly assigned experimental condition. Our treatment arms modified programmers' incentives, training data, awareness, and/or technical knowledge of AI ethics. We then assess out-of-sample predictions from their algorithms using randomized audit manipulations of algorithm inputs and ground-truth math performance for 20K subjects. We find that biased predictions are mostly caused by biased training data. However, one-third of the benefit of better training data comes through a novel economic mechanism: Engineers exert greater effort and are more responsive to incentives when given better training data. We also assess how performance varies with programmers' demographic characteristics, and their performance on a psychological test of implicit bias (IAT) concerning gender and careers. We find no evidence that female, minority and low-IAT engineers exhibit lower bias or discrimination in their code. However, we do find that prediction errors are correlated within demographic groups, which creates performance improvements through cross-demographic averaging. Finally, we quantify the benefits and tradeoffs of practical managerial or policy interventions such as technical advice, simple reminders, and improved incentives for decreasing algorithmic bias.