ACRIP
Machine Bias
Article

Machine Bias

There’s software used across the country to predict future criminals. And it’s biased against blacks.

Julia Angwin, Jeff Larson, Surya Mattu and Lauren Kirchner
ProPublica, 2016
Read the article

Read or listen offline

Amazon Kindle
自动生成的音频
  • 0.5×
  • 0.75×
  • 1×
  • 1.25×
  • 1.5×
  • 1.75×
1×
自动生成的音频

Editorial Rating

8

Qualities

  • Controversial
  • Analytical
  • Bold

Recommendation

Imagine a computer program that can predict if a criminal will strike again. Now, imagine that program isn’t accurate. ProPublica senior reporters and editors poke holes in one of the dozens of risk assessment programs that determine the fate of defendants across the United States. They found that one popular tool incorrectly flagged black defendants as likely reoffenders twice as often as it did for white defendants, exposing some to longer incarceration times. getAbstract recommends this revealing analysis to anyone enthusiastic or skeptical about the use of algorithms in the American judicial system. 

Summary

Police arrested Brisha Borden and Vernon Prater separately for petty theft. But a computer program said Borden, who’s black, was more likely to reoffend, although Prater, who’s white, had a worse record. While Borden stayed out of trouble, Prater ended up serving time in prison for subsequent burglary and theft. 

Dozens of US courts – and maybe federal prison soon – rely on risk assessments at each stage of the criminal justice process to assign bond and decide sentence length, although the results don’t have predictive value. Merely 20% of more...

About the Authors

Julia Angwin, Jeff Larson and Lauren Kirchner hold senior reporting and editing positions at ProPublica. Surya Mattu is a contributing researcher.

Comment on this summary