Algorithmic Bias in the Real Word

Abstract

Algorithms are increasingly shaping and informing human decision-making; in some cases, they are even making decisions that were pre,viously the sole province of humans. At the same time, there is increasing recognition and understanding of the challenges that algo,rithms can create. One notable set of worries focuses on the phenomenon of algorithmic bias: the ways that algorithms can embody, im,plement, maintain, and even create ethically, psychologically, and societally problematic biases. Algorithms have the potential to c,ounteract human (cognitive and emotional) biases, but the presence and/or perception of biases in an algorithm can serve to undermin,e this impact. This project will develop a novel computational framework for representing the mechanisms that generate bias, includi,ng how those biases manifest in statistical data. This framework will be used to develop algorithms that can discover bias-generatin,g mechanisms from statistical data so that appropriate mitigation responsescan be designed and deployed. This framework and algorith,ms will improve the Navys capabilities to produce Equitable AI systems as designated by the Department of Defenses AI Ethical Prin,ciples. This project will then provide a proof of concept test of the algorithms on simulated personnel evaluation data, informed,by background knowledge about the types of biases that can occur in Naval personnel assessment.

Document Details

Document Type
DoD Grant Award
Publication Date
Apr 01, 2022
Source ID
N000142212113

Entities

People

  • David Danks

Organizations

  • Office of Naval Research
  • United States Navy
  • University of California, San Diego

Tags

Fields of Study

  • Computer science

Readers

  • Educational Psychology
  • Neural Network Machine Learning.
  • Systems Analysis and Design

Technology Areas

  • AI & ML
  • AI & ML - DoD AI Strategy