Exposing the many biases in machine learning

Q2 Economics, Econometrics and Finance

Business Information Review Pub Date : 2022-08-15 DOI:10.1177/02663821221121024

Sharon Richardson

引用次数: 4

Abstract

In recent years, there have been numerous articles highlighting issues with bias in machine learning algorithms underpinning the use of AI in decision making. Specifically, algorithms trained on historical real-world observations. However, less is written about the many ways bias can be introduced into the machine learning process. This article outlines 12 different types of bias that can occur during the data science process, from capture through curation to analysis and application.

查看原文本刊更多论文

暴露机器学习中的许多偏见

近年来，有许多文章强调了支持人工智能在决策中使用的机器学习算法中的偏见问题。具体来说，是根据真实世界的历史观测训练的算法。然而，关于在机器学习过程中引入偏见的多种方式，却鲜有报道。本文概述了数据科学过程中可能发生的12种不同类型的偏见，从捕获到管理再到分析和应用。

本文章由计算机程序翻译，如有差异，请以英文原文为准。

求助全文

约1分钟内获得全文求助全文

来源期刊

Business Information Review Economics, Econometrics and Finance-Economics, Econometrics and Finance (miscellaneous)

CiteScore

2.50

自引率

0.00%

发文量

期刊介绍： Business Information Review (BIR) is concerned with information and knowledge management within organisations. To be successful organisations need to gain maximum value from exploiting relevant information and knowledge. BIR deals with information strategies and operational good practice across the range of activities required to deliver this information dividend. The journal aims to highlight developments in the economic, social and technological landscapes that will impact the way organisations operate. BIR also provides insights into the factors that contribute to individual professional success.