# A Survey on Bias and Fairness in Machine Learning

**Authors:** Ninareh Mehrabi, Fred Morstatter, Nripsuta Saxena, Kristina Lerman, Aram Galstyan  
**arXiv:** [1908.09635](https://arxiv.org/abs/1908.09635)  
**Date:** 2019-08-23  
**Summary page:** https://www.summarizepaper.com/en/paper/1908.09635v1/

## Key points

- Widespread use of AI systems and applications has raised concerns about fairness and bias
- It is crucial to consider these issues while designing and engineering AI systems
- AI systems are often used in sensitive environments where they make important decisions with life-changing consequences
- Discriminatory behavior towards certain groups or populations must be avoided
- Recent work in machine learning, natural language processing, and deep learning has addressed bias and fairness challenges in different subdomains
- Researchers are becoming more aware of biases present in AI applications as they become commercialized
- A survey investigated real-world applications that exhibited biases and identified different sources of biases affecting AI applications
- Researchers created a taxonomy for fairness definitions to mitigate existing biases in AI systems
- The survey examined domains and subdomains within AI to understand unfair outcomes observed in state-of-the-art methods and how researchers have attempted to address them
- There are still many future directions and solutions to explore for mitigating bias in AI systems
- Addressing bias and ensuring fairness can create more equitable technologies that benefit all individuals

## Summary

The widespread use of AI systems and applications in our everyday lives has raised concerns about fairness and bias. It is crucial to consider these issues while designing and engineering AI systems, as they are often used in sensitive environments where they make important decisions that can have life-changing consequences. Discriminatory behavior towards certain groups or populations must be avoided. Recent work in machine learning, natural language processing, and deep learning has addressed the challenges of bias and fairness in different subdomains. As these systems become commercialized, researchers are becoming more aware of the biases that can be present in AI applications and have made efforts to address them. In a survey conducted by Ninareh Mehrabi, Fred Morstatter, Nripsuta Saxena, Kristina Lerman, and Aram Galstyan various real-world applications that have exhibited biases were investigated. The survey also identified different sources of biases that can affect AI applications. To mitigate existing biases in AI systems, the researchers created a taxonomy for fairness definitions as defined by machine learning researchers. Additionally, the survey examined different domains and subdomains within AI to understand how unfair outcomes have been observed in state-of-the-art methods and how researchers have attempted to address them. However, there are still many future directions and solutions that can be explored to further mitigate bias in AI systems. The hope is that this survey will inspire researchers to tackle these issues by building upon existing work in their respective fields. By addressing bias and ensuring fairness in AI systems we can create more equitable technologies that benefit all individuals and avoid perpetuating discriminatory practices.