# Fairness And Bias in Artificial Intelligence: A Brief Survey of Sources, Impacts, And Mitigation Strategies

**Authors:** Emilio Ferrara  
**arXiv:** [2304.07683](https://arxiv.org/abs/2304.07683)  
**Date:** 2023-04-16  
**Summary page:** https://www.summarizepaper.com/en/paper/2304.07683v1/

## Key points

- AI has made significant advancements in healthcare decision-making, medical diagnosis, and other domains.
- Concerns about the fairness and bias of AI systems have been raised, particularly in areas like healthcare, employment, criminal justice, and credit scoring.
- Biased AI systems can lead to unfair outcomes and perpetuate existing inequalities.
- The paper offers a comprehensive overview of fairness and bias in AI by reviewing sources of bias such as data, algorithm, and human decision biases.
- The societal impact of biased AI systems is assessed with a focus on the perpetuation of inequalities and the reinforcement of harmful stereotypes.
- The paper explores various proposed mitigation strategies for addressing AI bias while discussing their ethical considerations for implementation.
- Interdisciplinary collaboration is emphasized to ensure effectiveness in mitigating AI bias.
- Current approaches to mitigate AI bias are also provided including data pre-processing, model selection, and post-processing.
- Addressing bias in AI requires a holistic approach that involves diverse and representative datasets along with enhanced transparency and accountability in AI systems.
- Alternative AI paradigms that prioritize fairness and ethical considerations should also be explored.
- The importance of increasing diversity in training data collection methods to address biases at every stage of the AI pipeline is emphasized.
- Developing frameworks for responsible use of AI that include transparent documentation of training data and modeling decisions is essential along with diversifying teams involved in developing and evaluating AI systems to ensure fairness and justice.
- Future research should aim to establish ethical and legal frameworks governing AI systems while addressing privacy, transparency, accountability concerns as well as addressing biases in federated healthcare systems.

## Summary

Artificial Intelligence (AI) has made significant advancements in healthcare decision-making, medical diagnosis, and other domains. However, concerns about the fairness and bias of AI systems have also been raised, particularly in areas like healthcare, employment, criminal justice, and credit scoring. Biased AI systems can lead to unfair outcomes and perpetuate existing inequalities. This paper offers a comprehensive overview of fairness and bias in AI by reviewing sources of bias such as data, algorithm, and human decision biases. The societal impact of biased AI systems is assessed with a focus on the perpetuation of inequalities and the reinforcement of harmful stereotypes. The paper explores various proposed mitigation strategies for addressing AI bias while discussing their ethical considerations for implementation. Interdisciplinary collaboration is emphasized to ensure effectiveness in mitigating AI bias. Through a systematic literature review spanning multiple academic disciplines, definitions of AI bias and its different types are presented along with discussions on the negative impacts of AI bias on individuals and society. Current approaches to mitigate AI bias are also provided including data pre-processing, model selection, and post-processing. Addressing bias in AI requires a holistic approach that involves diverse and representative datasets along with enhanced transparency and accountability in AI systems. Alternative AI paradigms that prioritize fairness and ethical considerations should also be explored. The paper concludes by emphasizing the importance of increasing diversity in training data collection methods to address biases at every stage of the AI pipeline. Developing frameworks for responsible use of AI that include transparent documentation of training data and modeling decisions is essential along with diversifying teams involved in developing and evaluating AI systems to ensure fairness and justice. Future research should aim to establish ethical and legal frameworks governing AI systems while addressing privacy, transparency, accountability concerns as well as addressing biases in federated healthcare systems.