Make your data fair: A survey of data preprocessing techniques that address biases in data towards fair AI

Abstract

During the public trials of ChatGPT, it was highlighted that the language model can generate racially discriminatory responses. This issue, however is not new to AI. Several models and networks exhibited sexism, racism and other discriminatory traits in their output. Needless to say, discrimination and biases in AI must be addressed. The urgency of addressing this issue, however, is becoming more evident and pressing with the widespread adoption of AI solutions across different aspects of our lives. This paper is a gentle introduction of Fairness in AI and a survey of existing solutions. The root cause of unfair AI, is the data used to train and test the algorithms. As such, our survey focuses on data preprocessing techniques that address biases and discrimination in the data consumed by AI.