Big Data Vs for Detecting and Preventing Biases in AI Algorithms: A Systematic Literature Review

The negative effects of AI algorithm biases on society have become a major concern recently. Many proposed solutions to address these biases have often failed. Our study responds to calls for using big data characteristics in designing training data to detect and prevent AI biases early. A systematic literature review examined the extent and details of algorithm discrimination, drawing on data from 8 databases and social media research sources. Results showed potential solutions for reducing discrimination bias based on big data characteristics such as volume, variety, velocity, veracity, variability, and value. Specifically, a larger volume of big data increases the likelihood of preventing AI bias. The diversity of big data allows for a wider range of training data, drawn from various formats and sources, helping to combat biases. The velocity of big data supports ongoing training, which should help eliminate bias. Veracity ensures the accuracy and truthfulness of training data, aiding early detection of biases. Variability allows for the use of changing, volatile data over time, helping to deter bias. Lastly, the high quality, or value, of big data promotes the use of superior training data to eliminate biases.