Addressing and Resolving Biases in Artificial Intelligence

This paper hopes to determine what are the best strategies through which programmers can reduce bias in artificial intelligence (AI). Alongside this goal, the paper covers fairly introductory content and can be helpful for beginners to start their journey in AI and utilize its powers. To determine the best strategies to mitigate bias, we looked at diverse datasets, hyperparameter optimization, in-processing techniques, and post processing techniques which are all used in the industry. The results that we got were particularly favoring diverse datasets and hyperparameter optimization which means that adjusting weights and the initial data of a model have the biggest impact on accuracy rather than adjusting the final outputs. The best overall strategy was keeping a diverse dataset and allowing the algorithm to independently set its weights based on the best model possible. These factors show that the setup for your model is just as important if not more than actually being fed an output and adjusting the answer through a series of processes like in-processing or post-processing.