PBIAS: A Principled Approach to Data Bias Management in Data Pipelines (2025-2028)
Abstract
This project aims to tackle fundamental problems of bias in data and Artificial Intelligence (AI), proposing the new concept of bias management. Being trained with massive amounts of human generated content, AI may reflect and reinforce human bias and stereotypes and may be used for malicious purposes. Urgent action is needed to support the average person in better understanding if the output of AI systems can be trusted or not. This project builds and evaluates novel methods to track, quantify, and deal with bias rather than to mitigate or remove it. This will empower end-users making informed data-driven decisions and will benefit Australia by accelerating investment in responsible AI and fostering greater social acceptance in AI.
