Machine Learning (ML), a subset of Artificial Intelligence (AI) has been in a pattern of rapid growth over the last decade, simultaneously evolving through the intersection of the needs of businesses and individuals, together with the combined, exponential increase of computer power, data availability, and network infrastructure.

The rise of ML products and services has led to advances in vital sectors including healthcare, finance, automotive, security, and more. These include expediting enhanced diagnosis in patients, strengthening cybersecurity measures, manufacturing automation, or leading to new technologies like self-driving vehicles, robotics, digital assistants, and so-called ‘chatbots’. However, the rise in the development of AI-enabled products and services has not been all positive. In parallel, there have been numerous documented instances of harmful impacts on individuals, communities, and the broader society.

This project focuses on understanding and mitigating negative, unforeseen, and even unconscious consequences of AI/ML by interrogating the presence of bias in the Machine Learning Operations (MLOps) process. Our approach is to better identify and address vulnerabilities at specific phases in the development of an ML product or service. Using strategic foresight methods, this project explores emerging AI trends and develops an array of possible future scenarios, through which bias and other areas of concern are studied to better understand their potential impacts.

As a product of this investigation, we develop an Empathetic Design Framework (EDF), employing a set of lenses and a toolkit that can be effortlessly incorporated into an ML cross-functional team’s agile practice in a bid to better identify ML risks and weaknesses, and reduce the occurrence of negative future scenarios.

Finally, this research aims to identify appropriate and impactful insertion points within the MLOps process for utilizing the EDF to mitigate negative potential biases during the ML life cycle.