Developing and designing machine learning systems requires multidisciplinary teams working together across the machine learning pipeline. However, information and values of different disciplines can be amplified or diminished depending on their positioning within that pipeline. So we chose to investigate this phenomenon and “plumb” the machine learning pipeline.
We developed a workshop format where the constraints and contextual conditionings surface during the decision-making process in which AI systems are developed. Through a gamified approach, participants can act out a fictional machine learning design scenario for an image classification system and reflect on how values are embedded and ‘lost’ in industry practices.