Lifecycle Narrative: Reconstructing the Development Process of a Machine Learning Model

Machine learning models are increasingly proposed for high-risk domains, such as healthcare. However, their growing complexity creates significant obstacles to transparency. This complexity arises from various factors; in this work, we focus on making the organizational development process of a model transparent. A model is often developed over many years and involves multiple version updates, carried out by different individuals, and based on varying subsets of data. To address these issues, we propose a new model reporting framework called a lifecycle narrative, characterized by three dimensions: first, how the purpose of the model has incrementally evolved over time; second, how the data used in the model and the model itself have been developed; and third, what inferences can be made with the model and what the consequences of its use are. We demonstrate these three dimensions using a case study of a high-risk application in healthcare. Our framework complements existing model reporting tools, such as model cards and datasheets, by reconstructing the full lifecycle of the model—a documentation requirement of the EU AI Act. Ultimately, the lifecycle narrative empowers external stakeholders, such as regulators, researchers, citizens, and national competent authorities, to holistically understand a model’s development process.