Who Decides When Artificial Intelligence Works? Algorithmic Production as a Continuous Evaluation Process

The intensifying deployment of AI systems in different spheres of our lives has raised concerns over their evaluation. The latter has often been discussed as a unified process that any model should undergo before its release and use. However, we argue that throughout their production AI systems are continuously assessed. A closer look at their supply chains reveal modes and practices of evaluation at each stage of their development. Based on empirical research of three sites of AI development, this article provides evidence for this continuous assessment. It points to the limits of common perspectives on AI evaluation and argues for a more grounded and comprehensive understanding of the ways in which AI systems can be and are assessed.