This paper presents research on how knowledge brokers attempt to translate opaque algorithmic predictions. The research is based on a 31-month ethnographic study of the implementation of a learning algorithm by the Dutch police to predict the occurrence of crime incidents and offers one of the first empirical accounts of algorithmic brokers. We studied a group of intelligence officers, who were tasked with brokering between a machine learning community and a user community by translating the outcomes of the learning algorithm to police management. We found that, as knowledge brokers, they performed different translation practices over time and enacted increasingly influential brokerage roles, namely, those of messenger, interpreter, and curator. Triggered by an impassable knowledge boundary yielded by the black-boxed machine learning, the brokers eventually acted like “kings in the land of the blind” and substituted the algorithmic predictions with their own judgments. By emphasizing the dynamic and influential nature of algorithmic brokerage work, we contribute to the literature on knowledge brokerage and translation in the age of learning algorithms.