In this brief note we explore some of the potential for modern “machine learning” methods to aid in the detection, and therefore defense, of collusive arrangements. We argue that where and when the problem can be formed as one of classification or prediction – are the observed prices during this period well-explained by an algorithm which does explain prices in another period? – machine learning algorithms may be useful supplementary tools. But we also argue that the analysis of collusion rarely ends with such questions. When we need to test a hypothesis, the statistical properties of more traditional econometric methods are likely still required.