<p>There is growing enthusiasm in some quarters for purging representation from the set of concepts necessary to understand intelligence, at least in some circumstances. But to date, the list of explicit alternatives remains thin, with few plausible examples of intelligent systems the operation of which can be clearly articulated in non-representational terms. I argue that the class of Reservoir Computer (RC) models from machine learning constitutes a rich source of such examples. While RCs can learn to forecast even highly complex dynamical systems, I argue that they cannot be said to represent those systems in any but the most trivial way. Specifically, RCs do not contain compositional representations from which may be extracted subrepresentations that correspond to portions or aspects of the targets they represent. In other words, RCs do not represent parts or properties of their targets and so do not achieve intelligent prediction by manipulating component subrepresentations. They are thus concrete examples of intelligence achieved without essential use of representation.</p>