We propose to study strategies to control chemical reaction networks, in order to use them to perform computations, with certain similarities to the computations by biological or artificial neural networks. In these systems, certain properties can be emergent when they arise from the interactions of a large number of components. We are particularly interested in the ability to perform some form of computation. Computation should be understood here as the ability of the chemical network to dynamically reach a certain final composition given an initial composition as illustrated in the figure. We are interested in robust computations in the sense small perturbations in the kinetics of chemical reactions should not affect the final composition.