Recent advances in Artificial Intelligence (AI) are linked to the unquestionable success of neural networks. But can this technology be applied to the ultimate challenge of quantitative finance, predicting future prices? In this paper we explain briefly how such systems could work. Then we present a simple experiment where we turn a neural network into a predictor of price returns, by using the significant amount of data available on the markets at high-frequency scales. We also explore whether the results depend on the complexity of the network (its ‘depth’) in this particular case.