Abstract

Although deep learning has historically deep roots, with regard to the vast area of​ artificial intelligence and, more specifically, to the study of machine learning and artificial neural networks, it is only recently that this line of investigation has developed fruits with great commercial value, starting to have thus a significant impact on society. It is precisely because of the wide applicability of this technology nowadays that we must be alert, in order to be able to foresee the negative implications of its indiscriminate uses. Of fundamental importance, in this context, are the risks associated with collecting large amounts of data for training neural networks (and for other purposes too), the dilemma of the strong opacity of these systems, and issues related to the misuse of already trained neural networks, as exemplified by the recent proliferation of deepfakes. This text introduces and discusses these issues with a pedagogical bias, thus aiming to make the topic accessible to new researchers interested in this area of​ application of scientific models.