Artificial intelligence systems are capable of unbelievable things. For example, it is able to predict a person's face just by voice. This system could be very beneficial, but it also has its disadvantages.

This artificial intelligence system, whose study was published in PNASis able to detect a person's age, gender, and origin from their voice. In this way, it gathers elements to create a face as close as possible to the real one.

The question that motivated researchers at the Massachusetts Institute of Technology (MIT) is to understand to what extent you can tell what a person looks like just by how he or she speaks. So they started a project to train an algorithm to generate the physical characteristics of a person from his or her voice.

The result of this project is called Speech2Face and, based on a neural network, is able to develop a virtual face very similar to that of a human being just by listening to a few seconds of audio of his or her voice, highlighting age, gender, and origin.

To be able to do this, the artificial intelligence system went through a training process that consisted of studying the correlations between voice and face of thousands of people through YouTube videos. In this way, the algorithm has secured a panoply of references that allow it to recreate a face without having knowledge of the original.

Although the system does not recreate exact images, the MIT researchers assure that this was not the goal either. In fact, the project aims for the artificial intelligence system to recreate a person's face by retrieving physical characteristics that correlate with their voice.

This kind of artificial intelligence system could be an asset for generating avatars of criminals, for example. However, it can also serve less laudable purposes, such as replicating a person's identity.