Machine Learning and Human Unlearning

One of the current problems in “learning machines” is the detection and elimination of unbalanced data. These data have their genesis in an endless number of variables that are related to models, types of algorithms, classification techniques, among others. In all of them, data continues to be the key element. A piece of data that is the input to a process to obtain information, which in turn can be transformed into data for a new process, in a finite or infinite loop, depending on the area of scientific knowledge that is being addressed. Processes that will be streamlined with the diffusion and democratization of the use of quantum computers, but they have and will have their Achilles heel in the data and/or information. Hence the importance of knowing those human factors that unbalances the expected results through learning machines. Our work presents a new heuristic metric born in the academic context called “tentacles-metric” (TENMET). The objective of the TENMET metric is to detect the intensity of the deviation of the data that is progressively and automatically incorporated into the learning machines, based on a heuristic and systematic evaluation of social networks. The analysis of the cases studied has allowed the elaboration of a vademecum that reinforces learning and minimizes anomalous data and unbalanced information in machines learning, from the perspective of the intersection of formal, natural and social sciences.