Machine Learning, or the learning methods used to constitute what is called Artificial Intelligence, is more diversified than is generally presented. Based in particular on the classification proposed by Domingos, the article begins by presenting different approaches to ML. By comparing these formal approaches with the practices reported during Kaggle challenges, we then show that the actual decisions of Machine Learners are primarily dictated by the available data and the scale of complexity of the problems to be addressed. This argument makes it possible to put into perspective the omnipotence attributed to Machine Learning but also, for the social sciences, to specify their possible collaboration, both as a preliminary work of exploring the dimensions of a problem and as a cooperation work with Machine Learners to reduce these dimensions as necessary.