Working at the convergence between the humanities and computer science, this text aims to outline a general grammar of machine learning and systematically provide an overview of its limits, approaches, biases, errors, fallacies and vulnerabilities. The conventional term Artificial Intelligence is retained although technically speaking, it would be more accurate to call it machine learning or computational statistics, but these terms would not be attractive to companies, universities and the art market. A review is made of the limitations affecting AI as a mathematical and cultural technique, highlighting the role of error in the definition of intelligence in general. Machine learning is described as consisting of three parts: training data ...