Artificial intelligence and deep learning approaches to solve problems in daily life and science are increasingly applied in historical linguistics as well. A recent post in Nature News offered some interesting and new perspectives on the problem of the application of these approaches in science. The first problem is the lack of insights, as the black-box character of machine learning approaches does only give an answer to a question, it does not provide insights on how it arrived at the answer. The second problem is the fact that the algorithms can apparently easily be betrayed, provided one learns the major patterns that trigger their answers. I discuss this in more detail in a recent blog post I wrote for David Morrison's blog The Genealogical World of Phylogenetic Networks, which you can find here.