The past few years have witnessed tremendous progress in the development of AI tools. Physicists are among those who have started to benefit from AI, for instance, in the classification of spectra or the detection of particles. But how far can the use of AI be pushed in physics? Skeptics argue that big data and its AI-based analysis fail to provide the kind of understanding that physicists value. Optimists point to examples in which the use of AI has apparently yielded progress in understanding. This talk aims to answer the question of whether, and to what extent, AI tools can produce understanding. I focus on explanatory understanding and artificial neural networks (ANNs). To answer my question, I confront recent philosophical accounts of understanding with ANNs. It turns out that ANNs make achievements that were traditionally based on physical understanding. Whether they do so because they have become sensitive to explanatorily relevant factors is often not known. This is why ANNs frequently do not provide understanding to physicists, as skeptics claim. However, there are examples in which ANNs can help to discover factors that matter to explanation. The understanding that ANNs can provide thus crucially depends on how they are used.