Brain imaging helps us to understand how the brain works, as well as psychiatric and neurological pathologies. The development of non-invasive imaging, notably MRI, and the creation of large databases over the last decade have provided an opportunity to improve our understanding of the brain through imaging. In this context, machine learning has made it possible to create powerful representations of the data, and then, in a second stage, to confront brain imaging with the rich semantics describing an individual's behavior and pathologies. We will present the key stages in these advances, based on our own work.However, artificial intelligence offers a particular opportunity to understand brain function: it provides models of cognition, sometimes inspired by brain function. This is particularly the case in two key areas: vision and language, where AI has scored its greatest successes over the past decade. We will show how comparing the internal representations of AI models with brain activity provides a deeper understanding of brain function, as well as the limits of this understanding.
