The goal of this talk is to show that Artificial Intelligence can help psychiatrists to avoid the most tedious and repetitive aspects of their work, thus saving time and energy for the most rewarding aspects of clinical practice. The underlying assumption of the talk is that mental health issues leave physical, machine detectable traces in the behaviour of people, especially when it comes to language (what people say) and paralanguage (how they say it). Therefore, machines can automatically detect such traces and infer from them possible pathologies with the help of AI. Two use cases – detection of depression in adults and identification of insecure attachment in children – will serve as a basis to highlight risks and opportunities in the relationship between AI and psychiatry, from the lack of explainability in state-of-the-art AI methodologies to the possibility of performing large-scale screenings of the population. A discussion on how to bridge the gap between AI and its users (especially in settings relevant to mental health) will conclude the talk.