Daniel Oberhaus opens the book with the tragic loss of his sister who died by suicide, leaving a note with the hopeless phrase ‘NO ONE KNOWS WHAT THIS IS’ alongside a list of psychiatric diagnoses. This becomes the foundation of Oberhaus’s core argument: that psychiatry lacks reliable diagnostic criteria in the absence of direct biological markers, and that this is an inadequate basis for large-scale AI automation of mental healthcare, what he terms Psychiatric AI, or ‘PAI’.

Oberhaus traces the development of the profession of psychiatry alongside the rise of AI, to set the scene for his critique. The book focuses on how PAI systems have developed to detect mental disorders by mining the ‘digital exhaust’ of our online activities as proxies of our mental health. Oberhaus warns that using these proxy markers for mental-health needs risks magnifying existing weaknesses in psychiatric diagnoses and a drift into ‘swipe psychiatry’, where clinical assessment is sidelined in favour of AI outputs based on the reductive data gathered about us.

Oberhaus does acknowledge that as AI models become more advanced the issues with diagnostic inaccuracy will decrease, but he’s also concerned that the pursuit of ever richer data-sets to feed PAI will usher in the total surveillance of our human experience, creating a ‘silicon shrink’ that will covertly shape our choices and experiences. His arguments do feel like scaremongering at times, but he offers some recent examples of companies that have quietly used sensitive mental health data without adequate consent, and highlights that companies are reluctant to open their data-sets to scrutiny because it would reduce their competitive advantage.

It’s not all foreboding pessimism. Oberhaus repeatedly stresses the potential of PAI to improve patients’ lives, provided there is strong ethical leadership and rigorous information governance and cybersecurity standards. He emphasises the importance of continuous monitoring to avoid AI systems introducing bias and injustice, or being misused to plug gaps in healthcare access. Oberhaus also leads you to imagine a future in which over-reliance on AI could reshape clinical practice, leaving professionals feeling that they need to justify decisions that conflict with opaque AI outputs, eroding the nuance of trying to understand the person in front of you.

PAI as Oberhaus describes it is edging towards reality, and we aren’t ready for it. He makes a strong argument for caution, but I wanted a deeper analysis of the ethical tensions that psychiatry will have to deal with. And, inevitably, the book is also showing its age only a year after its publication, as new hazards appear, including the normalisation of AI as companion or analyst.