Meatspace Press, ISBN: 978-1913824020, English, 206 pages, 2021, UK

Books like this should make a difference. There is so much hype surrounding AI and its (im)possible solutions to many of our problems, solutions which often border on madness. The historical delegation of responsibility we give to machines can have destabilising consequences. There is a recurring definition featured throughout this selection of critical essays, namely “snake oil”, an American term that describes a flawed, obscure miracle product designed to solve intractable problems. Applied to AI, it means aiming for something that “AI can’t do or should not do.” The texts here document these dangers primarily through case studies, a catalogue of frightening mistakes that can quickly shape parts of society, from the risks of algorithmic injustice (so-called “cheap AI”) to the failures of automated content moderation on major platforms. Humans have always suffered from the fascination of prediction, and now computers are also increasingly being used as unquestioned oracles due to their inhuman speed of elaboration. Industrial speculation with unreal results, supported by rapid dissemination through social media, encourages this process. To radically stop this development, we should question the predictive models at their root: “What if something is not predictable? What if randomness or unpredictability or pure good luck is the main factor and there is no possibility to elaborate a model?”