This excerpt from Stuart Russell's Human Compatible explores the past, present, and future of artificial intelligence. The book examines the potential risks of creating superhuman AI, arguing that a crucial oversight has been the lack of consideration for what happens if AI surpasses human intelligence. Russell proposes a new approach to AI design, focusing on provably beneficial systems that align with human preferences, rather than solely pursuing optimisation. The text also discusses related issues such as job displacement, the ethics of AI development, and the challenges of ensuring human control. Finally, it explores the complexities of human preferences and the need for robust safety mechanisms to prevent unintended consequences.