Will AI Go Bad?: The Wrong Question

In this insightful book, Boris Kriger explores the fundamental and often overlooked risks of AI development, revealing that goal drift is an inevitable and structural aspect of self-improving systems. He argues that human-designed objectives are inherently biased by biological factors, making them unreliable benchmarks for AI safety. Drawing on formal theories and empirical research, Kriger highlights that the most significant dangers may stem from subtle, cumulative distortions rather than dramatic failures. Despite these concerns, the book offers a cautiously optimistic perspective on the future of AI.