Human Compatible

In this pioneering book, AI expert Stuart Russell argues that this scenario can be avoided, but only if we rethink AI from the ground up. He starts by examining intelligence, both human and machine, and discusses the near-term advantages AI can bring, such as smarter personal assistants and accelerated scientific progress. He also highlights the breakthroughs needed before superhuman AI becomes a reality, along with the potential dangers, like autonomous weapons and viral manipulation, that we already face.

If superhuman AI does emerge, it would be far more powerful than humans. Russell suggests that we can prevent AI from gaining control by redesigning it to be inherently uncertain about human preferences. This new approach would make machines humble, altruistic, and dedicated to fulfilling our goals, not their own, ensuring they are both safe and beneficial.