Human Compatible

Stuart Russell sets out a different way to think about artificial intelligence and the future it may create. He argues that the dangers of increasingly capable machines can be addressed only by redesigning AI from first principles, with systems built to be uncertain about the human preferences they are meant to satisfy.

The result is a clear, forward-looking account of where AI stands now, what breakthroughs still lie ahead, and why misuse already matters. It will appeal to readers who want a serious but accessible view of the risks and possibilities around superhuman AI, and to anyone looking for a convincing case that powerful machines can be made beneficial rather than threatening.