ALIGNMENT PROBLEM: How Can Machines Learn Human Values?

Machine learning dominates every aspect of our modern lives as algorithms decide the news we consume, whether we get a mortgage, and which friends wish us happy birthday. But as these systems make ever more decisions on our behalf, how do we ensure that they do what we want and act both safely and fairly? This conundrum is the subject of this urgent and vital book on a subject that is often talked about but little understood. Brian Christian masterly surveys the 'AI fairness' community, introducing us to some of its historical roots in science, philosophy and activism, and crucially many of its quandaries and limitations. From the AI programmes and robots that exploit dangerous loopholes in their own instructions, to facial recognition algorithms that ignore black people, Christian explains how, as machine learning develops, we rapidly approach a collision between artificial intelligence and our most sacred human values. He unpicks the latest studies and features numerous interviews with top researchers. A revealing and helpful guide to an urgent problem in tech on how we can improve our lives rather than complicate or even destroy them. 476pp in large softback.