The Alignment Problem, Machine Learning and Human Values

Not many can claim to be a computer programmer, nonfiction author and poet, but Brian Christian has established himself as all three. Christian has just released his newest book, The Alignment Problem, which delves into the disparity that occurs when AI models don’t do exactly what they’re intended to do. The book follows on the success of Christian’s previous work, "The Most Human Human and Algorithms to Live By." Now a visiting scholar at UC Berkeley, Christian joined AI Podcast host Noah Kravitz to talk about the alignment problem and some new techniques being used to address the issue. The alignment problem can be caused by a range of reasons — such as data bias, or datasets used incorrectly and out of context. As AI takes on a variety of tasks, from medical diagnostics to parole sentencing decisions, machine learning researchers are expressing concern over the problem. Listen to the full podcast to hear about this and more — including Christian’s book club experience with Elon Musk and why he chose to double major in philosophy and computer science.