The Alignment Problem
Brian Christian
Recommendation inspired by:
How to get better at video games, according to babies - Brian Christian
There’s an apocalyptic thought experiment by Nick Bostrom where a company creates an artificial intelligence whose job is to make as many paperclips as possible. They forget to tell it to value human life though, so eventually, when human culture stands in the way of paperclip production, it eradicates humanity and converts the entire planet into paperclips. That’s an extreme and (hopefully) unlikely scenario, but it turns that throughout the history of artificial intelligence and machine learning, it’s been incredibly difficult to get the machines to do exactly what we want them to. For instance, we’ve inadvertently trained them to carry human bias into algorithmic parole decisions and facial recognition software. And as the algorithms become more complex and more empowered, we’re bumping up into scenarios where human safety and well-being are on the line. The Alignment Problem explores these ideas in fluid, fascinating, and accessible manner. You don’t need any background in programming to read it, as the book is full of real-world stories and anecdotes that make the concepts graspable and immediate. You’ll also learn a great deal about people and how we learn and think along the way, because the fields of human and machine learning are on a seesaw of discovery; a breakthrough in one often offers revelations in the other.
Recommended by:
Alex Rosenthal, TED-Ed Team
Want to download an audio version of this book for free?
Visit www.audible.com/ted-ed and start a free trial. Every free trial encourages Audible to continue supporting TED-Ed's nonprofit mission. Learn more.