When Machines Have Ideas, 2016
“When we build stories into machine intelligence systems, we should enable them to attach probabilities to their stories. Weighing alternatives stories is important, because it builds solid foundations. If the system weighs stories against each other and sometimes returns to earlier stories, readjusting them or reconsidering its beliefs, then the result is more stable and less likely to break and the systems will behave more like enlightened scientists — always balancing ideas.
If we build them this way, then when they do ultimately speak their minds they will do so with civility and always tell us how uncertain they are about the idea that they are conveying to us — making machines that can themselves have ideas worth spreading.”