Stuart Russell believes that as AI systems become more intelligent, they should be programmed with self-doubt to maintain human control over the technology. The flexibility and versatility of even a human toddler is still vastly superior to current AI capabilities.