I had the good fortune to meet Rodney Brooks in the late 1980s, before Roomba. He had some little toy cars he had fitted with programmable logic to do fairly sophisticated behaviors. It was actually an early inspiration for my own work in programmable logic. I kept up with AI over the years and my opinions tend to track Brooks. Some would say it is pessimistic, but I would call it realistic. In a nutshell, these systems will fail (perhaps a sign of intelligence itself) so you need to be ready for failures. A short, readable one from IEEE.