I really enjoy this paper/argument, it's fairly simple and produces an extraordinary conclusion. For anyone else who enjoys this work, he (Nick Bostrom) also released a book called SuperIntelligence regarding predictions/outcomes for various paths of AI.