Nick Bostrom addresses the topic of machines that think and the implications of superintelligence in his article:
1. Complexity of the Topic: Bostrom expresses the opinion that discussions about machines that think are often oversimplified or tied to science fiction clichés. He highlights the need for a more nuanced understanding of the topic, particularly in light of recent developments in superintelligence.
2. Current State of Machines: Bostrom acknowledges that machines are currently not very proficient at thinking, except in specific narrow domains. He suggests that they may eventually surpass human intelligence in certain areas, but the timeline for this advancement remains uncertain.
3. Superintelligence: Bostrom emphasizes the significance of superintelligence, which could be either the best or worst thing to happen in human history. He points out that the transition from human-level AI to superintelligence may occur relatively quickly.
4. Control Problem: The probability of a positive outcome with superintelligence largely depends on how well we can control it. Bostrom discusses the intrinsic difficulty of the control problem and the need for research in this area, particularly in fields like mathematics, theoretical computer science, and philosophy. He underscores the importance of recruiting top talent to work on addressing the challenges posed by superintelligence.
In summary, Bostrom emphasizes the complexity of the topic, the potential for superintelligence, and the importance of addressing the control problem to ensure a positive outcome for humanity.