Superintelligence: Paths, Dangers, Strategies · chapter 1

Past Developments and Present Capabilities

Bostrom’s historical chapter holds up well 12 years later. The AI milestones he cites (Deep Blue, Watson) are factually correct. His AI winter narrative matches standard historiography. Most strikingly, the 2014 Mueller-Bostrom survey prediction (50% HLMI by 2040-2050) has been validated by the much larger 2024 Grace et al. survey (50% by 2047) and compressed by prediction markets (Metaculus median ~2028-2033). His hardware extrapolations were directionally right but via a different mechanism than he anticipated. This is the strongest chapter in the book – mostly factual reporting rather than theoretical speculation.

So what

Chapter 1 establishes the historical groundwork for Bostrom’s thesis. His factual claims about AI history are accurate. His timeline predictions, based on the Mueller-Bostrom survey, have proven remarkably consistent with subsequent larger surveys (Grace et al. 2024). The hardware extrapolations were directionally correct but understated the actual mechanism of compute growth (GPU parallelism rather than Moore’s Law).

Verdict

MostlyAccurate

Claims checked in this chapter (4)

needs context⚠ re-audit pending
Expert surveys suggest a 50% probability of human-level machine intelligence by 2040-2050, based on the Mueller-Bostrom 2014 survey of AI researchers.
holds⚠ re-audit pending
AI has progressed from early chess programs through Deep Blue defeating Kasparov in 1997 to Watson winning Jeopardy in 2011, demonstrating steady capability growth.
holds⚠ re-audit pending
AI development has followed a pattern of hype cycles ('AI summers') followed by periods of reduced funding and disillusionment ('AI winters'), particularly in the 1970s and late 1980s.
needs context⚠ re-audit pending
Extrapolations of hardware trends (Moore's Law) suggest computational power sufficient for human-brain-equivalent processing could be available within decades.