Study Guide: Chapter 2 — Paths to Superintelligence

Core Idea

Five roads to superintelligence: AI (via seed AI + recursive self-improvement), whole brain emulation, biological cognitive enhancement (genetic selection), brain-computer interfaces, and networked collective intelligence. Multiple paths raise confidence in reaching the destination, though not all paths lead to equally powerful or equally controllable outcomes.

Key Terms

Seed AI · recursive self-improvement · whole brain emulation (uploading) · iterated embryo selection · brain-computer interface · collective superintelligence

Case Summary

C. elegans (302 neurons) has had its neuron-to-neuron wiring mapped since the 1980s — yet no emulation exists, because connectivity without synaptic strength/dynamics isn't enough. Shows how far even the "easiest" path (no theory needed, just scanning/simulation) still has to go.

Application Checklist

• [ ] When several independent methods could reach a goal, treat failure of one as inconclusive about overall feasibility
• [ ] Distinguish bottlenecks of raw throughput (bandwidth) from bottlenecks of comprehension (meaning-making) — solving one doesn't solve the other
• [ ] Watch for "sudden snap" failure modes where a system stays broken until one last piece clicks, with no visible warning
• [ ] Weigh generational/adoption lag when judging how fast a slow-but-certain technology can actually matter

Self-Test

1. Why does Bostrom think brain-computer interfaces are unlikely to produce superintelligence, even though direct brain-computer connections already exist?
2. What makes the evolutionary argument for AI's feasibility ("evolution did it, so we can") weaker than it first appears?
3. Why might whole brain emulation give more advance warning before success than the AI path — and under what condition would that warning disappear?