Summarizing Superintelligence: Paths to superintelligence (chapter 2)

  • Posted on: 30 October 2017
  • By: Juho Vaiste

Bostrom presents five different paths to achieving superintelligence level. Having different paths to superintelligence is important, because if one path is blocked, we can still progress in another.

Artificial intelligence
- programming-based approach
- capacity to learn is an integral feature (a child machine, "seed AI")
- "evolutionary process", but enhanced and guided by human engineering (far more efficient than the natural evolution)

Whole brain emulation
- producing superintelligence by scanning and modelin the computational structure of a biological brain
- progress needed: scanning the brain, inputting the scanning data to automated image processing, the neurocomputational structure is implemented on a powerful computer
- no need for fundamental conceptual or theoretical breakthroughs, but some advanced enabling techologies should be developed

Biological cognition
- enhancing the functioning of biological brains
- manipulation of genetics, selective breeding, iterated embryo selection, reproductive cloning
- weak forms of superintelligence could be produced (however, weaker forms of superintelligence could help progressing towards stronger form, even if we were fundamentally unable to create machine intelligence)

Brain-computer interfaces
- computer interfaces are connected to human brains (f.ex. implants)
- the possibility for this kind of connection has been proved, but seems unlikely that problems related to this approach will be solved anytime soon

Networks and organizations
- connecting individual human minds with one another
- an invidual intelligence would not reach the superintelligent level, but the network could ("collective intelligence")



This is a series of blog posts summarizing Nick Bostrom's well-known book Superintelligence. My goal is to present the key points of each chapter in the shortest possible form. More about the writer