In Part 1, we established that superintelligence isn’t a question of “if” but “when.” Now let’s get into the “how.” Nick Bostrom maps out three primary routes to superintelligence, and honestly, they read like different flavors of the same existential dread. Each path has its own technical requirements, timelines, and peculiar nightmares.
Path One: Artificial Intelligence
The classic route. Build a machine that thinks. Simple, right?
The argument here is seductive in its straightforwardness: human brains are physical systems operating according to the laws of physics. There’s no magic juice, no soul-stuff that makes cognition possible. Therefore, cognition can be implemented in other substrates. Silicon, photonics, quantum whatever—the medium doesn’t matter, only the computation.
The timeline argument from Moore’s Law is worth considering, even if extrapolation is a dangerous game. Supercomputer performance has been doubling roughly every 1.5 years for decades. We’re now in the exascale era—machines capable of 10^18 operations per second. That’s getting into the ballpark of estimates for the human brain’s computational capacity.
But raw compute isn’t the bottleneck. The bottleneck is software—specifically, the algorithms that transform computation into intelligence. We still don’t know how to write code that generalizes, that understands context, that exhibits genuine common sense. Every time we think we’ve cracked it, we discover a new edge case that breaks everything.
Path Two: Whole Brain Emulation
This is the copy-paste approach. Instead of figuring out how intelligence works from first principles, just scan a human brain at sufficient resolution and simulate it in software. The philosophy here is brutally pragmatic: evolution already solved the intelligence problem. We don’t need to understand how the brain works to copy it. We just need good enough scanning technology and enough compute.
The technical roadmap reads like science fiction made concrete. First, you need scanning technology that can capture the connectome—the full wiring diagram of the brain—at the level of individual synapses. Then you need simulation infrastructure: models of neurons, models of synapses, models of the chemical signaling that modulates neural activity. Finally, you need enough compute to run the simulation in something approaching real-time.
Bostrom estimates these capabilities might be available around mid-century, with large uncertainty intervals. And here’s an interesting wrinkle: the first emulations might not be perfect. What does it mean to be a slightly broken copy of a human mind? What rights do you have? The philosophical and ethical landmines here are dense enough to qualify as a minefield.
Path Three: Biological Cognition
The road less traveled. Instead of building artificial minds, why not enhance the ones we already have? This path leverages advances in genetics, neuroscience, and reproductive technology to push human intelligence beyond its current ceiling. Unlike the other approaches, this one doesn’t require us to solve the hard problem of consciousness or figure out how to build a mind from scratch.
The numbers here are genuinely startling. Bostrom walks through calculations for embryo selection—choosing which embryos to implant based on genetic predictors of intelligence. Selecting from 10 embryos could yield IQ gains of about 12 points. But the real gains come from iterated embryo selection: creating embryos, deriving stem cells, generating gametes from those stem cells, and repeating the selection process across multiple generations—all in vitro. After ten cycles of selection, you could theoretically produce individuals with IQs 130 points above the current mean. That’s not a smart human. That’s a different species of intelligence.
| Selection Method | IQ Points Gained |
|---|---|
| Select 1 in 10 embryos | ~12 points |
| Select 1 in 100 embryos | ~19 points |
| 10 generations iterated selection | <130 points |
The Intersection Point
Here’s what makes this interesting: these paths aren’t mutually exclusive. Research proceeds on all three fronts simultaneously. AI researchers build better algorithms. Neuroscientists improve our understanding of brain architecture. Geneticists expand our knowledge of the heritability of intelligence. The race isn’t between these approaches—it’s between human preparation and technological capability, regardless of which path gets there first.
In Part 3, we’ll explore what happens after human-level intelligence is achieved: the intelligence explosion, the takeoff dynamics, and the cognitive superpowers that would make a superintelligent system effectively godlike compared to human capabilities.
Previous: Part 1: When the Curve Goes Vertical
