Multiple roads could lead to superintelligence

Bostrom surveys several distinct technical paths that could produce a machine intelligence exceeding humans across virtually all cognitively demanding domains, rather than betting on just one scenario. These include artificial intelligence built through conventional software research, whole brain emulation (digitally simulating an actual biological brain in enough detail to replicate its function), and biological or genetic enhancement of human intelligence directly.

He treats the specific route as less important than the underlying claim: intelligence isn't uniquely tied to biological brains, so there's no principled reason multiple independent paths couldn't eventually reach or surpass human-level general intelligence, even if we can't predict today which route will get there first or when.

This multiplicity matters strategically — it means superintelligence isn't contingent on one fragile research bet succeeding; it means safety work needs to be robust across possible pathways rather than narrowly tailored to whichever approach happens to be fashionable in AI research at any given moment.