You're assuming that the first AIs will be grown via a GA-like process. This is *one* strategy. They may also be reverse engineered from existing intelligences like ourselves. They may be incrementally built over time by computer scientists. They may be constructed with multiple technologies (e.g. growing an organic neural net on an artificial substrate optimized for I/O).
Moreover, human-like AI is only one goal. Swarm intelligence in bees and ants is constrained by the natural environment. Take away those constraints, and swarm intelligence might grow much more complex than anything seen in nature.
As in nature, of course, intelligence and awareness are separable issues. As swarm intelligence or an enhanced GA might be significantly better at solving problems than we are, with nothing resembling human awareness at all.