Given I'm reading Stross's
Neptune's Brood right now, which takes place in a post-humanity universe inhabited by the AIs we left behind, worth posting some of
his perspective.
If you can construct a human-equivalent intelligence in software and switch it on, it's probably going to start by spending a few months emitting white noise and trying to eat its left foot.
Saturn's Children, the first book in this series, stared a sex-bot commissioned shortly before the extinction of humanity in such a method. They basically grew a mind in a rough analogue to human childhood, only grooming (deliberate term that) her to have the traits desired for her role in society. Once they have the fully grown mind they can copy it into as many bodies as they desire. The newer book features an accountant grown in the same way (only with far less sexual abuse of a childlike sentient intelligence) who has then forked off new instances who take decades of specialist training in similar roles.
There are people who have argued we could potentially grow sentient AIs using genetic algorithms, essentially letting them evolve over millions of rapid generations of thousands of prototype procedurally generated intelligences. This seems like it could work. It also raises the terrifying question of when an AI deserves or demands rights. At which points can you wipe the intermediate generations as redundant data and at which point does doing so become genocide?