The other concern is that even if the simulation is good enough, there's still an ick factor to "tricking" a baby into forming an emotional connection with a machine that doesn't have real human feelings.
If we build caretaker robots with software that closely models the internal function of human brains, to the point that we're satisfied that their outward behavior is accompanied by complex and human-like feelings of caring, that problem goes away too. Whether that's even possible is more of a philosophy of mind question; depends on whether you think internal experience depends on the exact nature of the physical substrate.
(Of course if it is possible, there's yet another layer there with the ethical issues of designing a slave race of beings with real feelings, see e.g. Robin Hanson's ideas about "ems.")