I'm kind of a believer in Kurzweil / Moore's law stuff that you can predict when it'll turn up by extrapolating hardware progress. That has always had the human level date around 2030.
Yeah. You can estimate roughly what you need for something functionally equivalent to a brain. It comes out around 100 TFLOPS give or take a couple of zeros so we kind of have that now which gives people five years to sort the algorithms if it's going to be 2030.