So, at best we are talking about some model of a human brain at some detail level. IMO, a timescale on it is silly when we don't know how complex the model needs to be to be useful.
Based on current rates of processing growth they can probably model a brain in realtime in 2045.
http://www.telegraph.co.uk/technology/10567942/Supercomputer...
However, while 'the most accurate yet' yet they did not model things like cell death and as ~10,000 neurons die a day it's still an important process even in adult brains. With children it's extremely important as over 1/2 of the brain cells your born with die by 18. Granted on a second by second basis cell death is not that important, but if you want to model a human brain vs say an abstract neural net then accuracy is important.
Sure, other things like cellular energy expenditure, hormones, orientation, gravity, blood flow, oxygen levels, immune response, might not seem important. But, again do you want to model a brain or make a fun simulation.
We can model the processing of a human now but it is exceedingly slow, and it is just toy calculations. So even if we can model a mind in 2045 it will may be a long time after that before it can be done in a meaningful way.
In the meantime we will probably be able to model humans in a more simplistic way. Our external interactions are simple. If we can record all interactions of a person over time we can develop a cognitive profile and develop a 'beta' copy of that person. Not a real thinking AI, but to someone interacting with that copy essentially the real thing.
People are working on this topic now but they're a long way away. I'm guessing 2035.