The human brain is mainly just a giant pattern matching organ (filling in gaps, resolving ambiguities, cross-comparing several senses, etc). Simulating how it works is one thing (that's what they've done). "Installing" existing memories, patterns, and connections is a giant step beyond what they're currently doing. This (and probably all) AIs will have to be trained. The hard part will be bootstrapping the first AI to usefulness, but then the next AI is a copy-paste operation away!

Ray Kurzweil notes that brain scan resolution and scan duration are both following Moore's law. So, within 50 years, it may be possible to take a full-resolution, real-time snapshot of a brain. Ray Kurzweil doesn't explore the possibility of snapshots being taken without consent.

Well if it uses MRI (although miniturised) should be pretty easy to detect/mess with strong magnetic fields around your head (a tin foil hat wouldn't do - but something ferrous may !).

Perhaps we should stop saying tin-foil hat and say "ferrous material hat" ;)