I mean, you're also ignoring the frankly larger issue that we still have no idea how to do this. According to recent research (see
http://slatestarcodex.com/2017/06/08/ssc-journal-club-ai-timelines/), AI researchers are very divided on how soon we can achieve AGI, and this is certainly an AGI-complete problem. The best we can do right now is deep neural networks and LSTM and algorithms like that, which are great, but aren't nearly close enough to AGI to do what you want.
So yeah, nice idea, but not much you can realistically do about it right now unless you have some sort of AGI breakthrough up your sleeve.