Google’s new open source language parser, SyntaxNet, seems to be nearing human levels of sentence parsing. Trained human linguists can parse sentences with 97% accuracy, SyntaxNet can parse them with 94% accuracy… basically SyntaxNet is halfway there. If it doubles in efficiency twice, it will surpass humans.
When it surpasses humans, how far will we be from general strong artificial intelligence? That’s hard to say… but if you combined the work being done on Go, language parsing, robotic movement in real space, and other areas it seems like we are very, very close. Possibly less than five years….