Where I come from 16% accuracy does not mean the problem is solved. Yeah, its a big step up from where we were, but the cat detector neural nets by no means solved the object recognition problem.
It was 85% accuracy on imagenet of pascal voc, using just a couple of computers with nnets vs 1000 for that google brain. And both of these were just experiments run by grad students, a proper industrial effort would solve it.
If applications, then they are literally immense. Robots that can perform any manual labor, thus increase in mnaufacturing ability (plus ability to conduct scientific experiments) by an immense amount. Marshall Brain has good articles about what computer vision technology can enable, including the short story Manna. Computer Vision is easily the most disruptive technology since the invention of the steam engine (it multiplies the work capacity of humanity by that much). It's a new gold rush bigger than the internet (robocars and humanoid robots including Baxter from Rethink Robotics are signs of that beginning). Startups should be all over it, web 2.0 and social networking are miniscule in comparison.