"I believe that once we have made progress on vision (not in the narrow-universe setting) to the point where generic visual scene understanding is effectively solved, there won't be much left that needs to go into the "ethereal" mind which cognitive scientists want to empower machines with..."

This is a great point but it may have implications that won't make the author happy.

If the field of vision computer vision is "nearly as big as" the field of general artificial intelligence, it may not be a good direction from which to approach general artificial intelligence.

The vision problem involves low-level problems of light-physics-etc, mid-level problems of object transformation and distortion and high level problems of cognition. Saying that vision is hard is saying we need to find one algorithm which can fluently mix these levels rather than using several algorithms that just sequentially deal with each level. And such a level-mixing algorithm sure sounds close to general IA.

It fits with argument of people like Jeff Hawking that there's a "Kernel" of general AI that computer vision is just an instance of (see http://books.google.com/books?id=Qg2dmntfxmQC&lpg=PA101&...). But there's nothing there that says computer vision is useful direction to approach that Kernel from - the technical considerations of vision are hard and you haven't shown it provides a particular useful division in a divide and conquer strategy.