Thanks to a link from Kaplan University Group's President, Greg Marino, I recently saw Peter Norvig's very nice TED talk about about the Stanford AI course he and Sebastian Thrun put together. It gives telling insights into how these two faculty members thought about their instructional design.
It reminded me yet again of a missed opportunity: while there are new things happening that are bound to help students, there is very little grounding in what the evidence about learning suggests will work. Some assertions about what is "bound to work" are not going to work (at least not as broadly as hoped), and other things with strong evidence aren't even on the conventional wisdom radar screen (if such a new field can be said to have conventional wisdom!).
What's needed is a "learning engineering" approach - rich understanding of what the evidence about learning says, used to drive practical solutions for learning at scale, just as other engineering disciplines do with related sciences and their own real-world, at-scale challenges.
There's more to do that's just waiting to be done.