Most of the projects at the Media Lab are not as novel as they seem. They just take some existing idea or product and tweak it a little bit to seem more novel. SOME stuff is very novel but really not that much.
But it's all implemented. It wouldn't be very interesting to say that you're going to make an animatronic teddy bear or a robot DJ and then not make it.
I'm not really sure what you're getting at anymore.
Entrepreneurs come up with the big ideas. Computer scientists tweak those ideas and try to come up with faster, more memory efficient, secure, better structured, etc. implementations.
In fact, if a computer scientist is working on a novel app, then his/her work would not be considered important as he/she is not working on the solution to a well established problem.
Not at all. Computer scientists invent new problems all the time.
Also, consider the criteria for having a paper accepted at a prestigious conference. Generally, there must be a non-trivial proof of some sort or strong empirical evaluation. This is a great way to discourage people from submitting novel ideas that are hard to evaluate.
There are lots of interesting ideas that don't work. The value of knowing an interesting idea does work is enormous. Evaluating ideas is the basis of the scientific method.
There are a few people who don't focus on implementation (e.g., the MIT Media Lab), but their work is not taken as seriously as traditional implementation-oriented CS.