It should start with math and formal verification (top down) rather than with the language (bottom up).
Start here: http://www.cs.utexas.edu/users/EWD/ewd10xx/EWD1036.PDF
Also, trying to beat people doing real work about the ears with ancient Dijkstra quotes stopped being helpful about 30 years ago. He's not the ultimate authority on how a programming language should look in 2011. If you're so sure you've got a good bead on the problem, why don't you go solve it for us then?
Do you honestly expect every PL researcher to all work on the same magical super-language? No- they're going to work on the problem in their area of expertise and create/modify a language to show off their ideas. Then a language like Scala or Haskell or Clojure can come along and improve on them or combine them.
Every field works this way. That's why there is more than one particle collider, more than one drug for each problem, more than one theory of particle physics, more than one programming language.
Your link talks about radical change over gradual change in computer science. It seems to support the idea of creating new and dramatically programming languages. The idea that "small" changes have large effects discourages the gradual modification of existing paradigms and languages.
It should start with a problem and then formulating abstractions to make the problem easier to solve. Or in other words with defining a language and then see what properties it has. Starting with the verifications gives you very well defined properties, but it doesn't solve your problems because you will then optimize your language for the wrong thing.
Mozilla's language [Rust](https://www.github.com/graydon/rust/wiki) is working on static verification (using typestate) and concurrency in a C/C++ level language.
The Haskell community has several people ([Conal Elliot](http://conal.net), [Luke Palmer](http://lukepalmer.wordpress.com)) who are working on new models of functional I/O such as functional reactive programming.
There is a lot of programming language innovation in parallel/concurrent programming that's just hitting mainstream- actor model, asynchronous programming, CUDA/OpenCL, etc.
The way you talk to the machine is an extremely important part of "how to get problems solved." If you can't express something without absurd amounts of overhead, you're not going to solve the problem that way.