A lot of the comments seem to be defending the necessity of the "chaff." That seems to miss the point of the article. The authors aren't criticizing the extra code (much of which IS necessary to make the code functional, readable, and maintainable), they're suggesting that recognizing that only a small subset of the code defines the core functionality can be used in interesting ways. Programmers already take advantage of this in a variety of ways: we have auto-complete in our IDEs, we use web frameworks that write a lot of glue code so we can focus on the problem at hand, and we (sometimes) use newer languages that remove the need for a lot of scaffolding code.
Their application section gives an idea of what they really have in mind: natural language programming for simple tasks, search for common tasks across diverse code bases, and summarizing code functionality using auto-generated "minsets." There are probably a lot of other tasks we could accomplish if we were reliably able to distill a large block of code to its semantic core.