The thing is, in general, a "person" might be able to hack together a small program that works for a particular problem most of the time, but it takes a computer expert to create a program that works essentially all the time, with good performance and strong reliability. If someone in the biology department is capable of doing that, it's probably because they're a "double-major", for some value of the term. A normal biology education will leave them unprepared, and trying to hack together a strong understanding of OpenMPI or something is certainly _possible_ but also extremely likely to leave gaping holes in their education. (Even if one has the brains to become both a biology expert and a parallel computing expert, one must question if that's necessarily the best use of one's time.)
CS keeps making it easier to write programs, but the cutting edge will remain CS's domain for a long time, and academia for some crazy reason tends to like the cutting edge. As specialization continues its march onward it will continue to become more and more difficult to be an expert in both $FIELD and computer science.
(NB: Nearly everyone in bioinformatics has to accept some gaping holes in their education -- who has time to become an expert in each of biochemistry, medicine, computer science, and statistics?)
The modern field of CS is too broad for a single department. Perhaps more universities will follow CMU, creating a school of computer science with specialized departments. Until then, expect to see CS undergrads and Master's students involved in informatics/ML joining non-CS Ph.D. programs.
YMMV
I anticipate CS will become a more narrow discipline, just as math has (remember that almost all of theoretical computer science and much of theoretical physics and theoretical economics was once considered part of applied mathematics...). I would not consider that a failur though. Those who wish to focus on the theory will stay with CS and those who wish to focus on the application will drift into the field of its application, again much as math, and much as you described.
I would hardly call that either a failure nor would I think CS would be reduced to teaching intro classes. First, remember that in the parallel of mathematics, people in math heavy disciplines will continue taking math classes throughout their undergrad. Again, there is a real demand for people with math degrees both working in applied mathematics for the industry and because it makes a good basis to branch out to other fields (I am getting a masters in math and work as a DBA...) So, I would hardly call that a failure, just a narrowing of focus, which can be a good thing. Those who want that focus will stay with it with less baggage in the applied side, and those who want the applied focus can target that with less baggage on the theoretical side.
I had a conversation with a friend only yesterday where I proposed exactly this. Some of the things that are now CS but I don't think will be in the future: graphics (as you said), HCI, robotics, and maybe eventually AI. I wouldn't be surprised to see both theory and systems (OS, parallel, etc) remain under the CS banner, though.
One "failure" scenario (to use "failure" loosely, since it's subjective) is, I think, the math one: CS becomes a narrow discipline of theory-of-computation, systems, and PL type theory, sort of like pure math. To get money, it spends its time teaching intro classes to other majors that the professors don't care about: the way math teaches everyone Intro Calc, CS will teach everyone Intro Java. And all the cool applied stuff, in this scenario, happens in other departments, which in 10 years will all also have computers, and many will know how to program. Not that likely, I don't think, but something to think about.