A Complete Understanding is No Longer Possible

I remember first encountering tree views in Borland Delphi 2. Took me a good couple of days before I got the hang of them, and how they interacted with image lists, to fully get how the recursive structure held together, how the dependencies interlinked.

Similarly with OpenGL. Took quite some time to get the hang of the sequencing of matrix operations, world and camera transforms, etc. - all just the very basics, before getting into anything interesting at all.

For real learning, 8 hours a day is simply unrealistic. It wasn't until the third or fourth time I tried many things that I really understood them, to the point that I didn't have to follow any recipes anywhere.

You probably couldn't handle the breadth of it, but the depth of computing in the 60s was pretty shallow. You had maybe a dozen levels of (pretty shallow) abstractions, even "higher-level" languages (like Fortran) were pretty uncommon, so you mostly had assembly, an assembler, machine code and the hardware you were running on.

You probably couldn't hold the whole machine in your head, but you could hold a good idea of its cross-section.

Such a complete understanding is eminently possible. It is as easy to understand as any other complex system. You work at a high level of abstraction to understand the breadth of complexity, and then work down into the component level to understand the depth of complexity. Rinse, and repeat for every subsequent level.

The caveat is that it takes time and dedication.

I can't comment on whether it's necessary or not. There might be scenarios where it is. There might be other scenarios where you're better off studying kung fu, because that might be a quicker route to enlightenment.

The problem is not that a single individual cannot master everything but that the reward for learning a part of the book pile becomes lower and lower in proportion to the required effort. This trend is clearly discouraging students to "jump" in some of these fundamental reference books. And teachers find it probably very difficult to provide applied exercises for their theoretical teaching.

"Which do you choose ? How do you know which one will be useful in 5 years ?"

My suggestion is that academia standardize in a virtual machine, like what Knuth started to do with MIX and MMIX. these VM can provide a stable, simpler target to help students and professionnals to enjoy "more complete understanding". I expect this to make learning a lot more fun.

When there is a clear breakthrough in technology (like multicore systems), then a new version of the VM can be specified. This can help to improve communication between academia and the real world : btw MMIX was specified with input from experienced chip archicture designers.

But the sort of understanding I'm referring to, and what I believe the author is referring to, is a comprehensive, infinitely detailed understanding -- the sort of understanding of OS X APIs after internalizing 5000+ pages of reference.

I think this sort of "complete" understanding is not what we mean when we say a expert has an excellent understanding of a subject; instead, when we mean that that expert can work knows what is important.

For example, you could be an expert (and have an excellent grasp of the material) of some programming language, but if someone asked you about the exact name of a rarely used library call, you might not know it off the top of your head. You would know where to look it up, and you would know how to use it, and you would know that that library call exists, but you wouldn't recall the exact name.

I think many people can obtain these expert understandings of complex systems, but that's not what the author is referring to (or if he is, then I think his argument is unsound because no one would approach gaining an expert understanding of OS X by reading 11,000 pages of reference material). The author is talking about a unnecessarily detailed "complete" understanding in which you know everything about the system.