computer science vs computer practice

Computer Science vs. Computer Practice

So you're an accomplished computer scientist are ya? We'll see about

that.

The difference between Computer Science and Computer Practice

What is a magician but a practicing theorist? -- Obi-Wan Kenobi

As we all know, there are a lot of ranty articles running around the

net about US IT jobs going overseas. Of course its a valid concern and

a lot of the posts are based heavily in frustration (which can

sometimes undermine facts). One point I see raised again and again is

the statement that programmers overseas produce "crap code" or "have

no idea how to design quality code". Or, the old fall back "don't

understand true computer science principles!"

Bummer.

Partly by force, hmm.. actually, completely by force, I'm still a

professional graduate student at Syracuse University. I don't live

there, but I travel there every few weeks. I meet with members of my

Ph.D committee whom, as you can imagine, are an interesting bunch of

smart professors. They see my dissertation (that I've worked on for

several years now), read it for the first time, and pull some obscure

embedded issue out of the depths of the idea. I'll never stop being

amazed at their ability to do that. Its like someone else telling you

your 10 year old kid has a 3rd arm - I never noticed that before!

Then again, that's their job. They are scientists - they can wrap

their head around new, complex concepts very quickly and immediately

assess the problem from many angles. Despite their brilliance, some of

them still don't use email. That's kind of the running joke isn't it?

The brilliant computer scientist that doesn't use email - or hardly

uses computers for that matter. The fact that this is a joke is a

clear misunderstanding of the line between computer science and

computer practice.

A lot of programmers seem to credit themselves pretty quickly with

understanding computer science. Surely some do. In some respects its

like the guy who has skiied all his life and then at middle-age takes

a skiing lesson. He does it to join a friend or just for fun. Often

enough he learns plenty. If nothing else, he learns other people's

viewpoints on things he figured out for himself. In fact, hopefully a

ski instructor is teaching from a source that has combined knowledge

from the world's best skiers over many years. Chances are you don't

figure out something on your own the same way that pooled knowledge

from hundreds of people before you figured out. You're bound to pick

something up.

The real truth is that knowing a lot about computer science isn't

seriously required for much today's programming needs. Personally, I

couldn't sleep if I knew an O(n^2) algorithm sat where a clean

O(nlogn) would fit - but thats my now inborn nature. Making a webpage

simply doesn't require much knowledge about complexity theory,

algorithms, or data structures. Now, before you go posting at the

bottom of this article that this is poop - I think you're forgetting

something. Surely there are implementations that would benefit from

running the best algorithms and the best quality code, but by far most

of those applications will run just fine with "good enough"

implementations too. That's the rub really.

Don Knuth (a pre-eminent computer scientist/mathematician for those

newbs) made a famous quote something to the effect of "I have proven

that this program is correct, but I haven't tested it". Sort of a

funny joke to prove something is correct but not be sure it will

actually work. If you've actually ever done any program proving

(probably in some computer science cirriculum), you know it can be a

tedious mathematical exercise. The only case I know of where

commercial software was actually proven correct was in the fly-by-wire

Boeing 757 aircraft (this is from memory, if anyone needs to correct

me please do so). That is, they didn't just write the code and test it

a bit to see if it works - they mathematically proved that the

software would work as prescribed. Sounds good to me since I'm on a

757 often enough.

There's a pretty interesting dividing line between science and

practice. Scientists prove, practitioners test. With all the hoopla in

our industry about "test-first" its pretty plainly a practical world.

Testing is fine, but its empirical and can miss details. Its rare that

you can test all possible input in an application. Maybe we all should

be more like computer scientists. Maybe we can abandon this

"test-first" mentality and adopt a "prove-last" one. I'd rather have

proven software than tested software. You can still test if you like

(as Knuth mentions). Then again, for a practical world its obvious

that despite the fact that empirical testing of software is almost

never perfect, clearly its "good enough".

Hopefully at this point, we can stop calling things like servlets and

web-services "computer science". I wear both hats, some days I'm a

groveling academic student and other days I'm a practicioner providing

solutions for a client. The most distinct element between science and

practice comes when I write academic versus business prose. As my

professor says about my dissertation, "you must be able to defend

every sentence". I've learned quickly (although sometimes I still make

mistakes when I context switch) about what and what-not to say in an

academic paper. Here's some examples:

* Never say anything is "optimal" unless you have concrete proofs.

We're not talking about amazing facts here, we're talking about

obvious stuff like "the shortest possible method is an empty method".

And darn it, your thingie better be optimal.

* Assume your reader is an accomplished computer scientist, but if you

use a word that isn't normal english, it better have a definition

first. Again, I'm not talking about "non-polynomial running time", I'm

talking about "parameter list".

* Empirical evidence is cute. But you better have a lot to make any

claims off it. Even then, you can only say what you have is of worth

"in practice". Unit tests won't get you anywhere with professors.

* For any algorithm that has any value in it - you best know whats

going to happen if that value were replaced by infinity.

Me: So the class has several methods and the algorithm examines each

one.

Professor: How long does that take? (he means algorithmically, not

wall time)

Me: Its linear - runs in O(n)

Professor: What if the class had BILLIONS AND BILLIONS of methods!???

HMM?

Me: Uh.. it can't. Java wouldn't support that. I think you can only

have like 256 methods or 32767 or something.

Professor: And what if it had an unbounded INFINITE number of methods?

What would happen THEN? hmm?

Me: It wouldn't because even if every Java programmer started typing

in methods in some grand coordinated method-creation effort and every

computer ran some method-creating program to help make more, we'd

still never get an infinite number of methods. Not to mention no

compiler or VM would accept such a class. Is this relevant?

Professor: And what if each of those methods had an INFIINTE number of

instructions and ran for an INFINITE amount of time? How long would

your little algorithm take THEN!?

Me: hrf.. um.. its linear.. runs in O(n). Which I suppose if

everything was INFINITE - it would be a really, really long fricken

time. But you and me and your dog too would be DEAD by then and we

wouldn't CARE anymore and my damn tombstone would say "DIED WAITING

FOR PHD DEFENSE" and yours would be like "DR. BOB PAINGIVER - I

SUCKED". Besides, most of the classes in question have only like 3

methods or so.

Professor: oh.

Unlike things like biochemisty, computers (and hence some computer

science) is actually fun to learn. Few people become accomplished

biochemists in their spare time. But plenty of folks gain useful

computer skills just by toying around. Its a pretty thin definition to

say that programmers with no formal computer science can't make good

computer programs. Of course they can. Languages like Java and C# have

built-in sandboxes to protect us from ourselves. Good algorithms are

buried away in collections libraries. How many programmers really know

how much faster a binary tree is over a hashtable for searching? Does

it really matter in commercial computer software? You might say yes,

but the truth is most software in existence today is written far from

algorithmically optimally. In other words, its "good enough". That may

stink, but its also the truth. It also may NOT stink - whats the

return on investment pristine computer science in business

applications?

The joke of the computer scientist not using email is not very

interesting to an actual computer scientist. Its like asking a

petroleum engineer if he knows how to drive an oil truck. Most

programmers today are practitioners because thats what businesses pay

for. If overseas programmers "suck" - don't worry, they'll get better.

And it probably won't mean anyone learns a whole lot more about

"computer science".

Printer Friendly Page