Entry tags:

science

On computer "science"
It is a periodic source of amusement contemplating the imbecilities of the various branches of computer science. AI is a favorite one.



You have to admire the pluck of "computer scientists" who think that, despite their inability to write decent high level compilers (like, say, natural language coding), or come up with a methodology which prevents common software bugs, they think they can reproduce the functions of the human brain with a tin can. I suppose it's a sort of giant solipsism in a sense; many people who attempt such things are often lacking in even vague knowledge of how human beings work.

Really, AI is the only branch of computer science with any pretensions of being a science these days; as such, the most interesting things show up here. It's kind of amusing to track via Peter Norvig's books on AI. They aptly demonstrate the active devolution of AI techniques. From reasonably interesting compiler tricks and techniques in knowledge representation in Lisp to insipid uses of Kalman filters (an averaging technique stolen from analog computing in missilery intertial navigation ... invented in the 1950s) and "logical agents" in Java. Just reading the books, you get the sensation of talking to an Alzheimer's patient who has forgotten more than he knows. And these are easily the best and most interesting computer science books I've yet read.

Finally some people are realizing what transparent quackery the whole field is. They went from grand predictions of being able to simulate the functions of a human brain by the year 2000 or so, to admitting they can't even write anything more sophisticated than 1960s style eliza programs in attempting to pass "the Turing test."

Even Marvin Minsky has finally come around to this view, "AI has been brain-dead since the 1970s.... For each different kind of problem, the construction of expert systems had to start all over again, because they didn’t accumulate common-sense knowledge.... Graduate students are wasting three years of their lives soldering and repairing robots, instead of making them smart. It’s really shocking.” The rest of the article is relatively good, and, as it was inspiration for today's rant, probably worth reading.

It's quite amusing how the average computer science graduate (who has probably never even peeked into AI ideas to see what they are all about) will loudly and indignantly insist that the computers are coming, that it is only a matter of time, blabbity blah blah. Such jingoism makes me laugh and laugh; not only do they have no sense of their own field's history, they have no sense of what they're talking about. Most of these guys don't even know what a real number is, or will assert something silly about them, like that they can be represented "just fine" by 64 bit integers, like they are in a computer. Anyone who tells me something like that doesn't understand the first thing about the world of dynamics. Real numbers (rather than continued fractions) have physical consequences that can be measured, actually. Go look up the Ford paradox if you don't believe me. Or, since someone mentioned simulated annealing today; go think about why it's so hard for a computer to do something which nature finds so easy to do (yet, it is powerful enough to do very impressive things, basically by representing very large chunks of information in a physical model, rather than a register -think enough about it, and you'll see why analog computers might be a good idea).

I suppose the average computer science geek is generally a sort of power mad fancy plumber. Such an analogy actually holds; plumbers in the 1800s were a lot like computer scientists today. They were and are clever working class guys whose cleverness bumps them up a couple of social and economic classes. In the 1800s, plumbing really came into its own, and it more or less solidified into the useful field it is today. But back then, it was seen as a high technology cure for all ills. Whether for piping water in, piping shit away, or piping power around in the form of steam; all technology was seen as involving lots of pipes in them days, and clever pipe fitters would come up with all kinds of contraptions. Now a days, it's all about computers. Just like in the old days, a piece of technology looked all the better with lots of pipes all over it, now a days, a piece of technology looks all the better with lots of microcontrollers or pentiums glued to it, running half-baked software. No doubt the old time plumbers thought they could do impossible things with their pipes, just as their modern counterparts do.




High tech with pipes

Really, whatever can be done now with a computer was in principle doable in 1965. There hasn't been any large breakthroughs in computer science cleverness since then. And I only chose that year as that's when Cooley and Tukey rediscovered (from Gauss; duh) the fast Fourier transform; the last big increase in computing power which came about due to an algorithm (barring that, 1962 or so was one of the last years of original computer science research). Everything else has been due to improved surface science techniques, making faster chips and better mass storage devices. So all those wonderful tasks of sorting, classifying and adding which modern computer scientists get paid to painfully teach computers how to do; they're all possible because some jerk knows how to make clean silicon wafers. Much like a lot of plumbing and steam technology came about because of improved metallurgy and machining techniques. Oh well; you can't blame the plumbers for thinking their plumbing is a source of ultimate power. For them, it is a source of personal power. But it ain't the answer to most questions.



There are basically two promising, unrelated (unless you believe Penrose, which I don't) areas in computer science today. Non Von Neumann architectures, generally phrased (generally wrongly) as "quantum computers" is one of them. The other one is neuro anatomy. Once one of them really begins to bear fruit, we'll see some real changes (such as happened when computers were first invented). Meanwhile, the latest fashion in objects or patterns isn't going to get you far. Hell, the guy who came up with the "patterns" idea is a Berkeley architect. And if you've ever spent time in Berkeley, you can see where that got us.