Slashdot Mirror
A Primer On DNA Computing And Software Breeding
There's been some interesting article published lately in the realm of DNA computing and software breeding models - kinda the land where 1s and 0s and Darwin meet. ArtsTechnica has got a primer on DNA computing which goes over the high points of DNA computing, and is accessible to anyone who remembers high school bio. Feed Magazine has got an article that examines breeding software and what that means.
There's more to it than doing actual computation. These are arguments that have been had a thousand times over in the past: "I don't see the benefit right now, so what's the point?" It may be excessively useful, it may not. The point is, we have to find out.
:)
Even if DNA computing proves to be too cumbersome to implement, we can still gain lots from it - for example, perhaps hidden deep within it is another model of computation. Maybe we can find something that's better then a Turing machine (i.e., it can check another program for infinite loops and find them). Hell, it may prove to be a useful storage device.
DNA does operations on data. It does remarkable things with them and it can do a lot of things at once. Studying this is not a bad idea, but so is putting all or eggs in one basket. If totally concentrated on shrinking die/chip sizes, we'll probably regret it at some point. Push the limits. It's fun!
Woz
I've done a little bit of work in DNA computing, and my impression of the state of the art is that we're only at the point where we're wiring the vacuum tubes together in order to program (if that).
It's kind of a "Nanotech-Complete" problem, in many respects. The repertoire of available enzymes is limited to those that are otherwise useful in molecular biology (endonucleases, ligases, methylases, etc.), but for some applications, "designer enzymes" are needed, and the technology to produce arbitrary enzymes is just not here yet (though a solution to the "protein folding" problem might be possible, given sufficient conventional computational power, e.g. something like distributed.net, or seti@home).
The biggest problem with DNA computers is I/O, primarily input, or what's called the "encoding problem". Representing arbitrary information as DNA sequences is a computationally hard problem in itself, since one must ensure that the encodings are unique enough to only interact with each other in the desired ways which contribute to the solution of the problem, and most importantly, produce a true solution (i.e. do not corrupt the data).
The output can be handled in a variety of ways, perhaps the most promising of which involves DNA microarrays (a.k.a. "gene chips"), which could potentially serve as an interface between DNA-based computers and conventional silicon-based computers.
But I'm afraid we're a long way from having a general-purpose DNA (or more broadly) molecular biology-based computer. And by the time we have the technology to build one, I suspect the other applications of nanotechnology may have rendered the point moot.
LaoK
"Software evolution is a controversial theory holding the unproven belief that random, undirected forces produced a world of better software. Use at your own risk."
Yes, they were really gonna put this in the books here. Fortunately, it was thrown out.---
DO NOT DISTURB THE SE
This may sound like flamebait, but it's not... lately, it seems that there is a lot of hype about biological systems and how it "exploits evolution" or "uses the solution mother nature has been using for all this time", or some such nonsense.
If you actually read till the end of the article, you'll realize that what they have done is merely to re-encode the TSP (travelling salesman problem) in terms of DNA reactions. As far as I can see, there is no gain whatsoever from doing this -- a computer doing a brute-force search on the TSP problem would have yielded the solution much quicker than the DNA method. The only potential advantage of the DNA method appears to be the "compactness" of data and the "stochastic, massively parallel" nature of it. But at the end of the article, it specifically says that this stochastic nature of DNA reactions (or any chemical reaction, that is) itself is the barrier -- it doesn't scale well. WTF??? That means that you have a method of solving TSP which is "massively parallel" (and therefore, by some strange fuzzy reasoning, it is "better" than silicon-based methods) but which doesn't scale well. Big deal, back to square one.
IMNSHO a lot of this hype is just riding on the unfounded assumption that "stochastic" is "better" because "that's what Nature uses". BS, I say. Just because something is stochastic doesn't make it any better (except by superstitious association with "stochastic" processes like evolution or some-such.) You can get something useful like strong cryptography from randomness. But you can also get white noise from randomness. Until it's proven that a particular process actually has its merits (and not merely duplicating what can already be done by an existing computer or other method), it's all just hype.
Granted, the "massively parallel" claim seems to be more credible; but in this case, they've just shot themselves in the foot -- the supposedly good "massively parallel" nature of the DNA method turned out to be a limiting factor, as you wuold need impractical amounts of DNA to conduct any non-trivial computation. Except perhaps for some hype value associated with the phrase "massively parallel" (boy, don't we love that term. Beowulf clusters, "massive" multi-CPU systems, etc.), I see no value whatsoever in this whole thing. As far as I'm concerned, somebody just hit upon something in the lab. Big deal, scientists have been doing that for centuries. Let's see something real produced before it's hyped like the Next Revolution.
Poll Mastah
So, if you repeatedly use legacy code in your new code in an attempt to ensure that your new code will support the same sorts of programs and environments as your old code, do you call that software inbreeding?
And if so, does that make Microsoft a bunch of software rednecks?