You're right, except that you're not recognizing that the boundary conditions of a designed circuit are going to be better understood than those of the evolved circuit by virtue of the fact that the designed circuit is an implementation of an abstracted understanding of the problem space. There's no guarantee that said understanding is complete, or is not significantly flawed; but the relationship between it and the working circuit is going to be fairly well understood as a matter of course.
In contrast, a genetic circuit's relationship with the problem space it is operating in is very poorly understood by those using it. It's initial condition is designed, of course; but as it evolves it becomes more and more negatively defined -- it relationship with the problem space is increasingly defined "merely" by it's failures, not by it's correct abstraction of the entire space.
Please forgive me, I'm having a hard time making my point in a clear and concise manner than doesn't involve pseudo-technical language.
My point is that a genetic circuit can't be known to be truly operating within the "true" problem space, it can only be known to be avoiding the space of whatever it is you've (probably poorly) defined as "failure".
Put another way, the aim in designing a circuit is to create one that will do what you intent it to do. It's ability to do so limited, of course, by your understanding of the domain of "what it's supposed to do". But that's a positive proposition, and is finite.
On the other hand, the negative domain of "what it's not supposed to do" is what defines a genetic circuit. And that domain is unbounded. You can never eliminate all failures, they are infinite.
People forget that there are far more failures in biological evolution (when a new algorithm is "tried") than there are successes. That works out well as long as life is ubiquitous, there is plenty of time, and the environment is more stable than not. But a designed solution would be better. Anti-evolutionists are making exactly the wrong conclusions based upon empricism: a creator, even the more abstracted "intelligent designer", would have created life that looked quite a bit different than what we see. To use an over-used but nevertheless good example, a "designed" biospere wouldn't have things like male nipples. Male nipples exist merely because they aren't a "failure condition". Evolution is not teleological, it isn't aiming for anything. This is a point that is completely misunderstood by the general public.
This relates to the whole GA thing because the concepts are similar. Whether or not there is a designer of the universe, there certainly is a designer of software and hardware. Designing is a better solution, generally. Except that the utility of design versus evolution reverses when one's comprehension of the problem space versus the failure space reverses. If you don't really understand the problem, you can't abstract it! But you can test for failures. There, GA becomes very useful and efficient.
Best of all is using *both* in combination. Use how GA navigates against the failure space to further your understanding of the problem space, and then better your design. Utilize "improvements" discovered by GA's only when they are well understood. Then let them continue to go searching the failure space.
To sum up: I know what I want a nuclear power plant control clurcuit *do*. Well, I think I do. If I understand this well enough, then I will understand the behavior of a circuit based upon this understanding failry well. In contrast, there's no end of things that I *don't* want my control circuit to do -- the only thing I really know about a GA is that it is avoiding those things it has already failed against. Do those things map out all of the failure space? Very likely, almost certainly, they don't.
Slashdot Mirror
"asteroids are extremely rare." Excuse me? What solar system do you live in?
...she works at RealNetworks. Heh.
You're right, except that you're not recognizing that the boundary conditions of a designed circuit are going to be better understood than those of the evolved circuit by virtue of the fact that the designed circuit is an implementation of an abstracted understanding of the problem space. There's no guarantee that said understanding is complete, or is not significantly flawed; but the relationship between it and the working circuit is going to be fairly well understood as a matter of course. In contrast, a genetic circuit's relationship with the problem space it is operating in is very poorly understood by those using it. It's initial condition is designed, of course; but as it evolves it becomes more and more negatively defined -- it relationship with the problem space is increasingly defined "merely" by it's failures, not by it's correct abstraction of the entire space. Please forgive me, I'm having a hard time making my point in a clear and concise manner than doesn't involve pseudo-technical language. My point is that a genetic circuit can't be known to be truly operating within the "true" problem space, it can only be known to be avoiding the space of whatever it is you've (probably poorly) defined as "failure". Put another way, the aim in designing a circuit is to create one that will do what you intent it to do. It's ability to do so limited, of course, by your understanding of the domain of "what it's supposed to do". But that's a positive proposition, and is finite. On the other hand, the negative domain of "what it's not supposed to do" is what defines a genetic circuit. And that domain is unbounded. You can never eliminate all failures, they are infinite. People forget that there are far more failures in biological evolution (when a new algorithm is "tried") than there are successes. That works out well as long as life is ubiquitous, there is plenty of time, and the environment is more stable than not. But a designed solution would be better. Anti-evolutionists are making exactly the wrong conclusions based upon empricism: a creator, even the more abstracted "intelligent designer", would have created life that looked quite a bit different than what we see. To use an over-used but nevertheless good example, a "designed" biospere wouldn't have things like male nipples. Male nipples exist merely because they aren't a "failure condition". Evolution is not teleological, it isn't aiming for anything. This is a point that is completely misunderstood by the general public. This relates to the whole GA thing because the concepts are similar. Whether or not there is a designer of the universe, there certainly is a designer of software and hardware. Designing is a better solution, generally. Except that the utility of design versus evolution reverses when one's comprehension of the problem space versus the failure space reverses. If you don't really understand the problem, you can't abstract it! But you can test for failures. There, GA becomes very useful and efficient. Best of all is using *both* in combination. Use how GA navigates against the failure space to further your understanding of the problem space, and then better your design. Utilize "improvements" discovered by GA's only when they are well understood. Then let them continue to go searching the failure space. To sum up: I know what I want a nuclear power plant control clurcuit *do*. Well, I think I do. If I understand this well enough, then I will understand the behavior of a circuit based upon this understanding failry well. In contrast, there's no end of things that I *don't* want my control circuit to do -- the only thing I really know about a GA is that it is avoiding those things it has already failed against. Do those things map out all of the failure space? Very likely, almost certainly, they don't.