I'd like to address these points from the other point of view.

You cannot predict where your targets are. ... Therefore, you cannot predict where your "immune system worm" will go.

This is a good thing. Any sufficiently-complicated system will be "unpredictable", in that given a surface-level snapshot of a given state and a surface-level look at the stimuli, you cannot with certainty predict the next state. All we can do is say "It is likely that the system will move in this direction, probably hitting these states along the way."

For example, we cannot predict the path a given packet will take to get from point A to point B, even in a small network. In fact, we don't want to be able to do that! We want the system to "think for itself", so to speak.

Now what we have is a piece of software with an inherently umpredictable migration pattern and that software will have bugs!

First off, you just described Windows. It's a piece of software that does way too much, we can't predict where it will be, and it has bugs. So what?

Every complex piece of software is reasonably certain to have bugs. The trick is to minimize the number and the impact of those bugs. You know the usual methods - packages, objects, testing, etc. This really is a non-objection.

Crackers are going to have fun taking your code and modifying it to seek out and destroy what they want.

This is a more complicated objection to deal with. The first, obvious, answer is to have your anti-virus make sure that any program that says it's the anti-virus passes certain criteria when checked. Things like size, checksums, etc. That would keep a lot of the silliness in check.

However, a better answer is to reformulate the idea. Instead of an anti-virus that goes around killing other virii, instead it should go around closing security holes.

Yes, I know this is an intrusion into other peoples' systems and changing stuff around, but my response to that is twofold:

1. Their systems are compromised, and thus are a threat to my systems.
2. They may not be aware that this loophole exists, and may never be aware that it was fixed.

Interesting idea, but too many problems.

That's what they said about GUI OS'es, the airplane, and the re-useable spacecraft, as well as irrigation, I suppose. I seriously hope that this isn't the answer we will give in the future. Instead of dismissing an idea, how about working it over? How about coming up with possibilities? Maybe, the idea as initially presented doesn't work very well, if at all. I know 99.999% of my ideas never work at first. It's only after they've been hashed out and tested and retested and thrown out and brought back that they become useable. E-bitch has a good idea, one I (and thousands of other programmers) have thought about for a very long time. Maybe it's time that a few of us actually do something about it, instead of giving up and saying "Interesting idea, but too many problems." Very frankly, I give props to Micro$oft for actually thinking about how to make this work. (Never thought I'd say that, huh? *grins*)

------
/me wants to be the brightest bulb in the chandelier!