jaac wrote:Try to make a gemini that is closer to Von Neumann universal constructor:
https://en.wikipedia.org/wiki/Von_Neuma ... onstructor.
It is true that you'll have many defective copies. But just think how much further you might get by making something that evolves on it's own!
...
I think that the next step is making it into the game of
life. Making it go further than our own imagination.
It's actually technically possible to build a Conway's Life pattern that evolves in this kind of way.
Unfortunately, it's first necessary to "crash-proof" the B3/S23 universe, which otherwise is so fragile that small pseudorandom changes to an EvoGeminoid's genetic code would inevitably cause the "giant cloud of junk" that Kiran describes.
Crash-proofing can be done by designing Geminoid universal constructors that can only interact with each other in a strictly limited number of carefully defined ways. Then evolution will happen only in the "safe" interactions between Gemini U.C.s, and not at the lowest level where a misplaced glider or block or single cell almost always causes irrevocable decay.
Unfortunately, it then turns out that the resulting system can generally be modeled by a completely different CA rule, treating each Geminoid U.C. as a single cell. That means that if evolution is what you're looking for, there's no particular reason to bother designing the Conway's Life universal-constructor patterns at all (!). Whatever interesting evolution might happen, it will happen exactly the same but billions of times faster, in whatever CA rule the U.C.s are collectively simulating. Might as well just set Golly up to run that rule.
For example, it's possible to design a Geminoid replicator that has several possible internal states, and checks the states of any neighboring replicators before deciding how to change its state, or whether to self-destruct or replicate. Something like this could function as a metacell, emulating Conway's Life or some more complex rule -- this is
calcyman's idea from a few years back.
But imagine that you set up and program your Geminoid U.C. replicator-metacell to emulate, let's say, Evoloop. You'll get some interesting evolutionary behavior, sure -- but it would be the exact same behavior that you'd get if you simply ran the equivalent pattern in the Evoloop rule.
Thus, oddly enough, an extra B3/S23 Geminoid layer in the simulation doesn't actually add anything interesting -- it mostly just makes it incredibly inefficient to run the evolution experiment.