> Interesting, I haven't seen a SF book exploit this possibility - yet it now seems the most likely way of a human-AI war to go ;-)
That's exactly the plot of "Code of the Lifemaker" by James Hogan. An alien mining spacecraft is damaged by a supernova explosion and crash-lands on Titan. It starts building mining robots, but some of their programs are damaged by the supernova explosion. So they start to evolve (eventually evolving full sentience).
Posted Sep 26, 2013 18:11 UTC (Thu) by nix (subscriber, #2304)
[Link]
Of course, it couldn't actually work that way. The reason natural selection has room to work is that mutations in nature are both relatively frequent compared to bit-flips in hardware, much wider in scope (there is no bit-flip analogue of translocation!) and very often do not break or even greatly affect the organisms they happen in, because the code is redundant, many genes are duplicated, and most gene products are parts of networks with multiple redundant paths to the same end. *None* of this is true of human-written software, ergo there is no evidence that it can evolve (and plenty of evidence that it cannot).
As for Hogan, the depth of his understanding of evolution's "modern synthesis" is suggested by the fact that he become a creationist. (The depth of his understanding of other things is suggested by the fact that he became a Velikovskian (!) and multiple conspiracy theorist. Truly a man with a mind so open that his whole brain fell out.)
Asteroid "mining" with Linux and FOSS
Posted Sep 26, 2013 19:30 UTC (Thu) by Cyberax (✭ supporter ✭, #52523)
[Link]
Well, we can imagine that robots use a pre-weighted neural networks or something like it. In this case small changes can, indeed, accumulate.
And yes, the later Hogan's books grow increasingly more and more bizarre. I stopped reading after the fourth one. But the first book in the series is quite brilliant.
Asteroid "mining" with Linux and FOSS
Posted Oct 2, 2013 14:29 UTC (Wed) by oever (subscriber, #987)
[Link]
there is no bit-flip analogue of translocation!
Yes there is: instead of moving the code, you flip bits in the address referring to the code.
*None* of this is true of human-written software, ergo there is no evidence that it can evolve
Posted Oct 4, 2013 11:48 UTC (Fri) by nix (subscriber, #2304)
[Link]
Yes, obviously one can design software specifically to evolve under directed mutation, and this has been done and is useful -- but unless we design *everything* that way (including OSes), we'll never get a machine analogue of evolution, and even software designed to evolve under directed mutation isn't going to evolve in the same way under random bit-flips, because the virtual machine that executes that is a) almost certainly much larger than what it's executing and b) not implemented in this fashion, so the first bitflip there and you're probably toast.
