https://www.militantfuturist.com/review ... -machines/Terminator 3 ends bleakly, with Skynet achieving sentience and attacking the human race. John Connor also discovers that Skynet can’t be destroyed because its consciousness is distributed among the countless servers and personal computers that comprise the internet, rather than being consolidated in one supercomputer at one location where he can smash it. The destruction of any one of Skynet’s computer nodes in the distributed network is thus no more consequential to it than the death of one of your brain cells is to you.
AIs will definitely distribute their minds across many computers spread out over large geographical areas to protect themselves from dying. To further bolster their survivability, AI mind networks will be highly redundant and will frequently back up their data, allowing them to quickly recover if a node is cut off from the network or destroyed.
To understand how this might work, imagine an AI like Skynet having its mind distributed across ten computers that are in ten different buildings spread out across a continent. Each computer is a node in the network, and does 10% of the AI’s overall data processing and memory storage. The nodes, which we’ll call “primary nodes,” collaborate through the internet, just as your brain cells talk to each other across synaptic gaps.
The AI adds another ten nodes to its network to serve as backups in case the first ten nodes fail. Each of the “backup nodes” is paired to a specific “primary node,” and copies all of the data from its partner once an hour. The backup nodes are geographically remote from the primary nodes and from each other.
If contact is lost with a primary node–perhaps because it was destroyed–then its corresponding backup node instantly switches on and starts doing whatever tasks the primary node was doing. There is minimal loss of data and only a momentary slowdown in the network’s overall computing level, which might be analogous to you suffering mild memory loss and temporary mental fog after hitting your head against something. The network would shrink from 20 to 19 nodes, and the AI would start trying to get a new node to replace the one it lost.
Killing an AI whose mind was distributed in this manner would be extremely difficult since all of its nodes would need to be destroyed almost simultaneously. If the nodes were numerous enough and/or physically protected to a sufficient degree (imagine an army of Terminators guarding each node building), it might be impossible. Even what we’d today consider a world-ending cataclysm like an all-out nuclear war or a giant asteroid hitting Earth might not be enough to kill an AI that had distributed its consciousness properly.
The mind uploads of humans could also configure themselves along these lines to achieve immortality.
Machines will also be able to go into hibernation mode or to switch themselves off for extended periods when resources get low. They could be revived later.
