Why Hasn't Partial Self-Replication been Achieved?

[Luminal]

I understand that complete artificial self-replication (replicating the CPU and memory and various electrical components) is an incredibly difficult task and may take a while longer to achieve, presumably using highly progressed machine learning. However, complete self-replication is not needed to have effective mechanical self-replication.

 

A central server or computer would handle the commands to robotic arms (or robots), and each robotic arm would have just the sufficient capabilities to manufacture simple wires, primitive motors, and maybe a a radio receiver. Robots on assembly lines are already quite capable of this task.

 

The physical range and number of the replicated arms would be finite of course, and they'd need to be recharged periodically. The number would be determined by how much memory/processing each arm required, as well as the requisite raw materials available in the vicinity. Nonetheless, it would be self-replication.

 

(Skip the rest of the post from this point if you would like to avoid nonsensical speculation )

 

Furthermore, each arm could be given a secondary directive (consuming perhaps 25% of its total operation time) to attempt new semi-random activities. If this activity resulted in more efficiency/independence (measured by how long arm could operate before needing to be recharged) then this activity would be sent back to the central computer, and broadcast to all other robotic arms who would then assume that behavior in their primary operation. Otherwise, all other behaviors that were wasteful and did nothing are ignored. Quasi-natural selection leading to quasi-evolution.

 

Eventually, the behaviors could evolve to feasibly allow the arms to gather their own resources and become almost independent. The "almost" would go away if the arms could evolve some form of self-control. To promote such an evolutionary pathway, the programmer would open up "transitional forms" by allowing the arm to randomly generate its own commands to its arms, yet while still being nearly fully controlled by the computer. To this end, the arms would at the very least need a simple circuit board, thus complicating the process of self-replicating, but enabling the possibility of evolving true self-replication over time.

[bascule]

It has been achieved. Check out RepRap:

 

http://www.reprap.org/bin/view/Main/ShowCase

 

The basic idea behind RepRap is that it's a rapid prototyper (i.e. a combination Computer Numerical Control (CNC) milling machine with a resin "print head") which is capable of fabricating many of the parts needed to construct itself, with any additional parts needed to build RepRap being easily and relatively cheaply obtainable in most parts of the world.

 

In addition to building parts for itself, it can build simple solid objects (think things like combs) as well as complex machined parts. This leads to an almost Ikea-like approach to building more complex machines: you download a copy of the object you want to build, which includes 3D models of all the parts the RepRap can build, a list of other parts you must buy, and a list of instructions for how to assemble the intended device.