0
1

[–] fusir 0 points 1 point (+1|-0) ago  (edited ago)

So we should also talk about what it takes for an electronic system to be self aware. I personally don't like the term self aware because if my ifconfig includes a 127.0.0.1 and if I can send to an api my hostname for a callback function then technically it is self aware. I programmed it to be self aware.

What we really mean is intellegent enough to independently become conscious. To do that with a machine you likely need a neural net, a simulation of an animal brain. I see three classes of mechanical system we need to talk about. A general computing device built out of mechanical parts. A machine calculator built for the math involved. Machinary that can perform in analog what a computer would do digitally but is simulating analog.

The way a computer does NN is mathimatically intensive so I think a mechanical general computation device would get run into the ground running NN code on it.

One way to speed this up is to instead build a physical unit for each node of your NN and each synapse of your NN and have them do the math for themselves. So you would still be treating NN as a mathimatical process but you would be parallelizing it with machines that can only do one kind of calculation, aposed to computing an NN on something with mechanical RAM and a processor.

The way NNs work is to simulate the analog processes in our brain. Therefore we might be able to do the whole thing in analog. There are two different types off NN. One is a type of filter/transform on data. Data goes in one layer, goes to another, and so forth until it reaches our end layer. Each node of one layer is connected to every node of the next layer and each connection has a weight for how much impact it applies. The impacts are summed on each node and the result is "normalized" bringing its result closer to zero or one but not absolutely zero or one. The communication and sum of synaptic signals could be either a series of differencials (mechanical) or the sum of applied pressure if it was piston based.

The next kind of NN is stochastic meaning it has random impulses that fire. It's "layers" aren't organized necissarily into actual layers. If an impulse is given to one node it makes it more excited and more likely to give off impulses itself. Each node has different properties about how recent an impulse effects its likelihood of giving an impulse. You would need to create some way for machines to behave probabalistically and dependent on "recent" events.

0
0

[–] fxm ago 

There's no reason why a mechanical machine can't be Turing complete, as Turing's own thought experiment showed. Any Turing complete mechanical machine should be able to emulate any electronic or biochemical system to any arbitrary degree of precision necessary. The main advantage to electronic and biochemical systems is speed. An awareness built from mechanical units would probably be a lot slower, and require a lot more energy to accomplish the same results, but there's no reason to suppose it wouldn't be just as self aware as you are.