Food for thought:
There exists a kind of programmable microchip, called an
FPGA. It is little more than a whole bunch of logic gates on a silicon substrate, that can be flipped on and off, to create a simulation of nearly any other kind of chip, on the fly. The most frequent use case is for testing CPU designs. Thus, the very same FPGA can go from being a DSP, to being a GPU, to being a CPU, just with a little extra "software." The resulting simulation is not emulation; the logic gates in the FPGA are given a configuration that is identical to the gate layout that would be inside those chips, and then it "Simply does" those things.
Why is this relavent?
Recent research suggests that memories are not stored in the synapes between neurons, but rather inside the neurons themselves.EG, each individual neuron develops and retains a small datastore, and a processing system. In other words, each individual neuron acts quite a bit like an FPGA. The way in which it fires, depends upon the configuration information it has stored inside it.
The argument that mind/body dualism is inappropriate, because the mind and the body (the brain) are inexorably linked, gets thrown a curveball by this. It means that if we provide enough horsepower in each of our synthetic neurons to handle every possible state/configuration a biological neuron can handle, and have a reserve pool of such synthetic neurons on our neuron simulation chip sufficient to meet or exceed the expected rate of neurogenesis vs apoptosis, we can simulate a brain on silicon, without resorting to software emulation.
The resulting hardware would be more easily scalable by design, meaning we could then tweak some of the parameters to allow the active simulation to become larger/more complex than current wetware offerings, and we wouldnt be bogged down by the slowness of serial processing. (And in case you were wondering, each neuron would likely operate orders of magnitude faster than organic ones, since organic ones fire quite slowly.) Ultimately, there would be a tradeoff, however. Synthetic collections of such "neurochips" (
which are already becoming a thing, btw) will run into inter-chip communication issues, due to signal propogation delays that are inevitable. This means that even with synthetic components, there will be a tradeoff between total array size, and effective processing speeds. (Bigger arrays are harder to keep usefully synchronized, due to this propogation delay. Sorry, but there is no good solution to this problem-- it will persist even on purely optical computers, since the speed of light cannot be violated for information transfer.) That's fine though, since the increased native speeds of the synthetic neurons over their organic counterparts means that substantially larger arrays are theoretically possible.
Also, just a nitpick--- Not all computers are digital logic. The earliest electronic computers were fully analog logic. Digital computers are simply easier to design and program, which is why we switched to them.
I fully expect that full sentience on artificial substrates will eventually become a thing, if humans keep doing what they are doing.