[jakobe]

> relies on the assumption that there is something special about the mechanisms behind the human brain's function

Imagine someone spilled the contents of a large trash in a pile on the floor. Now it is your task to create an exact replica of this pile.

There was absolutely nothing special going on when that pile of trash was generated, and yet it is impossible to replicate.

Sure, you can create similar piles of trash by spilling other trash cans on the floor, or you can carefully arrange banana peels an half-filled soda cans to create a pile that looks similar from the surface. But generate an exact copy? No way.

[graeham]

It depends on how 'exact' exact is. For a brain, there are a finite number of neurons with a finite number of dendrites and axons. A quick search suggests 2.5 pentabytes (2.5 million GB) of storage http://www.scientificamerican.com/article/what-is-the-memory.... So about 2.5 thousand hard drives, so not an unreasonable amount of complexity to deal with today.

Your argument seems to suggest there is some finite level of complexity which is the limit of engineering. What is this limit and how is it justified?

If our ability to scale to complex systems scales linearly, this could take a very long time. If it follows a power law (like Moore's Law), it could be feasibly much faster.

The trash argument is weak because the pile serves an identical function regardless of how exact or inexact it is.

Probably the challenge is not in making an artificial brain, it is in reading a current one (especially one that is living without damaging it).

[jakobe]

The argument with the trash I was trying to make was that we don't need magic to make things that are beyond our understanding. I do believe that our conscious derives from the physical structure of our brain, but that doesn't mean it is possible to "upload" it.

Yes, part of the impossibility stems from the impossibility of reading a brain. Non-destructive measurements aren't possible because the amount of energy required to scan the tiny structures in the brain would destroy it. Destructive measurements would destroy the brain while we read it, and we can only read a subset of the information before the whole brain is destroyed.

But even if it was possible to scan a brain, I think you are underestimating the task of creating an artificial brain. Storage is the least of our troubles. To actually simulate the neural net, each of these two thousand hard drives must be connected to every other one, requiring millions of interconnects, and then you need millions of processors, and every simulation step touches every byte of every hard drive. This is so many orders of magnitude beyond what we can do now that I doubt it is ever possible.

I think the only feasible way of uploading a brain will be to create a program that can convince everyone that the uploaded person is actually living inside the computer, similar to the Turing test. You'd configure the program by telling it anecdotes from your life and taking psychological tests, rather than "scan your brain". However, I consider even this variant unlikely, because it would basically require something similar to an AI.