Friday, 1 April 2016

Why I Think It's Impossible To Replicate Our Brains

In my view, the author and computer scientist Ray Kurzweil is one of the most interesting people around. Kurzweil is a bold futurist who has every confidence in the law of accelerating returns, a phenomenon that predicts a continual exponential increase in technologies in ways that will keep improving and enhancing human well-being. There is plenty of evidence to suggest he is likely to be right, although for various reasons, many of which are unpredictable, apparent exponential growth patterns can quite easily level out, and probably will in some areas of technology.

On a more specific note, in his book The Singularity Is Near, Kurzweil has made a very interesting prediction: that by the year 2040 we will be able to build the equivalent of human intelligence by scanning the brain from the inside using nanobots. Once we know the precise physical structure and connectivity information, Kurzweil says we will be able to produce functional models of sub-cellular components and synapses and replicate whole brain regions.

Kurzweil talks of "uploading" a specific human brain with every mental process intact, to be instantiated on a "suitably powerful computational substrate". Rather than an instantaneous scan and conversion to digital form, Kurzweil thinks humans will most likely experience gradual conversion as portions of their brain are augmented with neural implants, increasing their proportion of non-biological intelligence slowly over time. Quite how much time, he's not sure, although he offers a suggestion of 1016 calculations per second (cps) and 1013 bits of memory, plus the possible additional detail that such uploading requires, which could be as many as 1019 cps and 1018 bits.

This all sounds intriguing, but for my mind there is a possible problem that underwrites the above scenario - it doesn't seem possible to me to replicate the 'you' that you know as your first person selfhood, and ditto that for any unique human, for reasons I'll explain.

First, don't get me wrong, future advancements will astound us in all sorts of ways we cannot currently imagine, and there probably will be forms of artificial intelligence that we can interface with at a level similar to the human-human interface we enjoy. But I personally think that the 'you' and 'me' we each know to be our own mind is something of such unique complexity and evolutionary finesse that it will never be able to be precisely reconstituted in any kind of artificial intelligence. In a nutshell, the 'you' that makes up your first person selfhood is an utterly unique aggregation of mental machinery that can probably only be retained in the biological apparatus that you call your brain. I will explain why with a thought experiment.

Any time AI program writers try and extend the world of brain cells into the world of transmitter molecules they would then have to try and simulate hormones. And these, in turn, depend on genetic instructions which are themselves only partially programmed and highly adaptable. Simulating the human brain is not just about replicating the hardware that we see in the form of neurons and synapses, it is about replicating a lengthy evolutionary process that goes right back to the origins of biology itself. For example, some of our genes involved in the development of our cognition (even at the embryonic stage in the womb) are genes that go right back to the very beginning of all life forms billion of years ago - and these are genes and hormones that form the substrate of cognition prior to the point at which neurons and synapses become involved.

Let us suppose, though, that we overcame all those hurdles and developed the scientific wherewithal to attempt brain replication in an external agent - a computer or an android, as Kurzweil suggests. Here’s why I still think your own ‘self’ will remain unique to you. Suppose at this stage we can produce a short-cutting algorithm that enables us to reduce the human brain atom by atom and then reconstitute the exact atomic configurations of the original mind, reproducing the conscious cognition. What would that entail?  In the human brain there are 10^11 brain cells (that's 100 billion), and 10^14 atoms (that’s 100 trillion) in each brain cell - that makes 100 billion x 100 trillion atoms, which is 10^25 atoms (or 10,000,000,000,000,000,000,000,000). 

Now, suppose we were to replicate the brain at the rate of 1 trillion atoms per second. Even at that rate it would still take us 317,097 years to reconstitute the full brain. It sounds like all talk of reducing at the atomic level is too far, but it shouldn't be – after all, Alzheimer's sufferers experience degeneration atom by atom, one at a time, and babies are formed in the womb atom by atom, so evidently these effects do impinge on physical states, they are just processes of natural development that happen really fast.

It was once thought that consciousness was a precise configuration of proprietary parts, and that in assembling a brain from scratch there would come a point at which the first person perspective of consciousness would be switched on - rather like assembling a circuit board of cognition that lights up when all the proper connections are made.  But our aforementioned knowledge of degenerative diseases like Alzheimer’s shows that this isn’t the case – at least, it shows that atomic reduction can occur bit by bit and that the cognitive abilities of that mind operate with different degrees of composite integrity. 

The problem I see (in relation to compromising the ‘self’) is that if in our thought experiment we reassembled a brain from scratch one trillion atoms at a time we could not reconstitute the unique first person perspective because the first person self would at some point become aware of cognita being partially reconstituted, at which point (and from then on thereafter) a new set of conceptions and experiences would be taking place during the restoration process. So the simulated ‘you’ being uploaded would at some point begin to take on thoughts of its own before the real you in its entirety was uploaded - meaning that a partial you had begun to generate new and unique thoughts.

Ok, let's speed up the process. Even if you doubt the effects of atomic reduction, and we instead chose brain cells as our object of replication, and replicated them at the rate of 1000 per second (a feat that would take unimaginable technological sophistication), it would still take over 3 years to do the whole brain. In that 3 years, such an activity is bound to cause the simulated ‘you’ to take on thoughts of its own before the real you in its entirety was uploaded. However much we can reduce the execution time - months, weeks, even days, it seems impossible to replicate the unique 'you' or 'me' that make up the totality of that first person selfhood, because during the uploading process there surely will come a point at which the original you and the copied you each start to develop new and unique first person cognita - cognita, in fact, that is being caused by the experience of the replication process itself, as well as all the interference at a neurological level.