The three pound enigma // EIT Digital

The three pound enigma

The TOP500 for the last decade and some data for the HPCG (data collection only began recently). The dashed lines indicate a forecast. The main drivers of computational growth are also shown: Multicore CPU, GPU, and in 2016-2017 3D memory, and some new unknown technology in 2020. Will this growth be sustainable? Credit: Tim Dettmers

Almost 10 years ago I read a nice book, the three pound enigma, about the brain and how it "worked", that is processed data and made sense out of them giving raise to consciousness. Now that book is still worth reading today but it is completely superseded by the new knowledge acquired in these last ten years.

In the book The Age of Spiritual Machines (199) Ray Kurzweil, an authority in the field of prediction with very good statistics backing him up, predicted that a computation power equivalent to a human brain (20 quadrillions -US- calculation per second, that is 10 to the 15th power) at a cost of about 4,000$ (1999 equivalent, that is 5,700$ in 2015) would be available by 2019. 

As a matter of fact, Thiane 2, the fastest computer available in 2015, can crunch 55 quadrillion calculation per second, and we might expect that such a processing power should be available to a 5,700$ (2015 equivalent) computer in the first part of the next decade, hence the Kurzweil prediction seems to be not that much far off the mark. At the same time, it is completely wrong but that is not because of a lack of evolution in processing power rather because what we know today about the processing power equivalent of our brain is completely different from what it was believed to be back at the beginning of this century and even just few years ago.

If you are interested in this area you have to read the article published on July 27th by Tim Dettmers. I loved it and I bet you will love it too! (be careful of the numbers presented, there are some mistakes, like when he is putting the number of synapses per neurone in the cerebellum at 25,000 million whilst there are "only" 25,000).

Indeed, in the article Tim shows how the knowledge that has been harvested in the last five years has completely changed our (still far from complete) understanding of the brain pushing its processing capability to 1,000 EFlops, that is 10 to the 21st power, or 20,000 times more than it was believed to be 10 years ago.

Interestingly Tim also makes the point that it is not "just" the matter of higher processing capacity but a matter of "economics" to reach that capacity.

As I pointed out the Moore's law stopped its magic, from the economic point of view in 2014. Achieving a 1,000 EFlops processing power is expected to cost over 4,500 billion $, more than the Germany GDP! And there is no solution in sight that would decrease this cost in the foreseeable future. Which means, in other terms, that Ray Kurzweil singularity (the point where computers will be "smarter" than human brains, predicted to happen in the 2030 timeframe) is nowhere to be seen.

Of course you need to read the whole article to appreciate how much knowledge has been acquired about how the brain process information but to give you a first hint on where this amazing 20,000 times factor is coming from consider that:

- ten years ago the processing unit was considered to be the neurone and there are some 100 billion of them (85 billion estimated in an adult brain, many more in the young and many many more in the fetus, up to 1,000 billion).

- today we know that processing occurs both in the neurone and in the dendrites, in the axon and some preprocessing occurs at synaptic level. There are 300 billion dendrites and 10 trillion synapses and this multiplies the processing capacity of the brain. 

- today we know that neurones have a DNA that is different from the one we find in other cells in our body. Different neurones have different DNA. And this difference is the result of the adaptation of the brain to the processing of information. The actual processing is the result of interaction among proteins, they are the ones responsible for opening and closing synapses which in turns determines which signal gets processed.  There are some 10 billion proteins at work in each single neurone and they are produced by the neurone DNA. Scientists have discovered that the use of these proteins in the processing of information results in changes in the DNA. It is like the brain reprograms itself as consequence of its daily processing activity.

What is clear is that our brain has much higher processing capabilities than we believed to have just ten years ago. But, like ten years ago it still remains a three pound enigma.

Author - Roberto Saracco

© 2010-2018 EIT Digital IVZW. All rights reserved. Legal notice. Privacy Policy.