The Road From Logic Gates to Perceptron to Mind Uploading

Preface
Ever since seeing the famous Hubble deep field picture released from NASA,
I got a habit that every time I stare at the night sky, I sink into dreaming
that one day, mankind will no longer be a species that endlessly fighting
each others, only for the sake of a small piece of resource on earth. We all
know that we are in a 46 billion lightyear universe across, with billions of
galaxies that will never lack of any sort of resource. Every tiny little bit of
the sky we see, will contain more than hundreds of galaxies. According to
the Drake equation, each one of them will have more than 10 civilizations
which at least would be as advance as us. Sad thing is, we do not have
enough life span to launch interstellar traveling. We would be dead in our
space ships on our ways out of the solar system. As we all know now, we are
stuck at this solar system. The only thing we have is this fully colonized blue
planet earth, which has only limited resource. Wars, conflicts and shortages
breaks out everywhere in exchange of our lives, and we can only pray that
one day we can develop a way to get infinite resource. I look everywhere for
such hope: fusion power, global government, Mars colonization, water purify
projects, faster than light traveling hypotheses. None of this has been
proven with current technologies. Until I looked into Artificial Intelligent,
which could be able to produce human intelligence that can keep running
thousands of years, which could be able to contain human consciousness
regardless biological aging, and which could be able to enable mind
uploading, and our dream to spread our civilization outside the only earth
and solar system, could become true.


Introduction
This paper peeps into artificial intelligent at different scales. I am not trying
to draw any conclusion, but to reveal the possibilities of such technological
advancements. The first part will look into how computation machines evolve
from logic gates to Turing machines, then to neuron simulations. The second
part is to discuss the possibilities of brain simulations which takes neuron
simulations to another level. Third part of the paper is about the attempts to
achieve the 100 billion neuron simulation barrier(human brain) with different
approaches such as super computing or distributed computing. The last part,
will get into more generalized and philosophy topics such as the memetic,
the teme and how the development of artificial intelligent will help forming a
new kind of society.

 
Part I
A perceptron is the basic unit of the computation model of a neuron. By
using the differentiable sigmoidal activation function, and by using the
mechanism to find the local minimum of a derivable function, the
perceptrons will be able to remember or redraw a close pattern of the
information being fed. But our brains are definitely better than that. We can
not only to remember some patterns, but also be able to abstract them,
recognize them later on, and reason with different abstract ideas logically to
draw new ideas or conludsions. One of the pioneer in this area is Dr Geoffrey
E. Hinton at University of Toronto. Long before perceptrons and the neuron
net simulation being use widely, there are several significant problems that
prevents neuron networks from practical usage. One is that a single layer
perceptrons can not perform XOR logic, another is that it is impossible to
find out the correct outputs in the hidden layers of the multiple layers
perceptrons. Geoffrey co-invented the backpropagation method, makes the
former problems solved. Also he proposed the Boltzmann Machines method,
tried to overcome the local minimum by using randomize function to jump
out of the local maximum. Now that we are able to train neuron networks
with some correct data, then the neuron network can automatically
recognize the logic and apply to other data. However, this is no where to
match with human brain. In a recent paper by Dr Geoffrey E. Hinton, he
reveals some very interesting discoveries which shows how our brain
abstracts information without being trained on purpose.


figure 1
As the figure1 shows, he uses multiple backpropagational layers on neurons
to abstract the 28×28 pixels of images of numbers to 10 neurons, without
telling the neuron network any label, classification, nothing. Not really in this
example, since he does feed in the numbers by hand, thus telling the
network all of the different looking “0”s are indeed 0, but at the latter
example he shows that this kind of neuron network can learn the raw data
entirely without any human interpretation. The idea is all the nubmers are
abstracted into this 10 neurons layer, and he calls these “brain states”. By
feeding different numbers, the network recognizes the number and fires
different brain states in the last 10 neuron layer, thus he proved a famous
doubt in philosophy that information do have a specific representation in our
brains, which will leads to the terminology “meme” and “teme”, which will be
discussed in part 4. Now, symmetrically, Geoffrey adds another part of the
same model of neuron network to make the learning process completely
unsupervised:


figure 2
Then he feeds the network all of the 28×28 pixels images of numbers
without telling it what number it is. Amazingly, the network learns the
numbers’ image and abstracts the brain states all by itself, then reconstructs
a “better” image at its symmetrically output part. Geoffrey publiced another
paper at 2008 which he shows a t-SNE system that uses this kind of
mechanism to classify 2500 english words, and produce a shocking image
that english names, locations, verbs, nouns are being gathered to different
areas, gradually.

 
Part II
Our ancient ancestors never stop seeking the origin of the consciousness,
thoughts, spirit and soul. Various kinds of hypothesis are made. Some said
it’s from heart, some said it’s from eye, and finally by the advance of
dissection, logic and philosophy, we concluded that it’s our brain. Now we
have X-Ray, MRI scans and even invasive brain computer interfaces that
captures neuron signal directly, and scientist from different area started to
cooperate and understand the brain like never before. Neuron firing patterns
are being recorded, synaptic plasticities are being observed. With large-scale
model of brains being constructed, the technology enables the possibility to
simulate the brain in super computers. (Eugene M. Lzhikevich, Gerald
M.Edelman, Large-scale model of mammalian thalamocortical systems,
2007) So far, we have already successfully reproduct known types of
responses recorded in vitro in rats in computer simulation. With the
ambitious perspective, couple computer industry giants starts to get on the
race of the brain simulation. IBM lanuched blue brain project at 2005, and
now they have already successfully delivering a data-driven process for
creating, validating, and researching the neocortical column. By the Moores’s
Law, our computer power will be able to simulate more than 100 billion
neurons by 2015, which is as many as the human brain have.

 
Part III
Ever since Sony lanuched the project folding@home, distributed computing
or cloud computing have became well known by it’s affordable super power
gather from the mass. Trying to study protein folding mechanism and
ultimately find out the solution to protein mis-folding which leads to cancer,
the project indeed attracted millions of people to participate, and now it’s
one of the most power computer of the world. If curing cancer is that
important and all human should participate, why not a fully simulated brain
model, which will also potentially cures dozens of diseases and more? A
company from Toronto called “intelligent realm” is trying to break the 100
billion neuron barrier earily by using the power of distributed computing. The
project “100 billion and beyound” is simple. People go to their website,
download a 7mb small program which will run at the background, only takes
up CPU horse power that you don’t need. The prgram simulates some
neurons and lets the scientists behind tests different things with them. The
program is open source and they will public their research paper once a
while to keep the public interested. Although I suspect that they will really
create a “thinking machine”, the experience of such experiment is valuable
indeed.

 
Part IV
Before we discuss the defination and meaning of “meme”, we should first
talk about the replicator. Susan Blackmore states that why the universe is
not created by god and intelligent design theory is false science, is because
our world is created by a mechanism called “replicator”. Darwin discoried the
theory of evolution, which also dis-proves the god’s existence. But that is
not all of it. Susan states that the theory of evolution has three properties:
1. Something replicates itself; 2. Something propagates its properties to
others; 3. Something will be selected by the environment so that only the
better version will survived. This three properties make sure that anything
has them will get evoluted and become better and better. (Susan Blackmore,
Ted Talk, Feb 2008) Then we can look at the definition of the meme: it’s a
knowleadge or culture idea that passes through social activities such as
imitation between people, or internet communities, from one brain to
another brain. And teme is a abbreviation of tech-meme which its contain is
not brain but some machines, like artificial neuron networks. They also have
the three properties Susan stated at TED. Now let’s get back to the “brain
state” in part I. As Geoffrey’s model proved, information holds a abstraction
space in neuron networks, which makes one criticism about the existence of
“meme” invalid. The criticism argues that ideas do not hold any abstract
space in our brains, and Geoffrey’s experiment showed they do. So we can
say that the function of the human language is to carry “brain states”, or
“memes” around. The debate is, with A.I technologies adance enough to
decode brain states to memes directly, will language be still as necessary?
Will individual brains able to connect to each other directly? If it does, where
is the position for individual consciousness? How will our society changed
with such high efficiency of communication that is almost equal to telepathy?

 
Conclusion
From part one to part four, we can see that the simulation of human brain is
most likely to happen eventually. And with such technology, mankind no
doubt will have the chance to overcome its biological limit and start to
evolute into interstellar species. However, there will be potential unseen
dangerous ahead of us, that we can feel it coming, but don’t know what that
is. Nevertheless, the future for mankind looks bright and happy, and we
should be thankful that a new age is almost rised from the horizon.
Work Cited
1. Geoffrey E.Hinton, Learning multiple layers of representation, Trends in
Cognitive Sciences, Vol.11 No.10, University of Toronto, 2007.
2. Eugene M. Lzhikevich and Gerald M. Edelman, Large-scale model of
mammalian thalamocortical systems, The Neurosciences Institute, San
Diego, December 27, 2007.
3. Intelligence Realms Co.Ltd, Artificial Intelligence System Technical Paper,
http://www.intelligencerealm.com, 2008.
4. Susan Blackmore, Memes shape brains shape memes, Behavioral and
Brain Sciences, November 21, 2008.
5. Susan Blackmore, Memes and “temes”, TED talk, http://www.ted.com/
index.php/talks/susan_blackmore_on_memes_and_temes.html, Feb 2008.

Advertisements

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: