A new approach to Artificial Intelligence

muscleblouseΤεχνίτη Νοημοσύνη και Ρομποτική

19 Οκτ 2013 (πριν από 3 χρόνια και 9 μήνες)

91 εμφανίσεις

A new approach to Artificial Intelligence


There are HUGE differences between brain
architecture and computer architecture



The difficulty to emulate one system on another
system increases linear with the number of
differences in architecture between the two
systems.



The brain is an interactive information carrier.


Computer Emulation Issues:



Making compromises in reduced neuron complexity


Simplify the thousands of connections between neurons


Simplify sensory devices (camera no match for the eye)


Learning by feedback
-

a computer has no hardware for
measuring timing differences between millions of input
pulses. Learning is compromised


No computer is capable of emulating the entire human brain in real time at
this present time



Simulation of the brain on an IBM
Blue Gene /P supercomputer:


Model 1.6 billion neurons:

0.15% of real time

0.5% of capacity


* Source: IBM publication “The
cat is out of the bag”


Intelligent
robots?

FICTION


C3PO is an
intelligent,
interacting
robot….


R2D2 shows
initiative and
determination



To create this type of behavior would take
approximately 4000 times the performance of
current super
-
computers:


IBM Blue Gene/P


emulation of 1.6 billion neurons
and 8.9 trillion synapses at 0.15%

of real time. Ref.
IBM publication ‘The cat is out of the bag’


IBM Blue Gene/p


emulating 900 million neurons and
9 trillion synapses at 1.2% of real time



Human brain: 88 billion neurons and an
estimated 100 trillion synapses at 100% real time




REALITY
:


Violin

playing
robot

(© Toshiba)







Intelligent
robots?


Recreating intelligence by program control
involves a fundamental error in thinking:


The brain is not an analysis and control system


The brain is the only intelligent entity that we know
off, consisting of memory and integrating cores


Learning is a fundamental function of the brain


Artificial Intelligence systems need to learn to control
rather than being programmed to control


INTELLIGENCE is defined as the ability to acquire
knowledge and to subsequently apply this knowledge,
to combine it to form new solutions…


“Artificial Intelligence is the ability to perform
a task that would require intelligence if done
by a human being”


Minsky


We only use 10% of our brain (anonymous)


Bigger brains means smarter, and humans
have the biggest brains of any animal


We have to emulate the entire human brain
to
create intelligence



A new way to do Artificial Intelligence


The brain is an interactive information carrier.
Information is stored in synapses


Synapses are everywhere in the brain, therefore
memory is everywhere
-

Fundamental


Learning is essential, it is how intelligence emerges
and how the synapses are ‘programmed’, within a
DNA
-
built Framework


A digital architecture that is similar to the cells that
make up the brain, with analog synaptic parameters
expressed as multi
-
bit digital values


Wired as a cortical column, a single chip contains
10,000 neurons and 500,000 synapses




Consists of digital cells and synaptic memory


A Synthetic
Neuro
-
Anatomy will function like
the brain when its neural cell emulation is
sufficiently realistic and the network is the
same as the biological model


The model needs to operate at real
-
time
speed to allow learning to take place, in sync
with the real world around us


Learning is the method by which a neural
network acquires function




How To Create An Artificial Brain