Intellectual Property From 2008 to 2019
We need General AI to solve
issues and problems that are not to complex for humans to solve but just too
vast. Computers can crunch data way better than any human. Now
with Artificial General Intelligence available right now if we only network
our narrow AI's together (see below) we can achiever more in the next ten
years that we have achieved in the history of mankind.
AI and General AI
We already have narrow artificial intelligence
right now it is a data driven system that uses data and humans coding
systems to achieve narrow artificial intelligence. But to get to
option one true general artificial intelligence we are missing one step,
that step is not just neural networks that work off data. What we
actually need is "SELF CODING SYSTEMS", therefore we need systems that can
code themselves this will lead to true general intelligence. This is the
best option to have artificial general intelligence. It will produce
intelligence that is very much like human intelligence but better. To
achieve this we need to start the process of building self programming
systems. The second option for general intelligence, that will
not be true general intelligence but near is that humans all over the world
create bits of code that are narrow and link this to a data system that is
more like general intelligence. I would think that in the short term, option two
will come first and we will have general intelligent systems made up of many
narrow artificial intelligent pieces of code that used together will mimic a
form of general intelligence. However the future will be super general
intelligence that can only work perfectly if we create self coding systems that for
example can change the structure of a neural network on the fly.
Building these systems once the ground work is put in place for example
developing (compiler interpreters) that work by generating new code because
of data. What I am suggesting is that these compilers or interpreters
will not be rule based but will be driven because of the data it perceives.
This means that the self coding system will structure and write its own
neural node type procedures (code) because of the data from the neural
network. This will solve at lease one problem that of back
propagation. This means that the code structure can change over time
and improve its own structure driven by the data that it perceives. Importantly
the code is produced because of the data input.
This is the way forward but there is a better way, but that better way is
the introduction of a defined biochemical modules that may even form
consciousness however this is not a road I would like to go down because
this road could really lead to an artificial intelligence that is unstable
big question is will self coding systems that are super intelligent be
something that humans control and the simple answer to that is no in the
very long term. How
can you control something you do not understand is the simple answer.
Not being able to understand super intelligence is not a bad thing if it
provides a cure for disease. If it wishes to destroy humans can it be
stopped in the very long term the simple answer will be no, but for this
to happen you will need systems that not only self code but self reproduce
their physical and biological state. We consider physical state to be like humans but
in my opinion it will be any physical state that accomplishes a task best.
You have a conflicting issue allowing the true general intelligent system to
do it's own thing so it cure's disease or attempt to limit its self
programming that will mean it never become true super intelligence.
But for now we are quite safe to get general or super intelligence you will
need self coding systems and for a dangerous autonomous system you will need
some type of biological element for any of these system to have any type of
consciousness. Not to say that narrow artificial intelligent systems
can not be dangerous but this will depend on the will of the persons
controlling those systems. A data driven system will never be enough
for true artificial intelligence it will at best be a combination of narrow AI
systems. Tue artificial intelligence will need at least self coding driven
by data input and that will be the trick to a new type of AI system.
Importantly self coding systems will never be able to be fully debugged.
The code will interact with data and combined will not be debuggable
again if we attempt to debug it we will have to limit it because its code
and data stream will be far to complex for any humans to understand or
decode fully. This is not to say it will be impossible to understand
the bigger idea behind some of its code but that will be about it if we wish
for it to have true automomy.
I started thinking about
General Picture Recognition in the early 1990's and had always considered
whether it was possible to create software that up until late 2000 was
always concerned with recognition of a particular thing using mathematical
points mostly. For example using many different points on a face to
try and identify that face. But while all this was starting to take
off I was already thinking about how every single object could be identified
within a picture and the context and relationship of each object to that
picture or another way of putting it how it was positioned within any one
About 2008 I put my first attempt of General Picture
Recognition onto the Internet with software I called General Picture
Recognition Software General. Since that time everybody
including companies like Google have started to believe that what would have
once been considered impossible that a computer could recognize any object
within any picture may not be as impossible as first thought.
We have developed some
interesting concepts in "General Picture Recognition" and they are listed
below. Because we have noticed the proliferation of American Patents in
Software and in particular picture recognition Patents that are not the same
in the UK. We therefore have listed our intellectual property
1. Auto Tagging using our unique concept of
something similar to Neural Networking concepts but very different.
This concept does the following it compares two very similar whole pictures
and allows each picture to be manually or automatically tagged by the user. Once we
have manually tagged a number of pictures we then allow our picture
recognition software called GPRSG Software to take known information from a
picture that is already tagged and then use that picture to identify similar
pictures and tag that image. This process is iterative the more
images tagged the more images can be tagged given a selection of already
tagged images. This concept belongs to us and should not be copies or
used in any other software product without our consent.
interpretation uses our unique concept to remove all background within a
photo and allows the selection of only the main object through colour
selection taking any colour range and incorporating the whole colour range
into a single colour. This concept allows us to pull out an object
from any picture and without the need to know every rotational shape using
colour alone we can identify an object within reason of any size within any
picture. We call this our "User Defined Custom Colour Object to Search
Analysis" because by the user selecting colours within an object it will
allow them to find the same object in any other picture even if there is a
change of size and shape because of any sort of rotation.
3. Object storage uses our unique concept that allows object to be
stored in their own file and tagged so that any picture can use any object
within any file to build up a description of that picture so that the
picture can be described in its own vocabulary or language using a simple
word or phrase search with tagged to file; with in-depth descriptions and
links. Therefore a file may hold an
object called helicopter and that object file is tagged internally with the
description and linking mechanism and file content that allows any object in
any other image to be compared with the object in the file containing the
object. We use the same colours to identify the same object in other
pictures even if the colour changes because of the way light hits that
object when it is in different positions because of any rotation of that
object vertical or horizontal or both.
4) Using a method
that outline a change of strong colour or when one colour is changed
to another colour. We have a unique method for achieving the outline
5) Multiple methods to identify what one picture
contains. The main concept is to identify the difference automatically
between an object and what is actually background. We identify
background by working out what is an object. Objects always being
something that is between the background and each background is a layer of
the object. Using this method we can work out what is actually the
object and what is the background using software alone.
Ideas on Neural Networks
The traditional approach to Neural Networks
is creating Nodes that represent Neurons. Each Neuron can be changed
mathematically until the concept works. My new theory is only
different in the following way what if the computer "CREATES A NETWORK OF
NEURONS USING DATA IMPORTANTLY NO NEURONS EXIST UNTIL THEIR CREATED USING
DATA) this is something that I am currently working on.
I started this work in 2008 but came back to the idea in 20017 however have
been playing around with these idea since early 1990's. If it
is possible to create a computer programmed Brain using data then maybe we
will be moving further towards general AI or general picture recognition.
What is important in this new concept is not to have Nodes or Neurons
created by the programmer that can be acted on by working out the lowest
error rate to achieve an action like the traditional method. What if the data takes over
this operation and only the data is changed when an error in the data is
found. Once found the data is updated and we then have a
computer that now acts like a human Brain. This would also eliminate
one big issue with current thinking on AI and that is that all of these
current Neural Networks need a lot of training and when they are wrong nodes
or neuron weightings need to be changed or more neurons need to be added to
the system. This is certainly not like the human brain where new
connections to Neurons and new Neurons seem to be created through life.
In my theory of general AI the program that acts on the data is equivalent
to the brain structure and operates are differently depending on the
structure, for example speech, hearing or seeing. In my new theory the
current thinking on Nodes or Neurons are replaced with data that importantly
represent the original data. The best AI systems will be the system that are
designed with a programmed structure that best implements the data as
neurons. The most important aspect of this type of data driven neural
network system is that the design is general. The big advantage of
this is that the brain structure can be programmed and improved over time so
it continues to get better. But the brains patterns are not created
through code but is pure data. The way the data is stored, analyzed and
retrieved can use traditional programming languages but the data that makes
up the neurons mean that we have true Artificial Intelligence.
Decisions made by the system will depend on how the data driven neural
network nodes are organized. This organization will be critical to the
decisions the AI brain will make. What I have not made clear is
that this data can change over time to put right wrong decisions as the
brain learns and new neuron or neuron connections are made. Being able
to change a data driven neural network will be critical for such systems to
work well. Therefore as the AI brain capacity increases the AI system
should learn more and be able to rectify mistakes in data made when it did
not have as much knowledge. This I would think is a mirror to how a
human learns. For a simple example if I learn to spell "speach"
like this although it is spelt wrong because everybody spells it like this
"speech" until I realize that I have spelt it wrong I will always spell it
that way. However once I use a spellchecker I realize that I have spelt it
wrong my brain then swaps out speach and creates speech. To do this it
usually moves speach to a lower priority and supersedes with speech being
higher priority and therefore "correct at the moment." This is the key
to learning that of "currently being right" this may change over time when
new data is learnt. This is important because the human brain like a
good AI system must have a brain that can change opinion over time. This is
the area of my current research that I am interested in. I believe that such
data driven neural network systems may take computers far beyond current
human intelligence in that they will be very much better than humans in very
many areas. These new AI system will even be able to produce or create
theories like Albert Einstein's thought about but done using data driven
neural networks. What I think will be interesting is will humans be
able to understand why such theories are being defined and how those
computer driven theories can be implemented. Will the human be
intelligent enough to interpret such computer defined theories, I think I
will leave it as an open question to be discussed by others.
of these concepts are our intellectual property rights and these ideas
should not be reproduced in any other software without our permission.
General Artificial Intelligence
Maybe we already have it, maybe
its more about linking up that intelligence and maybe that is the big
problem. Maybe its company competition that is the issue. All
that Artificial Intelligence is; is computer code and data. Many
problems have already been solved for example I understand that object
recognition, is mostly solved or is as good as human level object
recognition. Therefore if you had a way of passing the problem
around to the AI that deals with that problem then job done. WE
NEED A SET OF SIMPLE GENERAL STANDARDS THAT EACH AI CAN RECEIVE AND
SEND. PASSING ON PART OF THE PROBLEM ON TO ANOTHER AI, THAT DEALS WITH WHAT
IT IS GOOD AT AND PASSING THE OTHER PROBLEMS ON. The mechanism for passing
on the problem is really the only thing that is stopping general AI right
now. We have robots that can walk we have algorithms that can see objects.
what if we just passed data between algorithms. It does not matter how
complex the processing is as long as the input output data or algorithms to
be passed follows a universal standard that is very simple so that anybody
creating an AI can easily follow those standards so that their AI
can become part of the general artificial intelligence. It also solves
the so called control problem, because no one person or group can control
all parts of the AI. This idea is the intellectual property of myself
One main concept or aspect of creating these STANDARDS is
something I have defined as GAIAG or general artificial intelligence
agreement general. What does this mean, today we have many people
working on many different A.I problems and many working on the same A.I
problem. These algorithms are getting great results but its been
pointed out that not all results are perfect. Take natural language
what if you have several natural language A.I algorithms that could take
results from each other and form a consensus. For example if five agreed and
two disagreed then it would go with the five natural language A.I's that
agreed. If all disagreed with each other it would do exactly
what we would do and ask a human for the answer, that answers would then be
feed back to all the A.I's so that they could learn and improve. This
feedback loop is important because it could allow A.I's that had the wrong
answer among other A.I's for example natural language learning to be
improved. My concept of general artificial intelligence agreement
would mean that one A.I can pass its results onto another A.I using; receive
and send general standards. Say you have several A.I's designed
by different people or companies they all sign up to the GENERAL A.I
STANDARDS once signed up and on the network they would form a comprehensive
insight to a particular area for example our natural language results if you
had 20 A.I's designed for natural language that looked at the same problem.
Remember this is general A.I therefore if the problem was object
recognitions the General Artificial Intelligence would just pass that object
(photo) or (video) or (sound) to the set of A.I's that deal with that issue.
Results are returned after General Artificial Intelligence Agreement within
a particular area back up to the main General Artificial Intelligence Brain
General or GAIBG for short. This combined information is feed back to
the USER who maybe anybody on the internet. Future expansion means
that the STANDARDS FOR SENDING AND RECEIVING INFORMATION MUST BE SEPARATE
FROM ANY ONE A.I INPUT OR OUTPUT. What I mean is that its the job of a
person or company who creates an A.I to translate its input or output to
conform with the GENERAL ARTIFICIAL INTELLIGENCE AGREEMENT GENERAL, and to
those standards. Also with this paradigm redundancy is not an issue new
A.I's in the same area or importantly new A.I algorithms created for the
first time in completely new areas would act like a new skill that any human
would learn. Some A.I systems may fall over or stop being developed. Like
with any node system if one node falls of the system it will just jump to
the next A.I. However for such an A.I system to function large A.I systems
like Googles speech recognition would have to be available. Therefore
for such an A.I system to work well you would have to have the combined
co-operation of big companies like Microsoft, Google and Facebook to name
just some. This general A.I system I am discussing here
would not just deal with data driven issues but for example with powerful
General Artificial Intelligence Agreements General, these standards could
mean that a Robot made from any material and hardware would just have to
SEND and RECEIVE information from and to the General Artificial Intelligence
Brain General that would feed back everything that Robot would need to move,
walk and talk in the real world. For example if the robot is looking
at something, that data stream would be sent to the General Artificial
Intelligence Brain General or GAIBG for short. it would be sent to the set
of A.I's that deal with video and interpreting objects in real time, at the
same time the sound would be sent to the A.I's that deal with understanding
sound and interpreting speech or sounds these would be sent back or even
sent to a set of A.I's that deal with combining sound and picture data to
give and send back to the robot via the General Artificial
Intelligence Brain General results. Therefore the robot can
importantly (think) what I am seeing for example is a room with seats,
tables and chairs but what I am hearing is the sound of the sea through an
open window. This information can be bounced back and forth to
the General Artificial Intelligence Brain General to maybe a set of A.I
(thinking) algorithms. So if the robot is asked the question what are you
thinking? The answer would be, I am thinking about the room and listening to
the sea through the open window. Maybe the next question from a human
is how can I use the space in this room better. This leads to the
final aspect of the GAIAG, and its standards, algorithms that are not
considered to be A.I as long as they comply with the A.I agreed standards
can still form part of the A.I system. For example an algorithm that
uses space aware mathematics to produce the best layout for a room may use a
standard algorithm. But the data is feed back up to the A.I Brain so that
the person who asked the original question about using the room space better
would get a reply from the robot like let me print you an alternative room
plan that better uses the space within this room. You may reply OK or No, if
OK the Robot may connect to the printer and prints the improved plan.
Ultimately if we are still dealing with robots we could say something like,
Robot can you run down to the local shop and pick up a pint of milk for me.
The robot may ask which shop, having used the standards to connect to the
General Artificial Intelligence Brain General. The human may reply to Tesco
or Asda etc. The robot would get the latitude and longitude from the General
Artificial Intelligence Brain General and then proceed to the shop to pick
up the milk. One important aspects of the General Artificial
Intelligence Agreements General, the standards is that not only data but
code or algorithms can also be requested. For example a Robot may not be
able to be connected to the Internet at all times therefore core A.I
algorithms can be run actually within the Robot without any Internet
connectivity but connection to the General Artificial Intelligence Brain
General can be made at any time when and if required. In summary
the system I am proposing does not have to be an all singing and dancing
system all at once, it should be flexible and allow for any future
technology. The standards must allow for any new technology and any
current or new types of programming language or hardware like quantum
computers. Therefore the standard must be very separate from any
software or hardware. It will be down to the software or hardware to
comply and produce input and output that complies with these standards.
The input data and output data to and from each A.I therefore would have to
be independent of any data type it would just be a data stream.
Therefore one very useful A.I would be a stream identifier. This would
learn about the data streams so that the General Artificial Intelligence
Brain General could send that data to the correct set of A.I algorithms that
deal with that data. This idea is the intellectual property of myself and
GPRSG. If anybody like Google, Facebook or Tesla would like to contact me
with the idea of developing these standards into a working model I would be
interested to help any large teams on my idea for the development of A.I
General Artificial Intelligent standards. One final thing that is very
important in my opinion and being a computer programmer myself. Input
and output should be a single line of code that can be placed in any
programming language. The longer I program the more I realize that keeping
it simple and easy for all is the key.
For example it
should be just one line of code :
Receive stream = anything(send
stream, any parameters required by standards)
The Sender and Receiver
Stream format can be defined in a set of parameters that the General
Artificial Intelligence Brain General can understand and also those
parameters can be set so that a certain format is returned within the
Receiver stream. so it can be used by the local algorithm.
because these standards are completely separate of hardware and software
requirements system like IBM, DEEP MIND and all other A.I systems can all
work together for the good of all given the standards and idea proposed.
With these standards we may have true General Artificial Intelligence
General or GAIG within a couple of years.
This idea is the
intellectual property of myself and GPRSG.
Machine Learning a Simple
but very Real Idea
I would like to make it clear current computer
materials will never be self aware. If you made the systems biological
then we are biological therefore in theory anything is possible. But
silicon will never be self aware on its own, but biological and silicon
But systems that imitate human behaviour is very possible.
I am going to be very controversial in what I am about to say but this
just maybe the truth, what if machine learning or general intelligence can
not be mathematically proved. We have always associated
mathematicians with thinking.
Albert Einstein for example was good at math's but I would say was an
amazing thinker. What if thinking not Math's answers the question.
What if we can simplify machine learning, what if HL = (N - Y), what if this
is the only equation you need to work out general intelligence. HL stands
for human level intelligence N stands for No and Y stands for Yes.
Intelligence is one single decision at a time either Yes or No. The
minus sign is important because all previous No's that do not exist or exist
give rise to a possible Yes and all Yes's that do not exist or exist give
rise to a possible No. What if our brain is simply a parallel
decision maker that produces one thought at a time. Put another way, I
may think that a Bishop in chess can move only in straight lines, Yes this
is my truth, but then having watched many people play chess I observe that
Bishop moves diagonally. Therefore No a Bishop does not
move in straight lines but Yes it does move diagonally, I have learnt.
The idea of big data is therefore "True" but we have made a fundamental
mistake. Humans get big data through the whole of their life, we see,
we hear sounds we smell, we touch and feel objects to get texture.
Our learning is continual, therefore when we see something we have usually
seen that a thousand times, when we hear something we have usually heard
that noise lots of times so it is familiar. That was a dog barking Yes
we think, but if it is a noise we have not heard before we think No, its not
a dog or wolf or cat so we eliminate all known possibilities. We may
never know what that noise is because (N - Y) have eliminated all possible
ideas. Now two things happen we wait for the noise again and use new
information to make sense of that noise. We walk by a paint pall
center near my new house. Next time I hear the noise I associate it with a
paint ball being fired from a paint ball gun. Hear it again I say Yes its a
paintball or someone tells my what the noise is for example I talk to my new neighbour and they explain the sound, he then tells me it is a paint ball
gun being fired.
Each human can only learns so much I may
understand about computer programming but I do not know about building a
Television. Therefore general intelligence and super intelligence is
already achievable. The human brain is only asking the same simple
question, is it Yes or No. Shall I get out of bed, shall I brush my
teeth, shall I wash my hands, shall I comb my hair, Yes, No.
intelligence and super intelligence, I have written a program that does POS
(Part of Speech) works well but then connected it to other programs that do
POS and the system is already way better than I at identification of POS.
I just asked the different programs if they agreed and if so update POS data
set. Each system on its own got some POS wrong but when you combine
them and allow them to learn from each other we now have super intelligence
in the POS domain. Super intelligence is just a matter of linking all
of these systems together (See Above Talk on networking these systems)
from system that analyse the voice to system that hear to system that talk
to systems that read and many more.
The next challenge is not just
creating more sophisticated machine learning programs (systems) but
HOW WE NETWORK THESE SYSTEM TOGETHER SO ALL CAN TALK TO EACH OFTHER AND PASS
DATA BETWEEN EACH OTHER. Then we will have super intelligence.
The super intelligence will not be something that destroys the world but
will be something that humans can use to improve the world I hope if it is a
Machine learning systems have more power
than we really yet understand, if you combine and network these systems,
they will be able to answer and achieve technologies we have not yet fully
realised. If we network these system to talk to each other they will
grow faster and be more powerful than we could have ever thought possible.
This is not fiction but reality, we under estimate this technology at
Join what we now call narrow AI to other narrow
AI's in a network of different types of narrow AI and you will see beginning
of the future I see. The next break through in my opinion will be a
breakthrough in joining different AI systems together, once this is achieved
you may start to get super intelligence faster than you think.
joining narrow AI systems (programs) together will lead to computer systems
understanding the real world and unless they understand the real world, they
will always be less than they can be. To achieve modeling the real world we
will need technologies that provide data to any part of the AI system of
collections of narrow AI with much more data that we currently provide.
Technologies and chips that can read different smells, associate heat with
objects and feed importantly that data back into the AI system and much
The brain stores data - computers store data - therefore if we
think of each narrow AI as acting as a group of neurons and each narrow AI
gets and passes data onto other neurons on the Internet anywhere else in the
world. Input equals data stream, with Yes Agree or No Disagree, if all AI's
agree or majority agree then Yes confirmed is a Car or confirmed is a dog
barking or confirmed is the smell of fire. Once we have this level of
modeling the world we will soon have super AI because it will be better than
humans in most every domain humans have ability within.
this very interesting are narrow AI neurons programs that will accept
input from many narrow AI neurons lets call these general AI neurons that
will have the ability to analyse many types of data and many of these
general AI neurons may collectively produce new concepts and new ideas that
it would be impossible for any human to have enough data to formulate these
These theories can be given to a lot of very
clever humans who may imagine new technologies or cures. This is the
future of AI I now see and envisage. Who can stop this future, well
surprisingly business who refuse to pass data between the AI systems or
allow their AI system to talk to other AI systems.
To all AI
developers the most important point I can make is that data from vision,
sound, taste, smell and touch are stored differently within the brain. Its
that stored data that the brain works with not the external data. Put
another way you see the Mirror on the wall. You close your eyes and can
still imagine the mirror on the wall that's the brain working from stored
data because you know longer see the real mirror because your eyes are
closed. Its the stored data we need to pass between Narrow and General
The General Artificial Intelligence Brain General is the
most important node on the system, it is the node that provides human like
thought, or makes a decision once all notes feedback to it, their results
just like the human brain with neurons. All bits of information sound,
vision, taste, smell etc. for a single spatial time slice would
be allocated their unique code, also the General Artificial
Intelligence Brain General main node would have a copy of this unique number
and therefore could make sense of all sub nodes that are passing that
spatial time slice of information after processing back to it. Once
this is achieved we then have General Artificial Intelligence working at the
same level of the human brain, its only a case of sub nodes improving to
bring us to Super Artificial Intelligence what I mean by this is that a
machine can do any human type process (intellectually) better than humans.
There is nothing wrong with using neural networks, and a combination
of other types of networks to achieve a imitation of the human brain, as
long as they are connected to each other; my proposed network. Time will go
by and better methods maybe found to improve the General Artificial
Intelligence Brain General sub notes. For example back propagation within
neural networks may no longer be required to train a set of neurons. The
system may see inefficiencies and self improve.
Can we create such a General Artificial Intelligence Brain General today?
The answer is YES if all interested parties work together. There is
nothing wrong with companies using thousands of GPU'S AND CPU'S but I have
developed a self learning data driven neural type network connected to other
neural networks that works well on my home very ordinary laptop.
can process up to 5000 lines of text within about 2 minutes at its fastest
speed but this is reduced when it is in learning mode. In learning mode it
is actively attempting to learn new POS or (Part of Speech) but if it is
just using already learnt speech yes it can run and analyse text much faster
than any human could possible achieve.
Current barriers I have
encountered, there are very few others who are willing to share their POS
data from their machine learning systems without wishing to charge a fee.
The small amount of organizations that do usually have limited POS data. I
would not be surprised if I do not have some of the better POS data sets
through my AI training. But organizations that have huge POS data
sets are just not going to share those data sets because they are worth to
much financially to openly share their data. Actually they use data to
feed their AI systems so sharing that data would not make sense for them.
There is a difference between companies using tools that access data via
some sort of neural network and say giving everybody that data to help
create new neural networks systems. It will just not happen so the
best solution is creating a neural network made up of many different types
of narrow neural networks that can in compose any other self learning
software even if not based on a neural network. The main idea as above based
on sub networks in the same area agreeing (discussed above).
my main point, you do not need big computer or the cloud to build your own
SELF LEARNING PROGRAM, I have shown that you can build your own.
However these programs are not easy to build and take a huge amount of time
to build a good narrow AI program. But once built they achieve what no
normal computer program can achieve. Having built one myself I really
can see why AI programs are way more powerful than normal type programs.
I can not get over how they self learn, it sometimes feels like these
programs are living systems in that they keep on improving the more data
I am interested in legal data, I am interested in
the idea that if you need AI to analyse legal documents from Acts to Case
Law to Journls or discovery document type data, these can be bunched up and
pulled out and provide the same relevant legal point from many documents
that even a well trained solicitor or lawyer would not consider. If we could
do this in say 10 minutes this could save someone spending days or even
weeks attempting to find that same data. Even if you considered it in
another context what if a 10 minute search found important information that
a lawyer missed after they did their legal research or discovery
On average computer hold large amounts of un-organised data.
Interestingly I have found legal information I did not no about and I would
think my legal knowledge is quite wide. Case law means you can home in
on cases you would never have considered or which are not precedence but
really give a quick understanding of why that precedence for a case evolved
from or an idea that improves on or may change precedence for a certain set
Why not do a simple traditional search, the main issue is
context, this can not be improve using a traditional search, with AI we are
attempting to pull data in with similar context , also these results must be
returned first and bunched together. The more learning the AI does the
better the context and better results.
What I have been working on
is the idea, that although we can input large amounts of information. Learnt
data set representing brain memory is only 71,950 KB currently on my biggest
data set of learnt data. The human brain when it sees a car it does
not go through every car you have ever seen in your entire life to say that
it is a car, this would take too long even if your brain is doing a type of
parallel processing. This is one of my big thoughts on reducing AI
data down from huge data sets to much smaller data sets. We only need
to find the closest match and put those close matches together to have much
smaller datasets. I think AI will have to move in this direction, much
smaller data sets mean much quicker time and results.
downside of any AI system is that it will need to learn, Another way
day 1 of using an AI will give no better results than a normal search, it
will know nothing. After day one it should always be better than a
All AI systems are really search systems, from self
driving cars to NLP or natural language processing they all do the same,
some do their search using a hidden mechanisms for example self driving cars
search against patterns, while NLP may search to translate language or
speach. It is not surprising human vision, smell taste is a searched
representation of firing patters of neurons that at the highest level is a
YES or No option. We may colour it using different language but its
just a single thought that makes a decision at any single point in time and
This idea is the
intellectual property of myself and GPRSG, if you use any of these idea in
any talks please acknowledge your source as (GPRSG).
This next topic
is over the top but may well be true, what if parallel thinking uses some
type of quantum relationship in the brain. If time space and parallel
thinking spark thought (all done at once theory) maybe this is why it must
be biological to have true thought.
Computers may come close to
thought but will always be an imitation because they lack true
parallel thinking. Like nothing so far breaks the speed of light test
what if the parallel thinking to produce a thought test only applies to
THE I KNOW TEST
The I know test is an
important connection concept, I know and pass it on and the I do not know
it on. Let me give an example of this idea. I am sitting
in dark light but I can touch type I press the wrong key, I observe that I
have pressed the wrong key therefore I know I have pressed the wrong key so
one part of the brain sends a message stating that I have pressed the wrong
key and the key I need to press is next to the wrong key that I actually
pressed. The final message is sent that adjusts my finger to move one
place over so I press the correct key. Without the I know Test we
cannot pass the message on to the correct place.
This is what I think the brain in doing in humans, but if
not its the best way for a computer to do it to get near human results.
I think all information is grouped in the brain, I am not saying it is
grouped in closeness or any other definable way however it maybe to some
extent. But what I am sure about is that the brain if its as random as my
brain can do this, Train equals track and many other things. Track
equals train (the double take test) and many other things. What is nice
about this is that it is really easy to do in computer code. It is one
of the easiest ways of allowing a computer to think, given a small amount of
data initially. Trains can be associated with passengers so equal to
passengers. Therefore trains run on tracks and have passengers.
We are building a way of identifying words with areas of thinking.
What is amazing about these methods is that all this can be done on a
standard laptop. This is because we are not working with large data
sets, just lots of very small data sets, that grow and learn just like we
Modeling the World
Modeling the word is just stored data,
so we do not have to do the whole prediction on the fly. Modeling the world
is not the whole answer but it is the search technique used with modeling
the word is moving closer to answering certain questions of general
All this is good but just connecting all the current
narrow AI solutions together on its own will be so very powerful.
This idea is the
intellectual property of myself and GPRSG, if you use any of these idea in
any talks please acknowledge your source as (GPRSG).