My Blog 08-December 2018

This blog will change over time and hold blogs removed.

Current Blog

I am currently not working that much with picture recognition, I have been spending my time writing software that analyses information. 

What is interesting is that I have been using Artificial Technology to achieve this to date.  What is interesting unlike these huge systems that Google and others are using to create such systems on huge networks my software runs on a simple laptop computer.

1) What does it do?

Quite a lot actually but just to name one thing I found interesting is the identification of nouns, verbs, pronouns, adjectives , adverbs, etc.  What the software does is given any text it just works out what parts are nouns, adjectives or verbs etc.  It continues to create lists of nouns, verbs, adjectives, pronouns. adverbs etc. Some would call this thinking, actually what I find interesting myself is that it does this much better than I can do it. Its accuracy levels seems to be about 100% and if for any reason it is uncertain which is not that often it will ask me what category that word belongs to.  Most of the time I check against a dictionary or Google just to check that what I think is correct.

I have now linked this to others who have POS big systems, to confirm my results.  This is even more interesting because I am networking different systems together.  Two narrow POS system, I have found give 100% confirmed accuracy.  This is with two POS systems.  Average humans would I guess get about 40 to 60 percent accuracy.  I say 100% because all the words I have checked using Google search or dictionaries have been correct it maybe that it is only say 99 percent accuracy with two competing POS systems put just say you add a third or fourth then I would think it would get it right 100%  without any argument or doubt.  But unless I find some words in the wrong category, I have to conclude its currently about 100% accurate.

2) Is this system more intelligent that a human

Yes again its artificial intelligence therefore like a calculator does math better than humans now A.I can identify nouns, verbs adjectives,  definitions etc. much better than a human.

3)  Do you need big neural networks to achieve this

Actually surprisingly NO, you can do this all on a laptop with the software I have developed.

4) Is your software in the public domain

No, its not for sale or in the public domain and I have no intension of putting it in the public domain currently.

5) Why not put it in the public domain

The software is very powerful indeed,  mainly because it does not need large networks of GPUs or CPUs .  I am amazed at what the software can do.  The truth is I do not know what I will end up doing with the software.

6)  Why does it not need large amounts of GPUs or CPUs to run the software

Its because its been designed to work more like a human being, its does not need huge amounts of training data that needs huge amounts of processing power.  It uses traditional computing techniques to create a system that works using information that a human would hold.

7) Does it use a neural network?

No, it does not use a neural network, let me say neural networks are great for things like picture recognition, not I think so good for this type of application, its needs a more human approach, which I have called the Yes, No network.

8) So how does it work?

It works like humans work if you think of a Car, you think about associated things.  It is that simple it just thinks like a normal human thinks. 

9) Was it difficult to program?

Yes, I think of it as an Artificial Intelligent Engine.  Its taken a very long time to program the engine and I would be the first to admit that its been quite a challenge.   Most difficult was performance improvements.  It can still get better parallel streams maybe help improve speed on new multi core laptops.  I am currently using a old laptop that is not very fast.

10) What language was it programmed in?

I used Java for one simple reason I wanted the code to be portable and if necessary scalable.   I wanted to be able to run it on Windows, Linux or any other platform that Java can run on using virtual machine technology.

11) What do I currently us it for

I use it for legal research purposes mainly because my two main interests are law and computing.

12) What is it good at doing?

Its good at finding information given a abstract question.  It uses two approaches one is key word search that seems very traditional but within A.I it find best results. For example if you have million pages of data with the word car in it you want it to find the appropriate car not just list a thousand links to the word car.  But also it goes further it actually uses inference like a human can do. So it will also return information relating to the initial search you gave it with no key words but that maybe still appropriate for your search.  

13) Why does it need to use inference?

It uses inference because in law and other things you may not know the correct question or key words to give the system.  Therefore over time the Artificial Intelligence within the system I have written uses inference to help get the correct information.

14) Is inference hard without big data?

The amount of data I use to create inference is very small, humans do not use billions of lines of information to create inference what they do is use targeted small amounts of linked data to create inference.  This is what my system is attempting do achieve.


15)  What do you like about your A.I Engine or System?

I like just watching it learn, its just amazing how it works out things and how it produces results.  I just love that its like a human learning a little bit about the English language every day.  

16)  Why are such systems important?

Its important because for example there is just two much law out there.  I am astonished that it finds new law that I have never heard about before or did not know that it even existed.  It helped me find related law going back to the 1800's I did not know related to current law and cases that I would have never known existed.  It also points me in legal directions I would not have ever considered or thought could be relevant to a legal position.  It reminds me of  the latest Chess computers that seem to give insight into new ways of playing chess.  Its a little like this in my view when thinking about law, it enhances your knowledge base and makes you aware of different arguments.

17)  Final Thoughts

The idea that someone playing chess on his own is always going to be beaten by someone who plays chess and uses a good chess computer.  Therefore my knowledge is Ok but given my knowledge and a good A.I System  combined is going to produce better results just about every time, even if my thoughts and the AI's thoughts are different or the same. This can often lead to new outcomes.  AI helping humans, intellectually or otherwise should always produce better results and that has been my experience at first hand programming an A.I. Engine.


For those who are interested in its ability to learn I watched the increase of words:

Analysis
Notes:
Zero means no data taken.
Time line is left to right but no time taken
This is a snap shot in a relatively short time frame
Notice that there are more then less then more this is down to that it is learning and if it realizes that for example a noun is not a noun but an adjective it gets removed. 
Cleared words, only when words are confirmed by the system that they are an actual category they are then considered perminant.
Permanent words with memory dislodge will be allowed in the future with a type of negative data re-enforcement to be implemented in the future. 
Lets take Nouns we had less accurate data at 4213 about 70% but this turned to 99.99% or 100 example accuracy at 4104 with none wrong to date found. If found could reduce to 99%.
Importantly very little mathematical modeling is required for the Yes No system to work well.  We humans do not use calculation to recall memory or put another way we to say pos 3 * pos 4 * layer 8 I store the thought there.  We just store it, we do not have to calculate where.   Also limits to any architecture are limits to the low level materials used.


Nouns = 4176    = 4213 = 4214 = 4220 = 4222 = 4208 = 4241 = 3914 = 3995 = 4104
Verbs = 0           = 1273 = 1282 = 1285 = 1287 = 1289 = 1299 = 1021 = 0 = 1159
Adjective = 0      = 695 = 771 = 721 = 724 = 727 = 766 = 954 = 0 = 992
Adverb = 0         = 0 = 693 = 667 = 668 = 673 = 706 = 676 = 0 = 685
Conjunction = 0   = 111 = 114 = 115 = 115 = 115 = 118 = 138 = 0 = 149
Pronouns = 0       = 36 = 36 = 36 = 36 = 36 = 38 = 40 = 0 = 50
Definition = 0       = 21 = 23 = 23 = 23 = 23 = 23 = 22 = 0 =
Names = 407      = 509 = 509 = 514 = 537 = 537 = 52 = 0 = 590
Interjection = 0    = 0 = 78 = 78 = 93 = 89 = 91 = 0 = 99










General Picture Recognition Software