flat assembler
Message board for the users of flat assembler.

Index > Heap > A speculation about Artificial Intelligence

Author
Thread Post new topic Reply to topic
Sean4CC



Joined: 15 Apr 2015
Posts: 14
Sean4CC
An animal cannot blend responses to its environmental situation. It cannot flee, attack and remain motionless at the same time. It must decide a clear sequence of states to enact.
You have a messy analog brain that must respond like a digital algorithm.

Next you have computer scientists trying to model the biological brain with soft computing methods that try to approximate the analog aspects of the brain with digital computer code.

You have digital code trying to approximate a messy analog system that is trying to be digital in nature.

In such a regime the estimates of the computational power required for simulating the human brain (for example) are going to be astronomical.

Of course the short cut is to do everything in the digital domain in the first place.
How can this be done, especially given the highly dimensional nature of the input data (such as visual data)? In the 1970's J.S Albus showed how the dimensional reduction could be done by simply using a hash table. However his work was incomplete in a number of ways. Likely due to the lack of reasonable computing resources at the time.
These days there are some additional tools available to help with dimensional reduction such as random projections. There are also greedy policies such as feeding in only a few bits of information at a time into the hash table, that results in decisions being made using only the minimum amount of information necessary (concision). To this there are various forms of reinforcement learning that may be applied or other ideas.

It is perhaps resonable to speculate that a simple animal with a few hundred neurons could be modeled with a hash table of a few 10's of thousands of bytes. A human with as little as a few 10's of gigabytes?????
Post 17 Apr 2015, 17:28
View user's profile Send private message Reply with quote
Sean4CC



Joined: 15 Apr 2015
Posts: 14
Sean4CC
Maybe there are some simple ideas here:
http://www.freebasic.net/forum/viewtopic.php?f=7&t=23309
Anyway I'm taking a break from writing code for a year. Especially anything to do with AI.
I'll go and do some engineering because you know engineering is cool.
Post 18 Apr 2015, 01:35
View user's profile Send private message Reply with quote
pabloreda



Joined: 24 Jan 2007
Posts: 100
Location: Argentina
pabloreda
I'm not study your code but perhaps have similar point to HTM from http://numenta.org/, you know this algo?

You have any site with the code? or all is in the FB forum?

numenta has some interesting test wih images sequences, predicting the next image, you can do similar test?
Post 18 Apr 2015, 02:37
View user's profile Send private message Visit poster's website Reply with quote
Tyler



Joined: 19 Nov 2009
Posts: 1216
Location: NC, USA
Tyler
Have you heard of neural networks? Or this (http://en.wikipedia.org/wiki/Blue_Brain_Project)?
Post 18 Apr 2015, 04:25
View user's profile Send private message Reply with quote
Sean4CC



Joined: 15 Apr 2015
Posts: 14
Sean4CC
The argument was that the biological brain is a crude and inefficient effort at implementing some digital algorithms with messy analog wetwear. I know some (sic) about neural nets. If a neural net is to switch cleanly between different states (as it must for survival as was said) that implies positive feedback and other things that are likely to require large numbers of neurons to do simple things. Well switch is a digital term.
I'm not really providing code though there is some on the FreeBasic site. I'm keeping a low profile these days.
I don't know too much about what numenta are doing currently, I guess I could check.
More broadly I am just pointing out that you can use memory to do things that are impossible with even trillions of machine instructions like add, multiply ....
It is possible to have thousands of 1 megabyte look up tables these days. If you start using hash tables in smart ways you can do some far out things.
Looking forward 5 or 10 years from now there will terabytes of non-volatile solid state memory that you will be able to access at DRAM channel speeds. One thing that will have to change if memory based computing becomes a big thing is the cpu cache system. At the moment it wastes large amounts of energy if you are doing random lookups in large tables. Without a cache, memory based computing would seem to be extremely energy efficient.
Post 18 Apr 2015, 14:36
View user's profile Send private message Reply with quote
bitRAKE



Joined: 21 Jul 2003
Posts: 2914
Location: [RSP+8*5]
bitRAKE
The brain seems to learn to learn. The senses clearly do some processing in the organs before forwarding information to the brain. Storage and retrieval seems to work in parallel across many temporal scales, and senses.

Definitely want to look at compressed sensing. Information is costly in terms of energy (transmission and storage). Complexity arises from energy contention in nature - the acquisition and storage of resources.
Post 19 Apr 2015, 10:50
View user's profile Send private message Visit poster's website Reply with quote
Sean4CC



Joined: 15 Apr 2015
Posts: 14
Sean4CC
Well actually I did look at compressed sensing. I kinda figured it out in an intuitive way. In terms of actually doing compression, you are not going to get a whole lot. As per usual they (ie the CS crew) are going completely nuts with the maths while not getting a engineering grasp on what is actually happening. The "maths" rather than the "why" if you like. Well, that's the way to get a good salary review in a university these days. What is useful from compressed sensing is the use of random projects as a means of mixing large data sets while preserving fully the information in the data set. That is one thing I can say google actually stole from me while I had information on www.code.google.com. I say stole because they never gave any credit as to where they got the information from. You can see some of what they are doing with it in their "FastFood" paper: www.cs.stanford.edu/~quocle/LeSarlosSmola_ICML13.pdf
One use is to feed a small number of bits of information at a time into a decision making process until it can just make a clear decision. You can create a "greedy algorithm" for learning and it helps keep things concise.
A hash table is also a sort of random projection if you think about it.
Post 19 Apr 2015, 23:59
View user's profile Send private message Reply with quote
bitRAKE



Joined: 21 Jul 2003
Posts: 2914
Location: [RSP+8*5]
bitRAKE
One could argue that the brain exists so that an animal may preform all the possible responses to a situation. ['All' in terms of that animals understanding of the world.] Of course, you are specifically speaking of the external action taken by an animal in an attempt to reduce it to digital. Which is a massive top-down assumption. The animal is analog-to-analog: every external response is different than any other.
Post 21 Apr 2015, 19:17
View user's profile Send private message Visit poster's website Reply with quote
Display posts from previous:
Post new topic Reply to topic

Jump to:  


< Last Thread | Next Thread >
Forum Rules:
You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot vote in polls in this forum
You can attach files in this forum
You can download files in this forum


Copyright © 1999-2020, Tomasz Grysztar. Also on YouTube, Twitter.

Website powered by rwasa.