[{red & blue} ca] - A New Kind of Science: The NKS Forum

A New Kind of Science: The NKS Forum

Pages:1

{red & blue} ca
(Click here to view the original thread with full colors/images)

Posted by: inhaesio zha

each subsequent column shows further history, there are gaps in time between the columns

Posted by: inhaesio zha

in attached .zip

Posted by: inhaesio zha

some more of those guys

Posted by: inhaesio zha

more examples

Posted by: inhaesio zha

...

Posted by: inhaesio zha

from 206849 in one of the sets above

Posted by: inhaesio zha

{surprising and cohesive} forms from 74824

Posted by: inhaesio zha

more of the same from 21186

Posted by: inhaesio zha

I would say that the surprising and yet still cohesive forms shown in the three previous images are qualitatively more surprising than cohesive forms I have seen in our other CAs and CA-like systems.

Whereas there is surprise/unpredictability in the position, size, and timing of cohesive forms in other of our systems I've seen, there has almost always been a crystalline regularity of the forms themselves (the triangles always have the same angle at their vertexes, the gliding forms are always replicated as carbon copies).

With these examples, we are farther toward the {surprise} end of the {surprise-cohesion} continuum, while still maintaining an undeniable cohesion.

Posted by: inhaesio zha

...

Posted by: inhaesio zha

[won't upload...not sure why]

Posted by: inhaesio zha

...

Posted by: inhaesio zha

..

Posted by: inhaesio zha

These histories were from homogenous systems whose cells pay attention to a 4-cell subset of an ant neighborhood.

Each cell follows the same rule as all the others, so they're homogenous in that sense...but the rule shape that each cell is using to calculate itself changes through time, based on local criteria.

The way the rule shape changes is that each cell keeps track of the number of times each cell in its possible rule neighborhood has changed states through time...so, given a 3x2 area in this case, each cell keeps track of how many times each of those cells in that area have changed state...and, in each moment, the cell decides what shape of (in this case 4) cells to apply the rule to, favoring the cells in the rule area that have changed states the most. So the cells prefer to take their inputs from areas in the rule template that are providing it with the most active input.

I think that's analagous to how we ignore input streams that are repeating themselves too much...either at a large-scale conscious level or perhaps at the neural-network level as well.

I had good luck generating conversations between two neural net-like entities named red and blue about 6 years ago, and a key component of how their brains worked was that components of the brain network preferred to be connected to parts of the brain network that were providing them the least repetitive input...so it was like a binary neural network whose cells reconnected themselves based on the variety of the output stream provided by each cell's neighbors (candidate inputs).

I just applied that idea to 1d CAs to make the {red & blue} CAs shown here. A whole slew of system types is suggested by this: systems whose cells use the history of their neighbors' output streams to selectively listen to (or use as input) certain of their neighboring cells. As mentioned, the systems here are homogenous; it might be interesting to apply the {red & blue} non-boring-input-selectivity concept to heterogenous CAs or other types of systems. And of course, a next step is to look at similar systems which use variously-constrained subsets of various rule templates: like systems that select {3} cells to pay attention to from a neighborhood shaped like {whatever}. Or maybe the # of cells a particular cell is paying attention to is variable, determined by a ,,variance/non-boringness,, factor such that perhaps no cells would be considered valid subjects of interest for the cell in question, or perhaps all the cells in the template would be paid attention to at a certain time...in a case like this, would the cell choose from different functions based on the # of selected inputs, or would it feed valid inputs and dummy inputs (all zeroes, for example) into a function of a fixed order, determined by the # of cells in the rule template? Or: we could look at various ways of determining that an input stream is interesting...the simple {more variety is more interesting} method I'm using in these examples is perhaps ridiculous!--at the very least, we should look at cells who have a limited memory of their neighbors' state histories...in these examples the length of the cells' memories of neighbors' state changes far exceeds the length of the system run (which doesn't seem very "local" to me).

David Jones has suggested trying a totalistic rule with this type of system, given how (A)NNs work.

This type of selective listening is present in consciousness-level processes...perhaps it is present in the elementary processes of our world.

Posted by: inhaesio zha

My friend Ashley was just telling me a story of her friend's niece, a four-year-old. The four-year-old pretended she was a monster and growled; Ashley gave her a big reaction, as we often give to four-year-olds because we know it will cause them delight. As four-year-olds often then do, the young niece repeated her faux-monster behavior ad infinitum, over multiple visits. The delight of this wore off for Ashley more quickly than it did for the niece, and the delight of it eventually wore off, albeit much less quickly, for the neice.

So, two things: 1) the tolerance for repetition decreased over time in both of the individuals. 2) the tolerance for repetition was greater in the younger individual than it was in the older one.

Maybe it would be interesting to apply these concepts to selective-listening-type NKS systems.

Posted by: Lawrence J. Thaden

It calls to mind thresholds in sensitivity.

Posted by: inhaesio zha

attached

Posted by: mdmd

great share , Thank you for this wonderful illustration

Forum Sponsored by Wolfram Research

© 2004-2013 Wolfram Research, Inc. | Powered by vBulletin 2.3.0 © 2000-2002 Jelsoft Enterprises, Ltd. | Disclaimer
vB Easy Archive Final - Created by Xenon and modified/released by SkuZZy from the Job Openings