Re: Neural Networks

[Judith Rosen <***>]

David Macy wrote:
> <x-tad-smaller>I feel that even many neural network approaches are likely to also be off the mark.  Mind and body are inseparable.  I now feel that if the mind is to be understood then this will likely entail moving the object of study from mind to brain and body and then on to life and organism. </x-tad-smaller>
>

I agree with you, on all counts. In fact, that's why my father ultimately settled on the task of figuring out why organisms are "alive" (from a scientific point of view); the answer to that question would be necessary to illuminate the behavior of living organisms in other respects, the behavior of communities of organisms, the behavior of organismal capabilities like intelligence and creativity, etc. The unexpected part of it was that the reasons organisms are alive (relational causality at work in a particular type of complex system organization) turn out to illuminate way more than just biological systems. Relational causality is actually a property that is a basic aspect of the universe and yet our entire scientific paradigm actively ignores it as,  literally, "immaterial".

However, if we analyze the human mind as a complex system then we have some interesting insights into its behavior. For instance, the mind is a separate anticipatory system contained within the first, but I would hesitate to call the mind "a living organism" in its own right. It's sort of a unique entity, as a system. We define ourselves by it, and yet, it's almost a parasite of the living body, in a way... the expression "some people are their own worst enemy" gives some hint as to the reason why. It's not their body that is being referred to, it's their choices, their behavior. Suicide is the ultimate example. Brain chemistry interacting with our experiences and thought processes can make a perfectly healthy, intelligent, perhaps even supremely talented human being decide that "life is not worth living". Clint Eastwood put it rather well in one of his movies (I think it was "The Outlaw, Josie Wales"): "Dyin' ain't hard for men like you and me, it's livin' that's hard, when everything you've ever cared about has been either butchered or raped." 

D.M.: 
> <x-tad-smaller>I "built" several neural networks (simulated) and ultimately came to the conclusion that, though the networks did in fact do something akin to learning, the networks themselves (as opposed to their synaptic weights) were unentailed.  How do we account for the "architecture" of any given network?  I also now believe that if learning (which seems to me must involve anticipation) is to be embodied in computational models that this absolute distinction between hardware and software will have to be circumvented.  Anybody got any ideas?</x-tad-smaller>
>

How do we account for the architecture? My answer would be to say that if it exists, then it must be entailed. In other words, it's not "an accident." But to find out what the entailments are, a whole lot of relational study needs to happen. Studying the architecture, alone, isn't sufficient. That would be sort of like trying to learn about a system called "chocolate cake" by studying some intermediate phase in its creation. We once had a discussion about organization, using chocolate cake as a metaphor, I recall. I said adding chocolate to an existing cake would not give us a chocolate cake and adding chocolate indiscriminately to some part of the process would likely not achieve the desired end result, either. It's entirely relational and it's organized into the cake from the beginning, via a set of relations. Similarly, there are many ways to create a chocolate cake, using different relations, different processes, different time constraints, but there is no way to get from one pre-existing type to the other. No amount of adding or subtracting aspects of the pre-existing cakes will transform them into one another. It has to be organized in from the start.

Just as an aside.... I seem to use a lot of cooking and gardening analogies. Most men use sports or hunting analogies, I notice. My father tended to use chess and poker analogies as well as the typical ones, including boxing, but he also seemed to have a whole bunch of uncategorizable ones. I had to ask him to explain some of the typically male analogies because I couldn't get the phraseology at all... There was a movie we were watching in which one guy delivers what was obviously a put-down, but I totally didn't understand why. The line was; "I crap bigger than you." My father laughed at it, so I asked him to explain the significance of it. He said it meant "You're so small and insignificant that my crap is more significant than you are." My interpretation had been; "I crap bigger than you do," and I wanted to know why that would be something to boast about!? 

I remember a farming analogy from Dad once; he quoted "Nature abhors a monoculture"... where that quote came from, I don't know. But I don't think it was his own phrase-- the word "abhors" wasn't one he used much. He was talking about disease vectors in densely populated human communities. It's kind of interesting to me that our species always seems to prefer monocultures. We actively create them or seek them all the time, don't we? Some people actually seem to abhor diversity, in fact! They act as if they feel threatened by it, at some gut level of being an organism. In any case, the reason Nature seems to abhor a monoculture, in my view, is because of the inherently interactive, relational nature of the universe. A monoculture is a system which has entailments based on its organization that don't exist in a differently organized system. That's all. A whole lot of something all in one place can be utilized as convenient raw material by some organism, for example. That's the case with a disease epidemic.

Anyway... to get back to the point: Your comment about the distinction between software and hardware was very interesting to me. With the mind/body interface, for example, the mind's intellectual activity would seem to represent software, and the brain would be hardware. But in this case, the software can build new hardware: new neural connections, new information, new experiences, new chemical inputs, the list goes on... Learning a new language can change the structure of the brain. So study of the capability of learning and the questions about architecture will need to be considered and pursued with this information... um... in mind, as it were. (Sorry, I couldn't resist!)

Judith

Web address: http://www.rosen-enterprises.com
BioTheory: An electronic journal of general science based on the Relational (Rosennean) Complexity Paradigm
On Oct 28, 2005, at 7:58 PM, David Macy wrote:

> <x-tad-smaller>Hey Jamie and Judith,</x-tad-smaller>
>  
> <x-tad-smaller>    Interesting posts.  I studied neural networks for several years as opposed to what's now called Classical AI, which is I suppose that disembodied super-intelligence that you refer to sometimes Judith.  People who gravitate towards neural networks are, I agree, much more likely in my opinion to stumble upon hidden entailments.  Especially if they are simultaneously looking at studies of real brains and their functioning.  I know that I did.  I spent several years thinking that the classical AI guys were way off, still do.  Now I feel that even many neural network approaches are likely to also be off the mark.  Mind and body are inseparable.  I now feel that if the mind is to be understood then this will likely entail moving the object of study from mind to brain and body and then on to life and organism.  Convenient huh?</x-tad-smaller>
>  
> <x-tad-smaller>I "built" several neural networks (simulated) and ultimately came to the conclusion that, though the networks did in fact do something akin to learning, the networks themselves (as opposed to their synaptic weights) were unentailed.  How do we account for the "architecture" of any given network?  I also now believe that if learning (which seems to me must involve anticipation) is to be embodied in computational models that this absolute distinction between hardware and software will have to be circumvented.  Anybody got any ideas?</x-tad-smaller>
>  
>  
> <x-tad-smaller>David</x-tad-smaller>
>  
>