Okay, I think I see what you're getting at. The issue with "decentralized"
anything would always become one of speed of transmission, hence the
limitations of materials and available power. Perhaps then, if we adopted a
perspective of layers, where each layer could theoretically represent a
particular context, as the intent and purpose of the layer. Let's assume such a
layer could be fashioned into a computer chip.
The architectural question would then arise as to which priority of all
priorities of a context would be aimed at by such a chip? This would be
relevant because an adaptive, complex system would probably had multiple
priorities of different "agents" within a single context. Based on a
methodological systems and analysis design heuristics, the potential number of
total priorities (as points of decision making, or gates) for a single,
normalized, decomposed context would be no less than 524,288. Again,
heuristically, at least 4 such sets of priorities would exist within such a
context = 2,097,152 minimum number of gates.
My new research on complex, adaptive systems indicates that the systems context
could be triple this size, which in exponential terms would scale to many
millions of decision gates for a single layer of the of the overall system.
Therefore, your point on compressionability is most relevant.
The 100 billion star problem in computing, I think it is called that, may be a
realistic approximation of the scale of the challenge in processing, let alone
design, faced by quantum computing. I include this perspective for us to keep
in mind as all elements of such a system should probably be of its optimal
compression value, meaning as short as effectively possible.
In theory, we could perhaps start by extracting certain architectural
principles for such a machine, e.g., Manage processing for, and on 100-billion
objects in near real time.
Such a challenge as this would provide some guidance as to the possible
characteristics, which would have to be ascribed to other elements of the
system, e.g., optimal state architecture and so on.
I may have diverted somewhat from your good points, but perhaps this is exactly
the type of hybridized thoughts that might emerge from this discussion, I hope.
> From: a...@listbox.com
> To: a...@listbox.com
> Subject: RE: [agi] Couple thoughts
> Date: Tue, 24 Feb 2015 15:22:22 -0500
>
> > -----Original Message-----
> > From: Nanograte Knowledge Technologies via AGI [mailto:a...@listbox.com]
> >
> > If I understood the first part of your thought correctly; agents could be
> > collaborative-type of quanta forming some kind of generative frequency,
> > relative to the properties of the agent form?
> >
>
> Well, yes having many similar agents communicating and coordinating via DNA
> quantum communication mimic in design and software. I know it's not
> determined yet how this works biologically but it doesn't have to be there
> is enough evidence and knowledge to work it that way and perhaps there are
> indications it will produce results - this model is a form of combined
> analog and discrete computation. Analog frequencies from discrete agents
> generating hybrid patterns...perhaps patterns of adaptive entropy within its
> operational complexity.
>
> > On the second part; each agent should be a hard-coded fractal, with the
> > scalability property being a replicated constant?
>
> Uhm - possible. I'm thinking agents would have complexity somewhere between
> that of a cellular automata and a simple unicellular organism, towards the
> lower end of course. Core defined by sets of mathematical tuples initially
> for example a finite state machine automata tuple, but having advanced
> dynamics of communication complexity, peer to peer, group and cluster
> forming capabilities...Could they be fractal based and maybe fractal based
> CA? I don't know. Tuples allow a vast spectra of higher order interlocking
> automata categories. But perhaps fractal based should be explored for the
> potential of recursive self-improvement.
>
> > Some type of soft-coded, management system gets all the pieces working in
> > the purposed direction?
>
> The structure needn't closely mimic biology, for example every agent needn't
> carry around it's DNA there can be a community copy. Agents might only carry
> changes for resource optimization. Oher biological pieces can be thrown out
> as not needed in software world like cell walls. And importantly a cell
> needn't be a "cell"...
>
> > Last, I noticed your point on "feedback".
> > If open-looped (probably recursive), closed-looped (probably self
> recursive).
> > To be learning, one would probably require both?
>
> The structure of individual agents should be flexible enough so that it
> communicates and incorporates systemic behavior. That would include feedback
> mechanisms. A main idea would be that it's self-organizing ability need not
> be biologically restricted but restricted along the lines of contemporary
> computer based technology. But like you say the system, aside from
> individual agents can be recursive that's interesting.
>
> BTW I'm just throwing around some ideas here on a complex adaptive MAS that
> has numerous similar agents. I'm sure there are many systems that could be
> built differently with just a few agents having highly specific functions or
> one agent with many different internal functions. I do suspect the many
> similar agent model could have more adaptive utility but I could be wrong I
> just don't know.
>
> John
>
>
>
>
>
>
>
> -------------------------------------------
> AGI
> Archives: https://www.listbox.com/member/archive/303/=now
> RSS Feed: https://www.listbox.com/member/archive/rss/303/26941503-0abb15dc
> Modify Your Subscription: https://www.listbox.com/member/?&;
> Powered by Listbox: http://www.listbox.com
-------------------------------------------
AGI
Archives: https://www.listbox.com/member/archive/303/=now
RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-f452e424
Modify Your Subscription:
https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-58d57657
Powered by Listbox: http://www.listbox.com
