My new A.I. idea

[Black Rose]

To see where this is going I will give you some back ground.

First when I was 12 I went to middle school and read at least 16 books on A.I and came up with this diagrams including this one.

Spoiler

Also I began thinking about cartoon digimon and game boy Pokemon.

In the last 2 years of high school I tried programing a neural network in Java with time hierarchy's by the book On Intelligence by Jeff Hawkins 2004.

Now for my idea.

A dodecahedron truing machine with 12 inputs and 12 outputs this way the configurations can fit in 3D space like a be hive. Second they reconfigure their output like a brain cell by deciding to put the 12 outputs where ever they want such as 12 on one pentagon can be done. They modify their output so they will receiver lest heat to distribute the energy received son that next time the receive an input it will be less radiant.

Time variable allow for inputs and out put to evolve with a algorithm to evolve algorithms inside the dodecahedron turning machine.

Entropy is the key factor in evolution, to evolve such as inside a video game require the amount of dodecahedron to be vast depending on the number of inputs and to replayed evolution game play many times over. The more you have the more goo to liquid then gas it becomes in the heat distribution inside layers of brain matter dodecahedron molecules.

The limit on computation is the limit of simulation 10 kilobyte turning machines but as they are all symmetrically be processed fast by a GPU.

EDIT: My 2007 report

http://www.challenge.nm.org/archive/06-07/finalreports/48.pdf

EDIT2:

As dodecahedron evolution algorithms select different shapes for the network in a batch of organisms trained to "survive" in trials where heat is the function of bad outcomes. The brain model uses self assembling axon structures via reduction of entropy by forming tunnels protected by their surroundings.

 

[Cognisant]

You read 16 books on AI, and that's what you came up with.

Wow, just wow, a slight breeze could bowl me over, my mind is blown.
That may well be the most amazing thing I've ever experienced.
OMGWTF, dude.

This thread has made my day.

 

[Artsu Tharaz]

You read 16 books on AI, and that's what you came up with.

Wow, just wow, a slight breeze could bowl me over, my mind is blown.
That may well be the most amazing thing I've ever experienced.
OMGWTF, dude.

This thread has made my day.

Useless response, disregard.

Can anyone breathe some life into the idea?

 

[Black Rose]

What does the Dodecahedron shape have to do with the computation?

It is the most efficient 3D volume compaction unit most resembling a neuron.

What to you mean by "Input"?

Who is they? Why do they want things?

dodecahedron have 12 sides so the surface area is the input / output.

What is the "goo"? "Liquid"? "Gas"? What is a Dodecahedron molecule?

A drop of watter is sticky a bowl of watter is wet so the more dodecahedron's you have computing the more powerful the new axon algorithms generated to disperse heat away from inputs.

The skull is a certain size perhaps a mouse for a Nintendo Wii and a fly for a game boy system

 

[Cognisant]

Hardly useless.

A drop of watter is sticky a bowl of watter is wet so the more dodecahedron's you have computing the more powerful the new axon algorithms generated to disperse heat away from inputs.

What is this, chemistry? Metaphorical? Bullshit?
It doesn't make any sense, either Animekitty is high or just trolling us.

I was responding in kind, y'know speaking troll to troll.

 

[Black Rose]

Hardly useless.


What is this, chemistry? Metaphorical? Bullshit?
It doesn't make any sense, either Animekitty is high or just trolling us.

I was responding in kind, y'know speaking troll to troll.

Metaphorical

what else does it matter being a troll is pointless.

I am not stupid I just represent ideas differently than you.

 

[Artsu Tharaz]

I just represent ideas differently than you.

Too differently, unfortunately. Don't take it to heart.

Mediating representations are needed first. ENTP perhaps? I'm unsure.

 

[Black Rose]

Too differently, unfortunately. Don't take it to heart.

Mediating representations are needed first. ENTP perhaps? I'm unsure.

Him or me?

I get the feeling that MBTI is not what we thought it was.

Jung was both Te and Ti I've seen heard of.

perhaps my loss of ego blends the unconscious with the conscious doing odd combinations.

 

[Cognisant]

No actually I understand perfectly.

What I don't understand is the relevance of a geometric dodecahedron, why having 12 inputs/outputs is superior to any other configuration, which even if I did understand would all be utterly moot because there's no explanation of the mechanics behind how these virtual neurons (I assume) decide where/why to allocate their outputs.

You say it has something to do with heat, but that's only a metaphor, for what I don't know.

Heck what is the basic unit of data being used here?

 

[Artsu Tharaz]

Him or me?

I get the feeling that MBTI is not what we thought it was.

Jung was both Te and Ti I've seen heard of.

perhaps my loss of ego blends the unconscious with the conscious doing odd combinations.

I meant an ENTP could be used as a mediating representer. Ne>Ti will be able to fill in the missing space better than Ti>Ne.

I'm unsure of how ego-death affects functioning. I suspect it may break the boundaries between function use, giving rise to undifferentiated function use. Observation of those with complete ego-death may shed insight.

Transcendental type, no doubt. But can the transcended interact properly with those who haven't? Insight needs to flow outwards gradually until it solidifies.

 

[Black Rose]

No actually I understand perfectly.

What I don't understand is the relevance of a geometric dodecahedron, why having 12 inputs/outputs is superior to any other configuration, which even if I did understand would all be utterly moot because there's no explanation of the mechanics behind how these virtual neurons (I assume) decide where/why to allocate their outputs.

You say it has something to do with heat, but that's only a metaphor, for what I don't know.

Heck what is the basic unit of data being used here?

In the blue brain project the simulation looks like swarm behaviors.

trainable Turing machines Hopfield Net trained by annealing(dodecahedron) modify rule sets when cellular automaton patterns enter them making the rules sets for those automatons.

 

[digital angel]

Interesting..haven't read it all yet.

 

[simichem]

Very cool... An idea really worthy of attention.
I'm planning to study chemistry and although AI isn't my cup of tea, this actually sounds like something theoretically AMAZING, plenty of nano and supramolecular chemistry could be added to the mix

 

[Black Rose]

After consideration I have come to the conclusion that geometrical attributes I specified are no longer needed for the network typology.

This is a discussion I had with a friend on Facebook recently.

Spoiler

can networks of code be manipulated from a visual interface where Links can be tested for desired effect?

Hmmm... very possibly. To find out, one would have to precisely define exactly what is meant by each of the terms you are using here.

Link in network is the input output between blocks of code. An avatar is run by an A.I. program that evolves by simulation of possible brain configurations and self models what feedback happens between code blocks and environment. Logs are stored of configurations for future reference. Interaction/development happens in social setting/games in physic simulation enviroments.

Yes, this is on the right track. Could you create a simple instance of such a network? Or, create an unambiguous diagram, and label the components, how they are connected, and how they interact?

What kind of math is needed for representation changing network typology? Like links on a webpage(node) Feedback by annealing (positive negative energy), links are finite say 100 per node but depending on feedback change address. Also self organizes to form loops within loops.

There are several appropriate choices. A Markov Random Field is an obvious choice, but that's kind of complicated. The simplest natural method would be a semantic graph with dynamically updated annotations. Another approach would be to use a hierarchical hybrid machine, where to bottom layer consists of simple decision engines, and the top layer is an adjudication layer; this general approach is sometimes referred to as "bagging", because you have a committee of "thinkers" that pass recommendations to a "higher authority" which is responsible for making the final decision.

Have you thought about creating a *precise* description of the problem you want to address? That is an essential step.

From our interaction I have formulated a new idea how an AGI can be created.

For the model of the AGI's cognition a "Markov Random Field" will be used consisting of ten million nodes, 100 links per node. This is a temporary limit/memory restriction of a gigabyte. These nodes will have an update time (spiking rate) about the same as a human 100 Hz. The links will be either inhibitory or excitement. If a link is weakened sufficiently by feedback(hebbian learning) it will be reassigned to a different node.

There will be a virtual environment where the AGI's will reside. Interaction between agents and humans will train them on becoming intelligent. A files system will save link configurations for testing in different templates of AGI's.

 

[walfin]

Eh. You are looking for visualisation, no?

Well as a start I suppose you could represent each neuron in the net with a sphere/cube and draw lines between them?

I suggest you just lay the spheres out in a grid first and draw the lines for a first prototype. That should be quite easy although it won't look pretty.

You could have an edit mode with the UI on the side of the window for the user to key in the numbers when they click a sphere.

The user can fly around (usual stuff). Some simple OpenGL/Open Inventor should do this.

 

[Architect]

I've glanced through what you have here and can't make heads or tails of it.

 

[Black Rose]

I've glanced through what you have here and can't make heads or tails of it.

@Architect post 14 is the most recent update and is very straightforward.

 

[Cognisant]

There is nothing to make sense of, it's a trollish game.

 

[QuickTwist]

Hi Animekitty,

Well the first thing you have to do is make sure every thing is in working order. That is not an easy step but one that needs to be done. Once you have everything in working order both the theory and the script, you need to test it. Testing it also might need to be done in order to make sure everything is in working order. I understand you have this written in Java code right? What/how have you transferred this information to a usable fashion/a device that can execute this code?

It seems as though there are some mathematical complications that you need to work out first. For example, in your frequency and amplitude diagram you have to understand that this is going to create a sign wave instead of a half of a sign wave (unless of course you are using an inverter [I suggest {if you are using an AC power source} using a transformer to convert to DC power so you don't have to over complicate it]). When Amplitude as a source is applied, it works in tandem with the frequency to give the difference in peak to peak value. Because the period is the inverse of the frequency, you will want to know what frequency you will be using to account for the time differential and how this plays into working memory.... Err... I think...???

I don't know if that helps but it might...

If what I am telling you is basic then what you are working on is beyond what I can help you with.

 

[Black Rose]

I've moved away from my previous designs of the network and have only the need for help in finding the mechanism for the change function. Each node is like a router that rearranges the links inside it when it reaches a threshold for how much energy the packets are delivering to it. A 3D cellular neural network. The brain grows axons in the configuration it does because of the way energy is flowing through the system. My network would do the same but not grow just rearrange the links inside each node.

This diagram has a fully connected network for each node but each link could be activated or deactivated when the energy flow exceeds a defined amount.

 

[Ex-User (9086)]

Why 8 input/outputs change from 12 i/o? Why not 17 or 255 i/o?

 

[Black Rose]

Why 8 input/outputs change from 12 i/o? Why not 17 or 255 i/o?

This is only an example diagram. It could be a hexagon in 2D or a cube in 3D. What is important is what happens inside each node.

 

[Nick]

Drivel.

 

[Ex-User (9086)]

Are you going to program it so that each of the instances interacts in a way you described, or is it that you want to build entire logic circuits with gates, adding the programming afterwards? Or you have some architectures in mind and you will buy them to assemble?

 

[Black Rose]

Are you going to program it so that each of the instances interacts in a way you described, or is it that you want to build entire logic circuits with gates, adding the programming afterwards? Or you have some architectures in mind and you will buy them to assemble?

It can be implemented in hardware or software. I don't have access to fabrication I will do it in software.

 

[QuickTwist]

Get a vehicle to execute this idea. Without a working frame, its nearly impossible to see how this could be applied.

 

[scorpiomover]

I've glanced through what you have here and can't make heads or tails of it.

I've been struggling here as well. Nice to know it's not because I'm a noob.

First when I was 12 I went to middle school and read at least 16 books on A.I and came up with this diagrams including this one.

The diagram is cool. It's a basic spatial layout of many of the common functions in the brain. The problem with making that into an AI, is how to make the parts. The sensors are easy enough. We can use photo-detectors to make up the retinal cells. The difficult bits are the advanced processing units, that changes it from a bunch of sensors, into a fully-functional AI.

From our interaction I have formulated a new idea how an AGI can be created.

For the model of the AGI's cognition a "Markov Random Field" will be used

Nice idea. Emotions in the form of neurochemicals in the brain, already act similarly to such probabilistic fields, and naturally evolve. If you want to implement an AI, you'll then be giving the AI "emotions". The field can be discrete or analogue. But either way, it will "feel" solutions.

 

[walfin]

My network would do the same but not grow just rearrange the links inside each node.

Not the links outside also?

 

[Black Rose]

These are my new equations for entropy in a network of dodecahedron

in the grid energy input changes how the dodecahedron makes or breaks connections.
when entropy is high it reconfigure it connections for an arrangement that decreases entropy

 

[Black Rose]

f = frequency
t = time

i might have it written wrong by phi in relation to f such that t will take f and at intervals of 1.618 sample it. delta is if a nodes links changes with phi being in the resonance of f. so if f matches phi within the t interval it is broken. phi is the threshold of the node. it makes sense to me because my math is dealing with certain assumptions about nodes links and energy. heat is entropy inducing so phi limits the heat to an equilibrium. adaptation is the nodes ability to form links accordingly

phi is also used as the ratio between nodes the link has attainment of.

 

[QuickTwist]

Hi Animekitty,

It looks like you're making some good progress and I think eventually it will come to fruition. Just keep working on it and it will get closer to completion.

[Edit] Also, Here is a site where you can buy electronic parts. If you can't find what you're looking for on that site, perhaps you could invent your own.

 

[computerhxr]

Why does AI have to be human?

 

[Black Rose]

Why does AI have to be human?

humans are optimized for there environment
a virtual human would be understandable to us
the point is that i don't know what environment the a.i. is in
its input output is all that matters and environment could be 3D or 2D

 

[Black Rose]

Is there any particular reason why you've chosen a dodecahedron? Is there some specific resource limitation on your target platform? I'm seeing a connectivity limit of 5 per node (unless the circles also carry signals) and 8 per cluster. That there are clusters at all, and that there are connectivity limitations to either, seems arbitrary to me unless it's a cost/benefit decision made to fit the system into a particular set of hardware limitations.

The phi ratio is used to calculate the entropy because the dodechedron in the network is the best shape possible for 3D compaction where rotation is the angle of the pentagram. Tetrahedron could be used in a diamond structure but its to uniform. Asymmetry is needed to create loops in the system. Icasohedron is also possible for the grid pattern but it has 12 nodes rather than 20 in the dodecahedron so stores less negentropy. The dodecahedron fits in a 3D space with 12 dodecahedron beside it. Energy input as frequency will allow the network to absorb information reducing entropy in a 3D space like a crystal hologram. It can then use energy to do computations / pattern recognition much like a perceptron but holographic.

The way i have it set up in my model does not need to be a specific typology but this typology is how i started to understand it. Entropy reduction can be done with any form information takes, it is just what kind of representations you use because i am using frequency of clusters which might be inside a 3D structure like a brain. Neurons are 3D but more complex. My model is simple but contains enough structure to reduce entropy in an environment where energy exists. Organisms have constraints of a 3D environment and energy consumption. Those organism change by building structures which links and nodes channel energy and entropy is the new configuration.

 

[computerhxr]

Computers use binary to represent a waveform. Just allow binary options to gravitate towards the closest value based on the waveform instead of depending on a ridged binary system.

The Phi ratio is a waveform (and a particle! Lolz). I use this in poker, psychology, physics, and the stock market. It's the basis of how the universe works in my opinion. However, I believe that it's used incorrectly to explain how things work. This is because we use a binary system to recreate the math and it is no longer a living system. There is life in it, but it can't express itself from a binary prison.

I have a lot of ideas on how to make this work, but I'm not really interested in AI. I feel like AI is inherently a system of thought and non-human. Look into hive intelligence and you will see what I'm talking about. Human is just one expression of thought so it should be added as an abstraction after the components of thought are created and functional.

These are my ideas regarding AI, so I don't think you will find much to support what I'm saying. I would be interested in seeing what would result from a system like this if anyone does create it.

 

[Cognisant]

What does geometry have to do with information processing?

 

[Black Rose]

What does geometry have to do with information processing?

information represents shapes that change, the process has a shape in time so knowing why things happen is dependent on that.

 

[Black Rose]

computerhxr

it may be that a computer is limited in expressing smooth waveform so because organic system are not limited in such a way they can be dynamic in ways that might be qualia. I want to improve how i think so i am meditating but it would be nice if i knew how thinking works based on phi. I think water in the brain is a crystal so it may have something to do with the dynamics of thought. Crystals are seen in New Age culture but they have a scientifically basis of how light sound become standing waves which is that everything can be seen as a crystal. Parallelism is good for increased intelligence because the more you can think of the more you can understand. Phi dynamics could be used to increase harmony so the parts can work together in parallel. A generalized formula could be made but i need to know the application before i make them. A technology for intelligence amplification has been produced already i have seen online but i don't know how it works mathematically. It uses ultrasound.

 

[computerhxr]

computerhxr

it may be that a computer is limited in expressing smooth waveform so because organic system are not limited in such a way they can be dynamic in ways that might be qualia. I want to improve how i think so i am meditating but it would be nice if i knew how thinking works based on phi. I think water in the brain is a crystal so it may have something to do with the dynamics of thought. Crystals are seen in New Age culture but they have a scientifically basis of how light sound become standing waves which is that everything can be seen as a crystal. Parallelism is good for increased intelligence because the more you can think of the more you can understand. Phi dynamics could be used to increase harmony so the parts can work together in parallel. A generalized formula could be made but i need to know the application before i make them. A technology for intelligence amplification has been produced already i have seen online but i don't know how it works mathematically. It uses ultrasound.

Crystals are a geometric representation of phi. E.g. sacred geometry.

If computers took natural bio-metric feedback from its surroundings, then it would be intelligently making decisions (nature vs. nurture). For example if you create an app that changes the music based on a heart sensor, distance traveled, and gyroscopic movements; it would be making intelligent decisions with a limited scope. There needs to be a checks and balances system to make sure that the app improves over time. And if it flowed naturally with rolling averages and volume like the stock market. It would tune itself to the operator rather than being programmed to follow a set of rules.

Creating AI would be 100% nurture. It's unnatural which is the biggest hurdle that I see with man-made AI.

Or just make a computer that randomizes based on the stock market and see what happens. I think that the human context is a leap so making micro-intuitive systems would emulate real intelligence and could eventually expand to human levels.

I consider networks an evolution that is based on necessity so the internet is already a living and intelligent system. It is just overly ridged compared to a natural network like a river that can flow as it pleases. The displacement of dirt and erosion is a memory system that prevents water from continually taking the same path. Without the moon, water would become stale and rot. The universe is one large living system that continually changes to adapt to new conditions.

Darwin has a book on plants that I'm interesting in reading. He alludes to plants having a form of intelligence based on similar rational as I mentioned previously. Plants also grow new nodes at phi. Plants also are controlled by water, which is controlled by the Sun and Moon. The moon cycles in approximately 30 day rolling averages if you consider the moon phased to be a curve.

I'm not really concerned for computer AI, because it will manifest itself out of necessity when the time comes. Forcing it to mirror human intelligence is something that the universe took billions of years to manifest. It's complex and I don't think that humans are capable of creating that level of technology given our current manifestation of reality. Baby steps and practical technology is more appealing to me.

 

[computerhxr]

Maybe you need to start programming for one of these?

http://en.wikipedia.org/wiki/D-Wave_Systems

 

[INTPWolf]

AnimeKitty, you have some truly amazing and intersting ideas, i wish i understood you and your other posts as well as i understand this one(i think i do anyway) but all you other posts make me feel like a total noob at using the brain.
I came up with a very similar idea while back with the help of weed and observing the behaviors of others and my own mind's background processes. Its basically a 3D array of 12 faceted basic binary functions using the 4 basic emotions to calculate thought patterns. Each dodecahedron consists of emotional values, memories, and past positive or negative experiences. As the dodecahedron are activated they total the emotional values of the previous are added and memories are compiled/recollected to create a action/response. The outcome of this at the basic level is recorded as positive or negative in the dodecahedrons, and will determine the probability of using that path again. Theres a lot more to it, phi fit in there somewhere but id have to dig up an old journal to accurately say more.
Or get some weed and rethink it up, but i don't really want to do that, while i was thinking this up i started seeing everything as programing, and i attempted to rid myself of cognitive dissonance and broke my brain somehow. Like force stopping Svchost.exe in windows. I was very quiet for about 2 months, my close friends, roommates, and coworkers noticed this and got really worried about me. It fixed itself somehow, maybe, and im back to normal, also maybe.

Though with a system this complex the only way we can make it think anywhere close to a human is to have it observe every action and every variable that makes us come up with that action. Otherwise it would be totally incomprehensible and nearly impossible to understand, while i think thats super interesting in itself that's not what i'm going for.