Life's purpose is to create Artificial General Intelligence

Got a great idea? Tell the world about it here.

Moderators: mvs_staff, forum_admin, Marilyn

Re: Life's purpose is to create Artificial General Intellige

Postby JO 753 » Sun Jan 28, 2018 5:17 am

Soundz more like an assumption than lojik.

And the notion that if there iz a common core we and the AI are based on, that they woud be comprehensible to us duznt hav to be true.
Tired uv Trump yet? I am:
http://www.7532020.com
JO 753
Intellectual
 
Posts: 611
Joined: Tue Mar 16, 2010 6:54 pm
Location: IL NOEEZ

Re: Life's purpose is to create Artificial General Intellige

Postby tvelection » Mon Jan 29, 2018 7:50 am

PGJordan,

I can only give you general advice and it’s a warning about abstract theories v. more pragmatic invention. For example if someone said they wanted to pursue “string-based artificial neural networks.” All this advice would be the same. An inventor may get caught up in the minutiae of abstract theories and impede their own progress by trying to substantiate a scientific theory instead of the project at hand. At that point the inventing stops and devolves into unsolvable complexities.

When I had read the terms “supersymmetric artificial neural network” I thought that your approach might seal you in. If some begin with supersymmetry then they may refuse to alter their course indeed end-up in a process of ever-reading. Supersymmetry will be a “wild goose chase” that will ruin your better efforts. And this is the key; if supersymmetry arises naturally in a “learning” network then you need not concern yourself until much later stages.

You should be careful not to make the effort about supersymmetry as much as a working artificial neural network . . . first and foremost. An example would be if someone (like the link) was steeped in both “string theory” and artificial intelligence. {As a skeptic of string theory, I think it’s a fantasy based on a need for extra-dimensions which makes it a delusion of unproven “hyper-dimensions.” Those abstracted physical extensions are reified (analytics claimed to be actual existences) then hypothetically extended, leaving no empirical basis for string theory except as non-correspondent mathematical speculations, simply a kind of coherent and consistent higher math set in that regard}. That “inventor” who believes in string theory then has an idea, “string-based artificial neural networks.” And so instead of building an artificial neurology he, or she, has books on alleged string theory and in fooling around with string theory’s analytics is unable to create a pragmatic machine through such extra-dimension fictions of speculation. Such an approach would never create a “string-based artificial neural network” but only lead to a slew of empty and speculative calculations (designed to be coherent even internally consistent) about how it could work if that physics (hypothesis) was correct in its explanation. The theory of supersymmetrics may inhibit rather than help you. Until then, asymmetry, or no duality, must be a sufficient and pragmatic means and if that (or the testable paradigm/model of the community of scientists) leads to a supersymmetry then it can be adopted. You're not going to prove it obliquely. What I mean is that your project is not to prove a theory with regard to developing a working artificial neural network.

I’d concede that many people who believe string theory are very intelligent and have the proper equations with an internal logic that’s coherent and consistent on paper. So they try to offer a bunch of equations and symbols and say, “See? It’s true. The math works.”That is not good enough for a skeptic (or non-belief, as you say). Yet it only works with numbers not things. First, it must be testable, measurable, repeatable, corroborated, predictive, and we still may not know the model for what is being measured. Another problem is that the math is made to be (correlated) predictive/adjusted for known data regardless if that particular model is correct. Supersymmetry is too abstract. Some may insist that it’s proven by definition. Then again I can describe a unicorn rationally too. Even if the math and measurements agree a model can be a misinterpretation. I looked at the links but I still don’t know. Those with a hypothesis have a way of finding it as the conclusion or try to say that they barred all other possibilities so that it must be the predicted phenomenon they wanted (like LIGO’s alleged “gravity waves”). That any measurement proves (not so much their predictive math) but their desired model in the LIGO case with an inexplicable medium. The point is that as a programmer you may simulate “realities” with code. So that even string theory equations can be simulated as if true, real in existence, even consistent mathematics for multiple universe fantasies (whether they exist or not). I’d forget about supersymmetry and focus on just constructing an artificial neural network.

Questions: {Edit, the 2 questions were meant as one, and then stated "in other words"}

Are you merely speaking of simulated/virtual supersymmetry in the software code or actual physical dual-particle effects with the hardware that utilizes actual dual particle and “string” effects? {in other words} Is this a product more of the hardware (physical supersymmetry that is) or a simulation of the theory/mathematics of supersymmetry in software code? Merely using the analytics consistent math of supersymmetry whether actual supersymmetry itself is true or false.

Much like, say, 6 dimensions ---we may simulate them whether or not they actually exist (some will even imagine that thereby they have proven them). But we cannot build a 6 dimensional “brain” because there are only 3 correspondent dimensions and the relative motion of mass(es).

One is actual sub-sub-atomic existences while the other a physics theory as a useful math-language for coding. That is, “We borrowed some supersymmetry equations in creating this code but do not actually claim ‘supersymmetry’ as real (only its coded simulation).” Because the human brain may not use supersymmetry at all rather that is just a useful way of describing it even if correlated. It can cause a pragmatic inventor to get caught up in a theory. If I proposed (an absurd example) a “10-Dimension neural network,” first, I will have ruined all realistic routes and outcomes and the pragmatic working approach will be blocked by irrelevant considerations and speculation (all-be-they internally consistent and coherent). Even if I successfully used a code that merely uses that math I have not actually proven anything about there actually being 10 dimensions. Rather it is semantics and what one really means is ten “characteristics” of measurable phenomena, e.g. a ten “dimensional” analysis of data. So one ends up writing papers, fanciful equations, reading, and coding but avoids letting the “evolution” of artificial network learning structures lead by calculation, trial and error, an adjustment towards optimal outcomes. This is what stops an inventor in their tracks, they begin with a hypothesis and all results in their mind only serve confirm the one hypothesis they wanted to believe. With regard to the art of invention, approaches and methodologies, I can only offer simple advice on your direction/approach ---I hope you succeed in your efforts just be ready to abandon "supersymmetry."

P.S.

https://physics.stackexchange.com/questions/31882/why-does-string-theory-require-9-dimensions-of-space-and-one-dimension-of-time
Nine dimensions? This theory is DOA. Although the math will be made to imitate any actual outcomes. I like how some suggest, or the side column suggests about "M" theory, that these alleged dimensions (pure fantasy) for which we have absolutely no means to even confirm are "folded" or "curled" or better "so small that . . ." You see, they don't even know what they're talking about in a physical sense. It's empty language. Space is not structured with dimensions we just analyze it that way. Perhaps they should search for hyper lines and points too. They discovered that string theory didn't work in 3 correlated analytic dimensions and "time"(relative motion) and so they just added 6 "extra" dimensions, and if took 20 dimensions then they'd add 17 hyperdimensions, no problem ---or rather big problem. It's an interesting faith.
tvelection
Intellectual
 
Posts: 369
Joined: Tue Sep 26, 2006 12:44 am
Location: Pittsburgh, PA

Re: Life's purpose is to create Artificial General Intellige

Postby ProgrammingGodJordan » Thu Feb 01, 2018 10:15 pm

tvelection wrote:Are you merely speaking of simulated/virtual supersymmetry in the software code or actual physical dual-particle effects with the hardware that utilizes actual dual particle and “string” effects? {in other words} Is this a product more of the hardware (physical supersymmetry that is) or a simulation of the theory/mathematics of supersymmetry in software code? Merely using the analytics consistent math of supersymmetry whether actual supersymmetry itself is true or false..


1. Cognitive science reasonably says that the biological brain may be described in terms of supersymmetric equations:

arxiv: Supersymmetric methods in the traveling variable: inside neurons and at the brain scale

2. It is typical in machine learning to imitate cognitive science:

Neuroscience-Inspired Artificial Intelligence

3. The above is what inspired the creation of the Supersymmetric Artificial Neural Network.

4. If you look on neural net history, the more sophisticated the matrix representation of weights were, the more the neural network could represent the input space.

History:

5. Apart from (1), from the history it is not absurd to hypothesize that we can do better at representing the input space, by seeking the usage of supercharge compatible special unitary matrices, or supersymmetry aligned matrices, i.e. "SU(M|N)". This is where the "Supersymmetric Artificial Neural Network" comes in!

6. And you're probably correct, even if Supersymmetry is not observed at LHC, I may be able to use formal methods from physics/string theory, namely "SU(M|N)" or supercharge compatible special unitary matrix or Supersymmetry aligned matrix, to build an artificial neural network in practice, reasonably in terms of software.
Last edited by ProgrammingGodJordan on Thu Feb 01, 2018 10:26 pm, edited 2 times in total.
Signature: I am interested in completing a novel learning model I call the "Supersymmetric Artificial Neural Network". (See this easy overview)
ProgrammingGodJordan
Thinker
 
Posts: 13
Joined: Tue Jan 02, 2018 1:40 am
Location: Jamaica

Re: Life's purpose is to create Artificial General Intellige

Postby ProgrammingGodJordan » Thu Feb 01, 2018 10:23 pm

tvelection wrote:PGJordan,
If I proposed (an absurd example) a “10-Dimension neural network,” first, I will have ruined all realistic routes and outcomes and the pragmatic working approach will be blocked by irrelevant considerations and speculation (all-be-they internally consistent and coherent).


You wouldn't necessarily "ruin all realistic routes and outcomes", by proposing or composing a 10 dimensional model.

Look at something called the "human brain project" by Henry Markram et al, which concerns "simplices" and other complicated structures.

Article: The Human Brain Can Create Structures in Up to 11 Dimensions

Paper from article above: Cliques of Neurons Bound into Cavities Provide a Missing Link between Structure and Function
Signature: I am interested in completing a novel learning model I call the "Supersymmetric Artificial Neural Network". (See this easy overview)
ProgrammingGodJordan
Thinker
 
Posts: 13
Joined: Tue Jan 02, 2018 1:40 am
Location: Jamaica

Re: Life's purpose is to create Artificial General Intellige

Postby tvelection » Mon Feb 05, 2018 12:00 am

PGJordan,

I wanted to make clear in the last post (but left out) that you know far more on this subject than I do. Your 6 points were succinct responses. Your knowledge, readings, and citations, of the subject is admirable and I don't doubt you are a talented computer coder who may be able to make a contribution to AI. My only concern was that you ought to have a realistic route to your goal of an artificial neural network and, as I've done myself sometimes, not to get into/believe overly speculative theories that may block the pragmatic progress that's usually a hallmark of inventions in development. If the supersymmetry approach works then fine (I'm not saying you are wrong) but be careful too because it could lead you astray. It can be like trying to build a (impossible) perpetual motion machine that always seems to work in design but never actually works in practice. Personally, I wonder what IBM's "Watson" platform software does and if they pursued a neural network design. It's also intriguing how the internet/web itself has the "feel" of a spontaneous neural network that simply needs "taught" to learn.

Thanks for the responses and links; I'll check them out and try to follow them and post if there are any more questions or comments.

Oh, and one more question. Will your artificial neural network start off "ignorant" and childlike and then learn from there? Can it ask questions? Seek input? Or will it be more of an "instant adult" like "Watson" that uses knowledge it has not really learned through experience or endeavor but is simply referenced and retrieved?
tvelection
Intellectual
 
Posts: 369
Joined: Tue Sep 26, 2006 12:44 am
Location: Pittsburgh, PA

Re: Life's purpose is to create Artificial General Intellige

Postby ProgrammingGodJordan » Mon Feb 05, 2018 3:42 am

tvelection wrote:PGJordan,



Oh, and one more question. Will your artificial neural network start off "ignorant" and childlike and then learn from there? Can it ask questions? Seek input? Or will it be more of an "instant adult" like "Watson" that uses knowledge it has not really learned through experience or endeavor but is simply referenced and retrieved?


The plan is to construct a reinforcement learning model (in virtual scenarios/games), that can perhaps benefit from learning supersymetric weights, i.e. weights that may store information beyond phase and magnitude data, as observed in unitary matrix based or complex valued artificial neural networks.

Here is a quick read on why games are important as test areas in machine learning.
Signature: I am interested in completing a novel learning model I call the "Supersymmetric Artificial Neural Network". (See this easy overview)
ProgrammingGodJordan
Thinker
 
Posts: 13
Joined: Tue Jan 02, 2018 1:40 am
Location: Jamaica

Previous

Return to Great Ideas

Who is online

Users browsing this forum: No registered users and 2 guests

cron