[Korrelan]

Should I make its brain and body nerves not just be code but also be in the physical word like korrelan showed in his video a flythrough render of his AI's brain neural network?

My simulations aren’t in a/ the ‘physical world’ they are in virtual space.  It’s the same principle as blender; you have to be able to design and manipulate the connectome models.  Because my connectome models are 3D by design it makes sense to edit them in 3D; I personally work better with strong visual feedback. 

I have just added a facility for placing labels/ markers at key locations because I keep getting lost whilst working in the connectome.  The models grow to vast sizes (equivalent to 10’s of miles at neuron scales)… it’s wise to take a packed lunch if I’m working in there lol.

How did you do this?

It’s called Math lol; Trigonometry and Geometry.  I wrote the editor from scratch incorporating the neuromorphic processor at the base level.  It can edit/ simulate any kind of neuron or NN topology in an unlimited virtual space.  I designed in parallel processing and support for any number of cores into the system from the beginning.

It’s one system that can be used to build/ combine/ explore any kind of neural network, recurrent, convolutional, etc with any kind of sensory inputs, audio, visual, tactile, etc. using any type of connectome or neuron model.  I can mix and match, copy, move, merge, etc neural models just like you use blender.

Blender is a 3D object/ scene modelling program; it won’t allow you to model millions of neurons and billions of synapse and the complex interactions/ dynamics between them.

Using a 3D physics engine to test an AGI is a false economy.  You are just adding more variables and uncertainties into the mix; no physics engine is 100% accurate.

Does it have an advantage over just leaving the skull empty and running the brain code?

I think you are thinking in blender model terms.  As above… my AGI is not in a skull; my AGI doesn't run on the computers… it runs on the wetware, neuromorphic processor simulation that runs on the computers.

I'm simulating biological structures and reactions; there are no heuristic/ binary/ computer/ gate logic equivalents to the processes/ logic I'm using to generate my AGI.

 

[LOCKSUIT]

Where is your AI's learning environment then? All you have is a network, there's no 3D world with physics...Where is this going

Blender is a 3D object/ scene modelling program; it won’t allow you to model millions of neurons and billions of synapse and the complex interactions/ dynamics between them.

Blender does a lot actually, how do you know that? Why can't Blender take the resources / or have the functions (even visually like your light-show-network)?

[Korrelan]

Where is your AI's learning environment then?

I use actual microphones, video cameras, torque sensors, accelerometers, gyros, etc (real world sensors) for inputs and displays, speech engine, servos, bot for output. It's learning environment is reality.

Blender does a lot actually, how do you know that?

I actually know a little about Blender and I’m pretty sure the last time I checked there was no plug-in for a Neuromorphic processor (I could be wrong (doubt it))… but if you think you can use blender to create your AGI then cool.

Why can't Blender take the resources / or have the functions.

It may be able to map and render the vectors for a few hundred thousand neurons etc, but it wasn’t designed to simulate/ emulate biological neural constructs/ systems.

You could probably design your AGI in python and then use a blender model as a real world physics engine… not sure about complex real time simulations though.



 

[LOCKSUIT]

This is an important message.

Got me there!

Is there an advantage by using a neuromorphic chip? I mean, you store senses/actions, link reference them to connect em.......does it save on code by using connections and gates?. . .I'd think it does the same thing just it's more "hardware-y".

Ohhh I've been looking around a lot, is that blender video showing that Python in Blender allows parallel computation/actions? I'm scared that Python in Blender wants to serial each motor and not allow many simultaneous motors / neural-fires!

I want to propose multiple (nice) checks to you for multiple little progress jobs, you on? In either Blender, or your reality-setup, or other. I'm the kind of guy that will make sure you not only get more money than worked for but also provide extensive ground proof that you'll get the money (which is also backed by "little progress jobs"). After the first (SMALL!) job for like 70 dollars, you WILL be amazed and see the results!! You'll love it! You can use my rigged baby. You're getting more money than usual, for something unimaginable, while have the skills to implement it.

[Korrelan]

Is there an advantage by using a neuromorphic chip?

Neuromorphic is a catch all description for biologically inspired/ based systems.  The specification/ operation of a neuromorphic system is subject to the designers ideas/ theories.  No generic neuromorphic chips are commercially currently available.

I'm scared that Python in Blender wants to serial each motor and not allow many simultaneous motors / neural-fires!

Unless you are using a dedicated multi core computer designed for parallel processing then you are stuck with serial computing.  Parallel programming however is a coding discipline/ schema that simulates parallel processing on a serial computer.

I want to propose multiple (nice) checks to you for multiple little progress jobs, you on?

Like the rest of the peeps on the forum I’m here to offer advice and help but I’ll have to decline your offer (again lol).

I’m way to busy with my own life and AGI research.

 

[LOCKSUIT]

I have the parallel 1060 GPU.........also the Skylake i7.....Blender has a switch button....



Problem 1 before was I wanted to create my AI, but couldn't learn coding (EV3 is easy, PLs are no good), and similarly couldn't even find the installers nor able to install them.

Problem 2 before was I wanted to find programmers and couldn't learn to prove it to them.

Problem 3 before was I wanted to pay programmers but couldn't learn how to trust them, get a English speaking person, get them to make it right, and not lose money, and do the whole thing.

Now I see an opening. On a moderated forum like Blender Artists, I have to find an English speaker, that's not on the "go-go" like on free-lancer sites - someone that's situated on the forum and trustable too, and as for not losing money + making it right + doing the whole thing > I will do little job progresses! Which them alone are amazing!

my AGI is not in a skull; my AGI doesn't run on the computers… it runs on the wetware, neuromorphic processor simulation that runs on the computers.

I think you mean:

run software simulation/real-thing-hardware on computer neuromorphic processor.

The chip is the brain, and if it's not, then it's simulation. Even if it's hardware, some small functions can be 2D/1D simulation and then complete the hardware brain's functions back on the chip in reality.

[Korrelan]

If this is the path you have elected to follow then I would suggest, (as others have on this thread) that you import your Blender models into Unity; and use that as the main simulation engine.

Blender was designed primarily for rendering high quality scenes.

Unity is a game engine and has been designed to run the physics you require in real world, real time simulations.

 

[Freddy]

There is a Blender game engine, but it used to be unloved and I think it was nearly dropped. Might be worth looking into to see if there is any movement.

https://en.wikipedia.org/wiki/Blender_Game_Engine

[LOCKSUIT]

I'm gonna pay.....but gotta find the engine now.....Unity costs iey iey

I must think what I need......I need good enough speed, parallelism, resources handled, o-k physics, (for now to make a AI with CNNs etc, just not necessarily the full thing, just enough to show it once done).

Here's someone's Blender:


a good one

[Korrelan]

Don’t forget there are other options. 

Bots like this can easily be controlled by a cheap Arduino and easily networked to a larger computer/ cluster.  Video cameras, microphones can be added as required, etc.

https://www.aliexpress.com/item/1set-17DOF-Biped-Robotic-Educational-Robot-Kit-Servo-Bracket-Ball-Bearing-Black-W-Servo-Horns-For/1531636741.html?source=%7Bifdyn:dyn%7D%7Bifpla:pla%7D%7Bifdbm:DBM&albch=DID%7D&src=google&albch=shopping&acnt=494-037-6276&isdl=y&aff_short_key=UneMJZVf&albcp=658432934&albag=36001185289&slnk=&trgt=66391061561&plac=&crea=en1531636741&netw=g&device=c&mtctp=&gclid=CJvW2frR7tICFcUy0wodXKcCqA

Another option is to buy a 3D printer and build your own from pre-designed plans.



or even…



Again... they can use a simple Arduino servo interface.

If you’re going to find someone to code your AGI and take this path; you need to make sure they are proficient in this type of project.

 

[LOCKSUIT]

Mine needs at least 20 somotosensory receptors though, humanoid baby body, good eye/s, hear, accelometer, voicebox. VR is way better if you don't yet have a enterprise corp. team working on it.

[LOCKSUIT]

@korrelan korrelan

Ok this is crucial, I need to know, of my 3D engine options, the things I'm picking for.



Do all 3D engines allow infinite/lots of resources so long as you have the processing/storage power? I could get so far and not be able to add any more parallel motion/mesh/operations.

Could some be missing operations I need? Ex. motors, nerve vectoring, etc.

Could some be way too slow?

HELPPPPPPPPPPPPPPPPPPP I need to know which engine I must go with, this is the time to help bigtime, this is all I need now. I'm gonna hire. I don't need any more $/AI/convincing-programmers. Just the baby - the engine. Start the engine, will you?

[kei10]

Do all 3D engines allow infinite/lots of resources so long as you have the processing/storage power? I could get so far and not be able to add any more parallel motion/mesh/operations.

Could some be missing operations I need? Ex. motors, nerve vectoring, etc.

You could do all of that if you know what you're doing. However, your computer isn't a super computer. When it comes to that, your normal computer, of course, will have limits of what it is capable of. Infinite/lots of resources isn't going to happen, and you'll get to a point where you won't able to add anymore parallel motion/mesh/operations.

[LOCKSUIT]

Yes but the thing I'm creating is going to be limited enough that it definitely will run on my computer / GPU/i7.

I need anyone/korrelan to help answer.

[Korrelan]

Kei10 is correct.

When it comes to running huge complex simulations you will soon hit the limits of using one computer.  That’s why I built my own cluster; no computer system has infinite resources.
 
My personal cluster comprises of six quad core over clocked CPU’s; with another six quads available when I really need the extra poke. You need specialised software to enable all the machines to run in parallel; they all have to work in perfect unison all the time.  You will require a master terminal that controls the flagging and load balancing of the separate CPU’s in real time… the list goes on.

You also have to consider access speeds and the like; just because I have 30TB of storage doesn’t mean I can access it all in an instant; these computers are still serial devices emulating a parallel system.

Yes but the thing I'm creating is going to be limited enough that it definitely will run on my computer / GPU/i7.

I would like to know how you are so sure before you have even started coding the AGI; you will be surprised at how quickly you will hit the limitations of an i7 rig.

Regarding the best game engine to use; I suggested Unity because I have seen projects that run along similar lines to what you require. It’s up to you to research the best engine for your requirements; I’m just saying consider Unity.

I’m not trying to dissuade you but what you are considering if going to be fraught with problems.  I see your lack of programming knowledge/ limitations/ techniques being a major bugbear.  You will request routines like…I need a bit that can save all the images it sees, access them all with a CNN and return matches.

Just impossible… consider that a PC can’t even search an image file in parallel; what ever method you use has to go through the image one bit/ byte at a time.  This is why the likes of ‘Google’ employ actual huge parallel super computers. 

I have a lot of experience writing bespoke software for various client/ corporates and I’ve said this before… you need a much, much better knowledge of how your idea is going to function at the program level.  You will have to specify the techniques/ algorithms/ system topology/ etc… you can’t just expect a programmer to be able to produce what you require from a vague idea.  Even utilising a CNN for what ever purpose you require will be a problem; even if the coder understands the algorithms there a literally thousands of types, schemas, techniques and methods of application to consider/ code/ bug test/ etc… ‘you’ not the coder will have to specify its design, usage, parameters, etc.

It’s because you have limited programming/ development experience that you are failing to see problems/ limitations you are facing.

The chasm between a basic theory and building a working system is vast.