Ai Dreams Forum
Software & Hardware => General Hardware Talk => Topic started by: HS on May 29, 2019, 04:48:30 am
-
Manufacturing electronics the size of human cells isn’t an option yet. Any artificial brain, therefore, would be quite unwieldy compared to the biological equivalent. The electronics could partially make up for this by being faster. Are there other ways to even the odds? Biological cells communicate with electrical impulses, and employ chemical reactions to modify cell behaviour. Those two communication channels offer great self-fine-tuning capabilities. The thing about electronics is that they can have way more channels than that. Do you think there would be an advantage to operating on many wavelengths at once? Suggestions about what would work well together?
-
Why are you trying to build a human brain using real nodes and not computer simulated neural networks? Computers can control all nodes and thresholds in parallel on GPU, and more, it's fully fine-tuneable.
Computers run a world of programs, from calculators to regression to object outlining. You need to work with computing devices and technologies. Everything from computers, to input wires, to output devices. Computers even store and upload from the internet all knowledge in text, video, etc. So, you need to stick to computer and not mechanical nodes... Computers can even brute force, etc. . .
-
Just a personal preference, I like the idea. I have a good chance of learning something new because nobody else is doing it. Also physical neural nets should outperform simulated ones at large scales, and since large scales are my eventual goal, it seems prudent.
-
Ok but, if you want to make the AGI make discoveries, or inventions, it will require teaching the AGI the stuff we learnt on the internet, and using the tools we have, and learn skills by watching others. How do you plan to feed in lots of text or video to a swarm of nodes? How does it save into the brain too? And how does it build a hierarchy of parts out of it and re-iterate to Refine that hierarchy?
In my brain, humans are made of limbs, then those are made of rectangles, then lines.....and my hierarchy gets refined over time to re-adjust itself...
-
Electrical transistors are smaller than braincells. Also delay memories are an option too, but I think its hard to hit high frequencies with them, not saying it would be impossible - Just youd be on your own putting it together because the industry doesn't employ them these days.
I don't think the hardware is the problem, the actual intelligence "method" is what is unknown.
Also, they used to make physical neural networks back in the 50's, they employed them for computer vision and called them perceptrons.
They are highly parallel and powerful, if ppl still made them today they would be amazing, but these days I think ternary cam hardware (in the form of a chip) makes more sense to manufacture, and has the same amount of parallel power.
[edit] sorry... u said electronic*S* the size of a human cell... note taken...[/edit]
-
Yeah, I want to bring back that technology and expand on it! I feel like we've chosen to go down the drab fork in the road, in the past, when deciding where to take technological development.
-
Well im here if you want to show me a random access memory chip schematic. O0
They are on the net if you look up google images, but understanding is worth more.
Converting that to a similarity matching chip (for knn hardware) is not much different, and that's the equivalent of a 1950's parallel perceptron. Open Ai, Greg Brockman, is talking about similar things on his youtube videos. Hardware does improve things at least by a square. (and u get on the cheap.)
If you don't want to go from complete scratch, FPGAs is the best option, they are better than GPUs.
but the perceptron is actually better, if u make it yourself, BillGates style.
-
Woah those FPGA things look cool, the tech could be really useful. It could enable tiny 3d electronics down the line. Just needs a slightly different form factor.
-
heres an fpga doing an old school perceptron type implementation. (faster than gpu.)
https://www.youtube.com/watch?v=QhaOEHQdpYk
-
If all the timbits in AGI's brain are not the answer it needs, then it has to invent the answer it needs. Your AGI has to Transform your AGI's sense memories into likely-true answers it seeks. You have to rethink here, is it small mechanical nodes you need, or actually a Big Data discovery engine you need? You have to"perform" operations on the data. AGI=data/actions/tools. You won't be able to make, repair, and control millions of nodes in real life...they require connections, 'programming', omg...
-
The cool thing about you Locky, is at least you think for yourself. The answer's there for the ones that believe in themselves.
-
Each brain has a motor, and it propels itself in a direction.
edit:
When they work in unison, bodies move faster.
-
You won't be able to make, repair, and control millions of nodes in real life...
Symbiotes! Immune cells & bacteria for biologicals, small autonomous bots for synthetics. AI bodies will need symbiotes as well! Actually more like replace than repair, but same difference.
-
You won't be able to make, repair, and control millions of nodes in real life...
Symbiotes! Immune cells & bacteria for biologicals, small autonomous bots for synthetics. AI bodies will need symbiotes as well! Actually more like replace than repair, but same difference.
yeh that could happen, little bots ontop of a big bot. has his own microorganisms.
So how do u make a robot reproduce? - or clone in a robots case?
You put the manufacturing plant in its body, in a compression shape, and then when it gets birthed, it expands, but even it too, needs a womb, a womb in a womb, a womb in a womb in a womb. so it doesn't work.
But I think if you make 3 different "sexes" they work together to produce the robot, and it gets over the problem. ;o
So instead of the womb in the womb, one just produces wombs by themselves, then the other robot comes, and chucks it into its wombless stomach, understood?
then you can take over the world, cause the robots are their own self supplying factory.