Ai Dreams Forum

Artificial Intelligence => General AI Discussion => Topic started by: Hopefully Something on December 25, 2018, 12:40:23 pm

Title: Tinfoil Hats!
Post by: Hopefully Something on December 25, 2018, 12:40:23 pm
Instead of making millions of individual artifical neurons signaling to each other with tiny leds powered by batteries charged by tiny solar cells suspended in gelatin in a periodically illuminated skull. Instead of going to all that trouble and expense…

Tinfoil hats! What you do is print a giant mesh of nodes connected with strips of conductive material on a flexible substrate like a plastic or rubber. Then you crumple it up like tinfoil and the neurons can start talking to each other and figuring out where they are. Now that we don’t have a power shortage issue more communication methods between neurons can be explored. Sound, light, vibrations, electromagnetism. Unfortunately, from a different point of view this is going to be a huge folded plastic electric blanket that’s plugged in and then kept in a cupboard for years. So it’ll still have to be emmersed in a liquid for cooling. No getting away from it, brains have to be heavy if we want compactness. Super conductors? I’m a bit wary superconductors. Dont get me wrong, they are all kinds of cool, but the situation with the temperature control feels like a nuclear reactor. Conditions change too much and BOOM, runaway meltdown. It's bound to stress the robot out if it's true agi.
Title: Re: Tinfoil Hats!
Post by: AndyGoode on March 07, 2019, 10:11:18 pm
I came up with basically the same idea years ago, though my variation was that one takes a lot of artificial neurons with many spikes protruding from each neuron, where each spike of which is an electronic axon, then one presses all those spiked balls together and applies either heat or electricity to get them to fuse together at numerous spots on the spikes. Then one does what you suggested: have each neuron figure out via those axons where it lies within the resulting conglomeration, who its neighbors are, then it figures out on its own which routes it needs to take in solving a problem. Although randomness was used in setting up the network, at least the creation of the network was simple, whereas the engineering of numerous 2D silicon layers stacked atop each other with engineered connections is prohibitively difficult.

(https://i.etsystatic.com/8511832/r/il/bcbca7/546483426/il_570xN.546483426_aw8s.jpg)

There is a terrific need for 3D chips, especially neural chips, like that. The recent thread on quantum computers was another example of this need: D-Wave has been cutting corners on complete connectivity of its qubits because it's too difficult to squeeze all those connections onto the surface of a 2D chip. If only they could make a 3D chip their problem would be solved.

With so many people needing some solution to this problem for so many years (this was an issue even back around 1990) one would think somebody would have solved the problem efficiently by now, but I haven't read any evidence of that. Your idea is a good train of thought, I believe, and maybe with a little more thought and experimentation you could develop a new, patentable technology, and your own startup company.

https://www.livescience.com/52207-faster-3d-computer-chip.html

(https://aidreams.co.uk/forum/proxy.php?request=http%3A%2F%2Fwww.livescience.com%2Fimages%2Fi%2F000%2F078%2F470%2Foriginal%2F3d-stacked-chip.jpg&hash=69dce78a7e4b07ea2a853683b084c7c1)

P.S.--If you're interested, I'll tell you the ideal solution to the heat problem in digital computers, if you don't already know about it.
Title: Re: Tinfoil Hats!
Post by: LOCKSUIT on March 07, 2019, 10:35:40 pm
Tell us how yes! Finally, how to remove the heat.
Title: Re: Tinfoil Hats!
Post by: AndyGoode on March 07, 2019, 10:54:26 pm
Tell us how yes! Finally, how to remove the heat.

The answer is reversible gates. Universal quantum computers naturally use reversible gates, but reversible gates can also exist in fairly conventional digital computers. To be honest, however, I've never understood the details of reversible gates to the extent that I understood why they can't be easily built.

----------

(p. 14)
For present designs, the slower the clock, the more
efficient the computing--and what we would like to have is ever-faster
processors. Hopefully, more efficient designs will be found in the next
few years, since by then reversible computing apparently will have to be
implemented for the reasons given below.

1.7 Reversible Gates

   When we calculate kT ln 2 for, say, 100o C, we obtain 3.6 x 10^-21
J(oules). This amount may seem negligible, but even with the absolute
minimum of one bit of logical information supported by only one bit of
physical information (electron state) in a gate, such a bit passes through
thousands of gates during each of billions of clock cycles each second.
When we consider that the number of transistors in a CPU (central pro-
cessing unit) may soon reach 1 billion, and that within a few years CPU
clock speed may exceed 10 GHz (see p. 135), we can estimate that the
maximal number of processed bits per second will exceed 10^19 bits per
second per CPU. Then we can easily calculate that unless we overcome
kT ln 2 per bit, no cooling could prevent such CPUs from melting. As
we have already said, the hardware will have to be changed to replace
today's clock pulses with resonating "swings." For example, electrons
could ballistically oscillate in silicon crystals or carbon nanotubes until
they hit a programmed lattice defect, which change change [sic] the paths
of some of them in order to perform logical operations. Such oscilla-
tions would need only a tiny amount of energy dissipation to keep them
swinging with a constant frequency.
   Reversible hardware will also require a new kind of software to imple-
ment and use a reversible logic algebra. It seems that development of
such a software is feasible. For example, it has already turned out that
a comparatively small number of oscillation delays would be required.
In today's computers, a series of delays of relevant clock pulses is always
implemented. In Figs. 1.1 and 1.2 we can see that a pulse (square wave)
cannot arrive at both a gate and a source at the same instant. We first
have to switch the gate on, and only after a delay can we let current
through. Each transistor has such a delay built into it. Groups of gates
in a computer are incorporated into sophisticated timed circuits. The
gates then "calculate" within the time window determined by successive
clock ticks. Any voltage-level change that occurs in response to a clock
tick must charge or discharge parasitic capacitances associated with a
transistor and its surrounding circuitry. The energy cost of (dis)charging
a capacitor is C V^2 / 2. In a conventional circuit, most of this energy is
dissipated resistively into the environment. In a reversible gate, on the
(p. 15)
other hand, the energy stored in parasitic capacitances is not dissipated
but is returned back to the circuit.
   The software for the first reversible processors has already been imple-
ented [De Vos et al., 2001]. Essentially, it is a clever way to implement
calculations by swinging electrons back and forth. At the end of each
swing, all obtained outputs are either copied and taken out [Bennett,
1973; Bennett, 1969] or reintroduced into the next swing. A classical re-
versible computer will most probably be a link between today's classical
computer and a would-be quantum one, when the shrinking of com-
puter elements hits the one atom barrier in about two decades. It is
no coincidence that reversible algebras underlying reversible computer
and quantum computer theories were developed in parallel and that
they have many characteristics in common: reversibility, control, and
universality. General software, on the other hand, diverged: classical
reversible computing software is just a technical blueprint for speeding
up already existing general purpose hardware, while a general software
for quantum computing still does not exist and is not likely to resemble
reversible computing software. Therefore, we will not consider reversible
computing software any longer but will present some details of its alge-
bra that will turn out to be relevant for the algebra of quantum gates
later on.
   We mentioned above that the logically reversible NOT gate is not re-
versible physically (in today's computers) because a voltage must switch
the gate in order for a current to pass through the source to the drain.
So it might be better to call the gate states within reversible computers
before and after rather than input and output as with standard com-
puters. These terms stress that we do not let gate current "through" a
transistor--we only redirect it so as to be able to reuse it. (This is why
reversible computing is sometimes called green computing.)
   In Fig. 1.4, we show the so-called controlled-NOT, CNOT operation,
which reuses the "gate current" of a NOT gate and serves as a reversible
logic gate. CNOT cannot be used to express and then reverse the NAND
operation (see Fig. 1.3, p. 13 and Fig. 1.4(a)) since it has only two
outputs and for a reversible NAND we need three outputs: one to give
its value and two to record the two inputs. CCNOT, shown in Fig. 1.4(b),
can do just that--as shown in Fig. 1.5(a) [Feynman, 1985]. Moreover,
it turns out that CCNOT is one of over 38,975 universal logic gates
[De Vos et al., 2001] among 8!=40,320 reversible gates with three binary
inputs and three binary outputs.
   Graphical representations of the kind presented in Fig. 1.4--reversible
circuits--are very common in both reversible and quantum logic, where
"logic" simply means a set of rules for handling gates. Expressions
(p. 16)
[figure]

Figure 1.4. Reversible circuits: (a) C(controlled)NOT gate (B~ = B_ for A = 1 and
B~ = B for A = 0); (b) CCNOT (Toffoli) gate (C~~ = C_ for A = B = 1; C~~ = C
otherwise).

on the left-hand side are "before" (inputs) and on the right-hand side
are "after" (outputs). Dots stand for controlling gates and mean that
expressions do not change by passing through these gates from left to
right. The gate they control is the (+) gate. Graphs can be concatenated
as shown in Fig. 1.5(b).

Pavicic, Mladen. 2006. Quantum Computation and Quantum Communication: Theory and Experiments. New York, NY: Springer Science + Business Media.
Title: Re: Tinfoil Hats!
Post by: Hopefully Something on March 08, 2019, 02:39:43 am
Thanks Andy, for the encouraging words. Maybe I'll explore the idea in more detail.

That's interesting . So it would be slower because of the constant backtracking. Bigger, because of doubled up diodes and the inductors & capacitors. But more efficient, because the surge of current (upon crossing the threshold voltage of the diode) gets stored and (used in reverse?) or easier to imagine, looped back and sent through the same diode with only a small energy top up every time. Hmmm... How to implement with a physical neural net...

I do like your spike fusion idea. It would naturally be faster then my emitter/receiver signaling. Duh, just use a solid connection. If designed smartly it would also be more flexible, better able to handle changes in geometry. The downside's are extractablilty, (need a way to replenish the system with fresh neurons), and I can't imagine a method of power distribution.
Title: Re: Tinfoil Hats!
Post by: AndyGoode on March 08, 2019, 11:28:43 pm

That's interesting . So it would be slower because of the constant backtracking. Bigger, because of doubled up diodes and the inductors & capacitors. But more efficient, because the surge of current (upon crossing the threshold voltage of the diode) gets stored and (used in reverse?) or easier to imagine, looped back and sent through the same diode with only a small energy top up every time. Hmmm... How to implement with a physical neural net...

You got me interested in reversible gates now. Here's a video that helped me understand them a *little* better:

Reversible Computing
https://www.youtube.com/watch?v=rVmZTGeIwnc&t=47s
Science University of Copenhagen
Published on Jan 7, 2016
length = 10:02

I think I'd need to go through a specific example of the hardware and software, like the example shown in the video, before I really understood reversible gates, though. I clearly understand that the gates have to have the same number of inputs as outputs, the input data must be reconstructable when the output is fed back into the output in the reverse direction, and that there must be some type of hardware that causes the electrons to surge back and forth (the electrons become sort of like ping pong balls bouncing back off a wall), but I don't understand how the gates would be chained together to produce any meaningful action, I don't know what the software would look like or how it would have to be changed, I don't know how the overall speed would be affected, and I'm not sure about the increase in the number of diodes (I think that would be a relatively small additional cost). I have an attraction for exotic types of computers, though, so this topic is much to my taste.

If anyone reading this post understands reversible gates well, maybe they can teach us all. Or maybe if someone is interested enough to do research, then we can teach each other what we learn along the way.

As for physical neural nets, I think that would be the least of your problems, because: (1) neural nets have extremely simple processors (only weighted sums followed by thresholding or compression before passing on the signal) so would need very limited circuitry; (2) any circuitry implementing the neural net would already be covered by the methods and technology worked out for reversible gates. From the book and video it sounds like the main problem is building some device that will cause the electrons to surge back and forth at regular intervals.

(https://aidreams.co.uk/forum/proxy.php?request=http%3A%2F%2Fwww.psych.utoronto.ca%2Fusers%2Freingold%2Fcourses%2Fai%2Fcache%2Fneur_ana.gif&hash=27ac7a40e32d15d75f3f6f80ba260a5b)
Title: Re: Tinfoil Hats!
Post by: Hopefully Something on March 13, 2019, 05:21:30 am
Hey Andy and anyone interested,
Cubes with small magnetic extrusions would solve all those problems. It would be the most compact design, have direct communication like Andy's idea, be shock resistant, have simple power distribution since it's a cube, and have extractable nodes. There would even be just enough gaps for a slow current of water to cool the system.  If you're still looking into reversible computing, "The Pleasure of Finding Things Out", by Feynman, has a section on it as well as some of his other ideas around computing.
Title: Re: Tinfoil Hats!
Post by: AndyGoode on March 15, 2019, 06:25:58 pm
Thanks, H.S., I'll try to find a copy of that book, starting first with libraries.

I think I could design a reversible circuit for doing the sums in a neural network without much problem, but the inherently analog transfer function (typically an S-shaped response curve) in a neural network is much more difficult unless an analog circuit (like in the 2nd diagram I posted) is used. That raises an interesting question in itself: Are any analog circuits reversible? My guess is yes, even more so than digital circuits, but I'm really getting out of my domain of expertise here, so I'm not sure, and I'd still be stuck on what to do with all those side functions and inverse functions I would computing when I am ready to piece together the circuits I designed.