Evolving circuits that learn and they have no idea how

a reply to: ChaoticOrder
Good point, ChaoticOrder. I often have trouble making these distinctions, as well. The words “natural/unnatural” also confuse me sometimes.

I’m not an expert on the subject, but from what little I know it seems kinda fascinating. The fact that self-organizing behavior is exhibited in both biological and non-biological systems doesn’t surprise me; actually it makes sense. It kinda reminds me of finding the path of least resistance. Nothing mystical there. I think some (maybe a lot) of the self-organizing behavior in systems is supported by Complexity Theory.

That’s all I know (right or wrong)! It’s cool, but it’s not magic.

Here’s a couple self-organization programs for anyone interested:

CALResCo - www.calresco.org... - Many Programs demonstrating Order from Chaos, Boolean Networks, Artificial Life, Self-Organized Criticality and Multi-Agent Simulations are currently available (QBASIC & Executables).

Rudy Rucker - www.mathcs.sjsu.edu... - Cellab, Cellular Automata (some self-organizing) & Langton's self-reproducing CA (Windows).
www.cs.sjsu.edu... - Other interesting stuff by Rudy Rucker

Great thread. Thanks!

No links to original articles, papers or any credible info.

Apart from that, this is the sort of basic circuit evolution that was being done in the 1980's.. nothing new here.

a reply to: ziplock9000

The work was was novel:

I studied for four years for an IEE accredited MEng in Microelectronic Systems Engineering from the University of Manchester Institute of Science and Technology. Under Phil Husbands, I completed a doctorate in the School of Cognitive and Computing Sciences at the University of Sussex entitled "Hardware Evolution." This was the first thesis in the field now know as "evolvable hardware" or "evolutionary electronics", and it was chosen by the BCS/CHPC Distinguished Dissertations award to be published by Springer (ISBN 3-540-76253-1). After a post-doc to further that research, I remained at Sussex as a lecturer in the Department of Informatics and an EPSRC Advanced Research Fellow, returning to normal teaching duties in 2006.

www.sussex.ac.uk...

The concept was pioneered by Adrian Thompson at the University of Sussex, England, who in 1996 evolved a tone discriminator using fewer than 40 programmable logic gates and no clock signal in a FPGA. This is a remarkably small design for such a device and relied on exploiting peculiarities of the hardware that engineers normally avoid. For example, one group of gates has no logical connection to the rest of the circuit, yet is crucial to its function.

en.wikipedia.org...

Source:

www.springer.com...

originally posted by: voyger2


Questions:
Suppose there is already a AI on the WEB lurking, studding and evolving, How would it be possible to identify it? Can we (as humans) recognize or identify a reality (new), concept or "living-form" without "formal" auto-presentation or direct contact?

We're no where near the point of developing "true" AI so there isn't going to be one lurking on the web somewhere.

a reply to: NiZZiM

I remember reading that article when it first came out in 2007. Damn interesting is one of my favorite sites to visit.

originally posted by: TheConstruKctionofLight
a reply to: BlackProject

When scientists scratch their head on a daily basis on what process to use to cure lets say cancer

Plenty of cancer cures out there. When scientists come up with a process its called a treatment. They never use the word cure.

I certainly agree, I did not really mean for cancer was just a thing to say for a use of such machines. However I am aware that there are cures costing pennies to use but they do not use them due to the money machine that is the pharmacy corporations at power in the world. Just like the great story told in a recent film Dallas Buyers Club where the victim had AIDS and was told by doctors he will die in a few weeks and that was it, they left him. However by looking into other avenues himself he found a non FDA approved drug which allowed him to live many more years more. It shows that such things do exist and the fact is even though we know it, is anyone doing anything about it?

Anyhow, yeah. Im sh*t-chatting, I mean chit-chatting.

Ok, so the article says he uses 100 logic gates. If these are standard logic gates, then it is a binary system. or like 2^100, or 100 bits, either on or off.

A 64-bit register can store 2^64 (over 18 quintillion or 1.8e+19) different values.


"a mere 100 logic gates" can store 2^100 (1.2676506e+30) different sequences.


What is wild is that in so few iterations, that the solution was found.
(4,000 iterations of batches of 50 100 bit bytes?) Even a million different combinations out the entire domain of possible sequences is a very small number. This would infer that there are a very large number of possible solutions to the problem within 100 logic gates, if the program worked reliably. However, the final program did not work reliably when it was loaded onto other FPGAs of the same type.


The weird or non-understandable part is the anomaly of the disconnected gates that caused it to work. Considering the extra variables of the flux phenomena; or, the idea of a grey bit, the odds are astronomical. I wonder what happens when you run the program again from scratch? Does it reliably find a solution from the original FPGA, or does it find the same convoluted solution?

a reply to: ogbert

I wanted to clarify a bit from my last post.

A 64-bit register can store "one of" 2^64 (over 18 quintillion or 1.8e+19) different sequences.


"a mere 100 logic gates" can store "one of" 2^100 (1.2676506e+30) different sequences.

If you want to simply print out the last lottery numbers; and, you know them in advance. You could use this algorithm to delete the losers and keep adding random numbers until the pre-set number is achieved. If it is the Florida lottery, somewhere between 1 and 14,000,000 tries, the number will be generated.

What I am saying is that there is no "ghost in the machine" or that somehow no body knows how AI has actually developed some kind of consciousness. Garbage in Garbage out. This makes a very interesting phenomena, but if it is not reproducible, on other chips, this may be a fluke.The algorithm is not like throwing darts till you hit it,- the failures are deleted from the domain of possible sequences. Obviously, it is possible to solve the problem with 100 gates, but how many ways? Eventually, with enough time all ways to solve the problem will be generated, which can be quite useful to find solutions that humans would not have come up with.

a reply to: ogbert
Actually this is AS not AI IMO. AKA artificial sentience not intelligence, and here's why. Say you could map every single neuron that fires neurotransmitter release and nerve impulse of a person doing a task, and then port it over to someone else's brain somehow and force it to try to run it's not going to work. And for very similar reasons really. Many of the very unorthodox solutions it came up with depend on things unlikely to be the same from chip to chip. (Now when I say the same I mean EXACTLY the same!)

The Macobserver.com, Oct. 18, 2016 - Thoughts About Apple’s Secret iPhone 7 Chip.

The iPhone 7 was taken apart by ChipWorks and they found a FPGA chip on the inside. The article speculates the reason and function of this chip.

Fujitsu’s new architecture on the other hand uses parallelisation to great effect, and unlike quantum computers boasts a fully connected structure that allows signals to move freely between the basic optimisation circuits, making it capable of dealing with a wide range of problems and factors – and still offer the speed seen with quantum computers.

Fujitsu says it has implemented basic optimisation circuits using an FPGA to handle combinations which can be expressed as 1024 bits, which when using a ‘simulated annealing’ process ran 10,000 times faster than conventional processors.
...
The company says it will work on improving the architecture going forward, and by the fiscal year 2018, it expects “to have prototype computational systems able to handle real-world problems of 100,000 bits to one million bits that it will validate on the path toward practical implementation”.

Techradar.com, Oct. 20, 2016 - Forget quantum computing – Fujitsu has a better idea.

So a little more than a year after this thread was created there is real world applications being utilized. The Fujitsu concept is rather awesome! And kind of insane which is why I like it! The "fully connected" aspect in particular. This seems to be the path of NP algorithm modeling but hey, artificial neural networks could just as easily be done!

originally posted by: TEOTWAWKIAIFF
The article speculates the reason and function of this chip.

FPGAs are useful for 'function goes here' expandability in a design. We often/nearly always put them in the I/O flow of a board, with some patch resistors to route around it for users who are on the cheap.

You never know when they might come in handy, if you're not really sure how the thing's going to be used in the future.

It's sort of a jump to assume that it's going to be used for some oddball usage like in the OP, as these results often don't play on identical parts, and Fujitsu's annealing algorithm runs on a 'sea of FPGAs', and annealing isn't useful for the sorts of things a phone would do.

We often use FPGAs such as this when we want to roll up a group of unrelated functions into one part to save space. However, for mass production on Apple's scale, you'd typically take the finished/tested FPGA program into a cheaper fixed function part, so Apple must be re-tasking the thing constantly, or they're not sure of the end function's specs yet and want to keep it flexible for now. Later phone runs might ditch the FPGA for a custom ASIC that fits on the same pads.

This particular FPGA is targeted for cell phone and pad use, and has a wad of off the shelf IP for sensor management, barcodes, USB and the like. It has a little 16 bit DSP as well. You'd use it to offload the low level crap from the main processor.

a reply to: Bedlam

Cool! Thanks for the info! I was thinking of something along the lines like "rapid prototyping for a fixed chip use" but just throwing one in a phone seems like a waste. The guy saying it was for encryption is kind of a joke (using a welder when a paper clip is all you need). I like the DSP aspect as well--put it to use like the Bose noise cancelling (or VCO dalek voice, "ExtermINATE!").

Also seeing Microsoft being all hot and heavy about them too. Like you said, good for I/O and load shifting. The Fujitsu idea is my kind of crazy. Did you see Stanford's Ising computer? That is a pretty cool device as well. [ETA: ATS link here]

Awesome you get to play with them! Looked and they have a PCIE board for 6K which is a bit more than I am willing to shell out.

Maybe... one day... a TEOT can dream!

Hey, FPGAs are fun. I started on a new personal Zynq project today.

If I ever get it to do what I want, one day I might reveal all on ATS.

The development board is $3600 but it'll be worth it if I can get it to do what I want.