Darwinian breakthrough in electronics

[Perspicuo]

Darwin on a chip: New electronic circuits mimic natural networks like the human brain
http://www.sciencedaily.com/releases/2015/09/150921133946.htm

Researchers have demonstrated working electronic circuits that have been produced in a radically new way, using methods that resemble Darwinian evolution. The size of these circuits is comparable to the size of their conventional counterparts, but they are much closer to natural networks like the human brain. The findings promise a new generation of powerful, energy-efficient electronics

Design-less chips that thus prevent(correct) design gliches. From research at University of Twente (pronun. ['twɛn tə]).

 

[Kharakov]

Reminds me of something that was posted here a little over a year ago... combine the two techs.

http://io9.com/brain-inspired-microchips-simulate-one-million-neurons-1569184586

By modeling a circuit board on the human brain, Stanford bioengineers have developed microchips that are 9,000 times faster than a typical PC. Called Neurogrid, these energy-efficient circuits could eventually power autonomous robots and advanced prosthetic limbs.

Bioengineers are smart to take inspiration from the human brain. It's a highly efficient information processor capable of crunching 100 million instructions per second (MIPS). Astoundingly, it only uses about 20 watts to power its 100 billion neurons. Today, our best supercomputers require a million watts to simulate a million neurons in real time (measured in terraflops). A standard desktop computer requires about 40,000 times more power to run and operates about 9,000 times slower.

This is really cool. 20 watts versus a million watts, very impressive.

 

[barbos]

Sounds like a dishonest comparison. Ordinary computers are not well suited to simulate neural nets.
There are special chips which are designed specifically for neural nets, they should compare to them.
So, no, not impressive at all.

Also they should double check numbers they publish. 100mips for human brain? that's awfully small number.

 

[Shake]

According to Ray Kurzweil in his book, The Singularity is Near, that while the brain is massively parallel, its computational speed is limited by the chemical reactions between the neurons. This happens in the millisecond range whereas computers now operate in the nanosecond range. This book was written a decade ago and I think some of his predictions may be somewhat delayed due to the fact that researchers have more recently discovered that our brains are far more complex than previously thought. I'm sure he'd still argue that it will still only be a few decades before some form of AI surpasses human intelligence.

 

[barbos]

Yes, brain is massively parallel and while it can't be directly compared to ordinary computers, 100mips is basically 100Mhz single core CPU.
Even older "dumb"-phones are probably faster than that.
Many companies have specialized chips which are many orders of magnitude better than conventional CPUs at neural nets.

 

[Speakpigeon]

Darwin on a chip: New electronic circuits mimic natural networks like the human brain
http://www.sciencedaily.com/releases/2015/09/150921133946.htm

Researchers have demonstrated working electronic circuits that have been produced in a radically new way, using methods that resemble Darwinian evolution. The size of these circuits is comparable to the size of their conventional counterparts, but they are much closer to natural networks like the human brain. The findings promise a new generation of powerful, energy-efficient electronics

Design-less chips that thus prevent(correct) design gliches. From research at University of Twente (pronun. ['twɛn tə]).

I suggest a bit of scepticism is in order here. Darwinian evolution took something like 3.5 billion years to produce a reasonably smart brain like ours using a lab the size of a medium-size planet called the Earth. I will assume that the chip here took at most a few months to produce and was "grown" inside a container not much bigger than a Petri dish. So while their conclusion is probably honest and perhaps correct, that does not really get us anywhere near producing an artificial brain as or more powerful than ours. Also, all other species should be regarded in this context as so many design failures in the Darwinian process of producing our brain (assuming our brain is indeed the top brain in the world).

Anyone wants to argue with that?
EB

 

[barbos]

OK, I decided to actually read the link. Sounds like a gibberish and probably is for the most part.
It does not explain how that chip operates. If I had to guess then I would have to say it is an ordinary programmable logic array which has some new way of programming.
They program it into NN and then train using "evolutionary" algorithms. That's not a new thing. And they are not designing general chips that way, they train NN.

 

[fromderinside]

..... all other species should be regarded in this context as so many design failures in the Darwinian process of producing our brain (assuming our brain is indeed the top brain in the world).

Anyone wants to argue with that?
EB

FDI here Speakpigeon Ready to argue. Is Newton physical model a failure just because it isn't the best overall? Of course not. Neither are Tigers a failure just because they can't eat many humans any more. Its a problem with ideals when talking in evolutionary terms which is all about 'good enough'.

As for the OP it seems there's a bit of hand waving when it comes to 'evolutionary' realization of a circuit. Yes failures may be avoided but set aims for design product are prescribed making the realization engine something other than an evolutionary algorithm.

 

[Juma]

How do you know how these things will react in untested situations? And then realize that testing will onlu
Testing will only cover a tiny number of real situations...

 

[Speakpigeon]

FDI here Speakpigeon Ready to argue.

Can't resist, hey?

Is Newton physical model a failure just because it isn't the best overall? Of course not. Neither are Tigers a failure just because they can't eat many humans any more. Its a problem with ideals when talking in evolutionary terms which is all about 'good enough'.

You missed the point. I was assuming that the design method was indeed evolution but clearly the engineers or whomever would still have some kind of objective, a purpose, which evolution doesn't and couldn't have by itself. So, good enough is good enough for evolution, I agree, and old posts of mine expressed the same ideas in the hope of educating the masses around here. However, for the engineer using evolution with the purpose of creating an artificial brain better than ours then my point stands that if they are indeed using an evolutionary process somehow similar to Darwin's they will get all sorts of things not better than a dog's brain, which, from their perspective, would be abject failure. And that, after a very, very, very long time so that they would die well before they get anywhere near human brain.

As for the OP it seems there's a bit of hand waving when it comes to 'evolutionary' realization of a circuit. Yes failures may be avoided but set aims for design product are prescribed making the realization engine something other than an evolutionary algorithm.

I guess.
EB

 

[fromderinside]

You missed the point. I was assuming that the design method was indeed evolution but clearly the engineers or whomever would still have some kind of objective, a purpose, which evolution doesn't and couldn't have by itself. So, good enough is good enough for evolution, I agree, and old posts of mine expressed the same ideas in the hope of educating the masses around here. However, for the engineer using evolution with the purpose of creating an artificial brain better than ours then my point stands that if they are indeed using an evolutionary process somehow similar to Darwin's they will get all sorts of things not better than a dog's brain, which, from their perspective, would be abject failure. And that, after a very, very, very long time so that they would die well before they get anywhere near human brain.

What was the point I missed. It must be somewhere in that convoluted thing you posted.

If its that I missed that you missed with your assumption then neither of us wrote anything useful here.

An engineer who got a result that mimicked evolution would declare success if his work did so even if it were a pigeon's brain capacity. That his result didn't come up to expectations would be a disappointment, but, not a failure. He could use results to do better which would be evolutionary in a kind of a unnatural, er scientific, way.

 

[Speakpigeon]

An engineer who got a result that mimicked evolution would declare success if his work did so even if it were a pigeon's brain capacity. That his result didn't come up to expectations would be a disappointment, but, not a failure.

I'm sure you could find some utility to an electronic circuit grown today by "Darwinian evolution", if at all possible, but you ain't going to get anything like a pigeon's brain anytime soon. Today's engineers will be long dead if ever it happens. Current geopolitical entities will have disappeared, possibly completely, maybe human kind will have evolved beyond recognition. The Earth itself could look very different, I don't know more like Venus or Mars.

Let alone anything like a human brain.

He could use results to do better which would be evolutionary in a kind of a unnatural, er scientific, way.

Sure, whatever, but nothing remotely like evolution, in a natural way, producing something like or better than a human brain.
EB

 

[Elixir]

Darwin on a chip: New electronic circuits mimic natural networks like the human brain
http://www.sciencedaily.com/releases/2015/09/150921133946.htm



Design-less chips that thus prevent(correct) design gliches. From research at University of Twente (pronun. ['twɛn tə]).

I suggest a bit of scepticism is in order here. Darwinian evolution took something like 3.5 billion years to produce a reasonably smart brain like ours using a lab the size of a medium-size planet called the Earth. I will assume that the chip here took at most a few months to produce and was "grown" inside a container not much bigger than a Petri dish. So while their conclusion is probably honest and perhaps correct, that does not really get us anywhere near producing an artificial brain as or more powerful than ours. Also, all other species should be regarded in this context as so many design failures in the Darwinian process of producing our brain (assuming our brain is indeed the top brain in the world).

Anyone wants to argue with that?
EB

I won't argue, but will question your comparison. "Darwinian evolution took something like 3.5 billion years..."
During (at least the latter parts of) that process, Darwinian evolution was time-constrained by generation cycles of decades. I assume that the process for producing circuits dealt with generation lengths MUCH shorter than that (story doesn't say, but could be as short as fractions of a second), and the selective criteria were 100% singular and consistent, unlike what happens in nature.

 

[Speakpigeon]

I won't argue, but will question your comparison. "Darwinian evolution took something like 3.5 billion years..."
During (at least the latter parts of) that process, Darwinian evolution was time-constrained by generation cycles of decades. I assume that the process for producing circuits dealt with generation lengths MUCH shorter than that (story doesn't say, but could be as short as fractions of a second), and the selective criteria were 100% singular and consistent, unlike what happens in nature.

Alright, it's a fair point. Yet...

Yet the chip's evolution here may thave to be compared with the evolution of cells, with much shorter generations (though not a fraction of seconds). As the chip would be grown into something a bit more like a brain, it would grow in size and have to interact with a much larger and much realistic environment which would be much more time consuming than the initial stages. It could go faster than for brains, perhaps, but not likely so fast that this could compensate for the small size of the environment and small number of the systems involved in this experiment. I guess it all comes down to the difficulty of doing things like nature does them and on the same scale.
EB