Researchers have created an artificial synapse that requires just 1.23 × 10-15 Joules per "synaptic event":
An artificial synapse that emulates a biological synapse while requiring less energy has been developed by Pohang University Of Science & Technology (POSTECH) researchers in Korea. A human synapse consumes an extremely small amount of energy (~10 fJ or femtojoules per synaptic event), and an entire human brain consumes only as much energy as a domestic light bulb, but can outperform a supercomputer in many aspects, according to the researchers.
The researchers have fabricated an organic nanofiber (ONF), or organic nanowire (ONW), electronic device that emulates the important working principles and energy consumption of biological synapses while requiring only ~1 fJ per synaptic event. The ONW also emulates the morphology (form) of a synapse. [...] The researchers say they have emulated important working principles of a biological synapse, such as paired-pulse facilitation (PPF), short-term plasticity (STP), long-term plasticity (LTP), spike-timing dependent plasticity (STDP), and spike-rate dependent plasticity (SRDP).
Organic core-sheath nanowire artificial synapses with femtojoule energy consumption (open, DOI: 10.1126/sciadv.1501326)
(Score: 3, Funny) by jmorris on Tuesday June 28 2016, @11:23PM
So now when the robots rise up they will at least be green.... until they are covered in our blood.
(Score: 2) by LoRdTAW on Wednesday June 29 2016, @11:42AM
So more or less brown then.
(Score: 3, Funny) by wonkey_monkey on Tuesday June 28 2016, @11:27PM
Suck it, Mother Nature.
systemd is Roko's Basilisk
(Score: 2, Disagree) by Dunbal on Wednesday June 29 2016, @12:12AM
Millions of years of evolution vs a nerdy basement-dwelling photophobe. Mother nature usually doesn't like wasting energy on useless stuff. For now my money is on "something you just haven't thought about yet" as to why higher energy = good.
(Score: 5, Insightful) by Immerman on Wednesday June 29 2016, @01:18AM
Actually, as a rule mother nature is horribly inefficient. Photosynthesis is typically 0.1-2% efficient at energy conversion, radically less than even the 10-15% of early commercial solar panels. Human muscle cells are typically 18-26% efficient at converting energy to work, much less than an electric motor which can easily exceed 90%.
It's one of the great weaknesses of blind, randomness-driven evolution that it gets trapped in local maximums, while intelligent design can simply look at the problem space and work toward truly optimal solutions, despite the fact that early implementations will generally be considerably less efficient than the cobbled-together crap that's been fine-tuned over millions of generations.
(Score: 0) by Anonymous Coward on Wednesday June 29 2016, @08:17AM
This article says 1st stage photosynthesis is 90% efficient AND badass quantum powered
http://www.economist.com/blogs/economist-explains/2014/12/economist-explains-1 [economist.com]
(Score: 2) by Immerman on Wednesday June 29 2016, @01:36PM
If true, then nature got lucky with it's starting point for that particular aspect of photosynthesis, it happens sometimes. Or perhaps it's simply one of those systems that has few local maximums to get stuck in. It doesn't change the fact that only a tiny fraction of available energy actually gets converted to chemical energy which is available to the plant.
(Score: 0) by Anonymous Coward on Thursday June 30 2016, @02:10AM
Electric motors don't build or repair themselves though. If you add portable nanobots and factories for self rebuild and repair it might not be as efficient.
(Score: 0) by Anonymous Coward on Thursday June 30 2016, @03:20AM
I'd like to see how efficient really is an electric motor or solar panel system that can self upgrade and repair while in operation.
There are plenty of inefficiencies but you can just look at a housefly and compare it to a quadcopter and know how far behind we are technologically. The former can self-navigate, find fuel and refuel, reproduce, sense and avoid collisions/attacks and has a far better flight time. People talk about AI etc but are we even capable yet of 100% simulating/modelling the thinking/information processing of a white blood cell? A model that is 90% accurate in observable behaviour does not necessarily prove you understand how stuff actually works. I can provide you a 90% accurate model of Stephen Hawking too, it's not like he moves that much ;).
(Score: 2) by dyingtolive on Wednesday June 29 2016, @03:48AM
I'm not sure how well it translates, but far as I'm concerned, my flabby, works-at-a-desk-all-day ass could use biological systems that consume MORE energy to keep active.
Don't blame me, I voted for moose wang!
(Score: 0) by Anonymous Coward on Wednesday June 29 2016, @03:51AM
Mother nature doesn't care about maximizing efficiency, it just has to be just efficient enough to not be too problematic.
(Score: 0) by Anonymous Coward on Wednesday June 29 2016, @01:34PM
Well, I guess that this artificial synapse cannot grow. The main way our brain achieves its enormous plasticity is by growing new synapses.
(Score: 1) by stretch611 on Wednesday June 29 2016, @03:59AM
Replace my neurons right away...
With 1/10th the energy requirements, i can be 10x lazier.
Now with 5 covid vaccine shots/boosters altering my DNA :P
(Score: 2) by LoRdTAW on Wednesday June 29 2016, @11:52AM
So we're the DTL/RTL/TTL of the neuron world then.
(Score: 0) by Anonymous Coward on Wednesday June 29 2016, @03:16PM
In principle, a major challenge for AI is the question of whether or not it is cheaper to run than humans.
On the face of it, you could calculate this in terms of energy, maintenance, training and other inputs and determine in simple output terms whether Burger King would rather employ teenaged paper hat wearers to flip burgers, or an AI-driven burgerbot.
Part of the problem is figuring out how much intelligence you need. The intelligence needed to flip and turn that eternal chain of burger patties is not human equivalent. In principle, you could train a dog to do it. In actuality, the actual burger-flipping part is probably manageable with a cockroach-equivalent AI. The rest of the teenager's brain is basically wasted at the grill - or at least alternatively employed, thinking about sex, new dirty jokes about the manager, and where to spend the minimum wage earned from the job.
The investment in training and proving out the burgerbot will be considerably higher, of course, but once done the burgerbot can be duplicated conveniently.
AI for more complex tasks, such as driving, navigating a cargo vessel, playing Go, or predicting the weather is higher order activity that requires a lot more neurons, but still falls short of full machine cognition. These problems are not AI-hard.
We still have no (published) structure or architecture that supports an AI that can produce or sustain a culture. We have bits and pieces, and unfortunately a lot of those pieces are big and expensive in computing terms. Sure, we're good at finding efficiencies, but even so there is no particularly cheap way of slapping together a driving AI. What's worse, if we use a neural structure, it's very hard to provide plausible engineering guarantees on its quality since we have no way of proving what a neural net actually does. This turns out to be a political, if not a practical problem (who'm I kidding, it's a practical problem as well).
Let's suppose that we conquer all those problems, and produce a human-equivalent AI for one tenth the maintenance cost of a human being, even allowing for big changes in the price of energy. What does this actually buy us in practical terms?
We can use AIs where it is not convenient to use people. Space exploration. Hazardous environments. Whatever. But the political problem of the encroaching unemployability of humans rapidly turns it into a situation where the negative externalities of mass indigence weigh much more strongly against AI than we would think at first. It's all very well to talk about a post-scarcity society or a universal basic income while the police are replaced with peacebots, but the unpredictable downsides of a population of lotus-eaters are likely to be vast and, based on a cursory review of human history, destructive in nature.
We have no plausible way of preventing the automation of burger flipping. But it would make a lot of sense to establish the idea that full AIs have employment rights equivalent to those of human beings, including full remuneration with consideration of overtime.
The real costs are almost never shown in the receipt.