According to this article, the brain has implemented epoll(4). Wake up the neuron when you want to use it. The upper bound is tied to how many active firings are happening concurrently.
What this article leaves out is that sparse coding is most common in neocortex (the wrinkly outside part of your brain). In structures underneath there are a lot of brain cells that fire away all the time: the thalamus, the basal ganglia, the dopamine cells in the midbrain that fire like pacemakers. I've recorded from the cerebellum in deeply anesthetized animals; there is still a lot of spiking going on in there. Presumably the same thing is true for humans. So sparse coding is nifty--and people have made some cool algorithms with it, cf. Cris Rozell's work--but it doesn't mean you're not using your whole brain
She totally does say they're cortical neurons. But it seems the point of the article is that most of the brain is dormant. I'm not sure that most of the brain is dormant.
> The quiet neurons might be doing more than refining perceptions. Though they spike infrequently, we know from intracellular recordings that they still receive inputs from other neurons, which causes their membrane voltages to fluctuate. The sum of these fluctuations and spikes constitute what’s commonly known as brain waves. Over the past 15 years scientists have begun to amass evidence that these brain waves play an active role in information processing, shunting some neural inputs while enhancing others, for example, or altering the timing of spikes. This suggests that spikes are not the sole information-carrying signal in the brain, and that, in turn, the “inactive” neurons are doing much more than it seems.
Software neural networks are a poor approximation of biological ones, but it may still be interesting to compare spike frequency distribution to see how general this observation is.
"The human brain, which accounts for less than 2 percent of the body’s mass, uses 20 percent of its calorie budget, or three bananas worth of energy a day. That’s remarkably low, given that spikes require a lot of energy."
What is amazing about this fact is that this is only 20 watts of power. The brain computes at 10 quadrillion calculations per second. While the largest supercomputer (China’s Tianhe-2) is at 34 quadrillion and has "beat" the human brain, it comparatively runs at 24 megawatts of power [1]
Brains don't do FLOPS. It is a completely different type of "calculation". Computational benchmarks are only directly comparable when the representations of input and output are the same.
That's very true, but in some ways it's even more remarkable: there exists an alternative form of calculation/computation that is "magical" in certain domains when compared to all the standard CS we know, and we don't have a clue what it is.
It's not an apple/apple comparison. It's an apple/unicorn comparison.
I don't know, if counted every modulation in every wire in a computer (CPU, RAM, cache activations, etc), you might get a signifant multiple of its FLOPS count, right?
Yes! This was the biggest flaw in The Matrix (which I otherwise love): it had machines using human bodies as power sources whereas they could have used human brains for computational power leading to huge power efficiency.
Think about somehow hooking up a brain to a computer which can interpret the neural code. Then you have the best known general purpose AI machine running on 20W. Now combine hundreds, thousands of them. Not human brains, of course, but say monkey brains. DARPA already conducted some experiments in this area (http://www.kurzweilai.net/darpa-links-brain-waves-sensors-an...)
>it had machines using human bodies as power sources whereas they could have used human brains for computational power leading to huge power efficiency.
That was apparently the original intent, that the Matrix used human brains as a neural net. But the script was changed to "humans = batteries" because of fears audiences wouldn't be able to comprehend it[0].
The brains-as-the-Matrix conceit would have been much more powerful. Damn. Lots of symbolic value, and a statement on human culture, human nature, and complacency vs. untapped potential.
I'm going to go ahead and pretend that's what actually happens. :)
It is possible that when Morpheus described the 'true purpose' of the Matrix, he didn't actually know what he was talking about. I think that's the excuse most people use. It was a best guess based on piecemeal information (and possibly deliberate misinformation by the machines.)
Perhaps it is a misunderstanding on the part of the humans in the film. They could be mistaking the human brains powering the neural network by means of providing processing capacity for powering it by means of providing energy.
Thanks for the link, fascinating discussion in these answers. I found this very relevant to this point: http://scifi.stackexchange.com/questions/12223/why-did-the-m.... One other problem with the humans as batteries thing (in addition to its inelegance) is that the machines could have used cows or pigs to much better effect.
This comment wraps it up nicely: "All the machines ever wanted was to be left alone, not destroy humanity. I suppose this is the most humane solution they could think of"
This is just speculation, but I'd think it would be important for the brain to have a way to manage it's memories... you wouldn't want memories just randomly appearing, all the time There has to be some way to orchestrate memories, and I wouldn't be surprised if the more active neurons are involved in that orchestration. In other words, couldn't those exceedingly active neurons be contributing to suppressing memories?
It would be interesting if active neurons were suppressing memories but I don't know of any evidence in support of that idea.
Most of the articles cited by the author don't deal with neurons that are directly involved in memory. The articles described the activity of neurons in primary sensory areas: the neurons in cortex that receive direct input from the periphery, or at least are the closest to neurons in the periphery in terms of number of connections between them. One exception is the article she cites about the hippocampus, but this area is required only for forming memories--once the memories have formed, they live somewhere else, but probably not in the primary sensory areas. Okay, maybe memories are a distributed network that includes neurons in primary sensory areas...but it would be weird to also have the main sensory area also be the place where memories live.
It might make a lot of sense actually. It could serve to speed up instinctive responses to stimuli which have been encountered before, i.e. whipping your finger away from a hot surface.
I think I understand what you're getting at, but the example you gave is not a learned act, it's a reflex action. It bypasses your brain entirely, you generally recognize the pain after you reflexively withdraw your hand.
If neurons are polymorphic functions, not keepers of state, it would stand to reason that certain functions at the root of a dependency tree would get more use than some obscure branches. Is it overreaching to suggest that object orientation has been detrimental to our understanding of cognitive processes?
I can think of so many memories of images of brains drawn into regions. Usually for comedic effect. But the symbol is powerful. Different 'parts' of the brain know or do things related to a specific domain. To imagine the brain does crazy things like biological macros is harder to draw in 2D, but likely much more accurate to the real world than having a portion of my brain dedicated to XYZ...
24 comments
[ 4.1 ms ] story [ 58.5 ms ] thread> This energy budget puts limits on how much of the cortex can be engaged at once
I passed over it during my first read through as well. But at certain parts the author did state explicitly that these were cortical neurons.
> The quiet neurons might be doing more than refining perceptions. Though they spike infrequently, we know from intracellular recordings that they still receive inputs from other neurons, which causes their membrane voltages to fluctuate. The sum of these fluctuations and spikes constitute what’s commonly known as brain waves. Over the past 15 years scientists have begun to amass evidence that these brain waves play an active role in information processing, shunting some neural inputs while enhancing others, for example, or altering the timing of spikes. This suggests that spikes are not the sole information-carrying signal in the brain, and that, in turn, the “inactive” neurons are doing much more than it seems.
What is amazing about this fact is that this is only 20 watts of power. The brain computes at 10 quadrillion calculations per second. While the largest supercomputer (China’s Tianhe-2) is at 34 quadrillion and has "beat" the human brain, it comparatively runs at 24 megawatts of power [1]
[1] http://waitbutwhy.com/2015/01/artificial-intelligence-revolu...
It's not an apple/apple comparison. It's an apple/unicorn comparison.
What exactly is that based on? Does it count anything other than synapses?
Think about somehow hooking up a brain to a computer which can interpret the neural code. Then you have the best known general purpose AI machine running on 20W. Now combine hundreds, thousands of them. Not human brains, of course, but say monkey brains. DARPA already conducted some experiments in this area (http://www.kurzweilai.net/darpa-links-brain-waves-sensors-an...)
That was apparently the original intent, that the Matrix used human brains as a neural net. But the script was changed to "humans = batteries" because of fears audiences wouldn't be able to comprehend it[0].
[0]https://scifi.stackexchange.com/questions/19817/was-executiv...
I'm going to go ahead and pretend that's what actually happens. :)
Not actually canon but it'll do.
This comment wraps it up nicely: "All the machines ever wanted was to be left alone, not destroy humanity. I suppose this is the most humane solution they could think of"
Most of the articles cited by the author don't deal with neurons that are directly involved in memory. The articles described the activity of neurons in primary sensory areas: the neurons in cortex that receive direct input from the periphery, or at least are the closest to neurons in the periphery in terms of number of connections between them. One exception is the article she cites about the hippocampus, but this area is required only for forming memories--once the memories have formed, they live somewhere else, but probably not in the primary sensory areas. Okay, maybe memories are a distributed network that includes neurons in primary sensory areas...but it would be weird to also have the main sensory area also be the place where memories live.
source: https://en.wikipedia.org/wiki/Withdrawal_reflex
I can think of so many memories of images of brains drawn into regions. Usually for comedic effect. But the symbol is powerful. Different 'parts' of the brain know or do things related to a specific domain. To imagine the brain does crazy things like biological macros is harder to draw in 2D, but likely much more accurate to the real world than having a portion of my brain dedicated to XYZ...