There is the Theory of Everything and we're nearly certain it exists, we're really really close, we can smell it. Until then, we have two not so simple almost rules, quantum chromo-du-jour and pesky relativity. That's just my opinion of course. Who knows, it could be cymatic patterns carving analog ai onto the nerve cell surface, or a schoolmaster with a wooden pointer.
Or the universe as we perceive it is a fabrication of our time-bound perception and any "Theory of Everything" is bound to fall short as reality is beyond our human perception. I would be interested to see where experiments with time crystals take us as they could be a key to discovering the true nature of reality not bound by time.
The title is clickbait. They've found something that seems to be true for all synapses (even though the article is too vague to make that out, and it only seems to be valid for pyramidal cells; these are important though), and it's about some aspect of plasticity: synaptic change can be predicted from calcium levels. This is not a learning rule, but a small part of the learning process. It also seems to offer a new direction in learning research, namely the influence of one synaptic change on other synapses. So it's interesting, but at a very technical level.
Interested to see how this fits with Hawkin's 1000 Brain theory which was focused on cortical columns as modelling units. This seems to step down a level to individual voting elements.
Well, step down a few levels, if you want to put it that way.
This model describes how connections between individual neurons (real physical cells, not abstract "voting units") work. Inter-neuron connections (aka synapses) are the basis of long term memories/learning.
The human cortex is a thin sheet (scrunched up in our head) that would stretch out to about the size of a tea towel, that has 6 layers of neurons, with a specific inter-layer connect pattern.
A cortical column is what you'd get if you took a cookie cutter and cut a small cylindrical section, like a cookie, through the cortical sheet. It'll have those 6 layers of neurons, but maybe 100 neurons total (not 6) depending on what size cookie/column you're talking about.
Any model of the brain (cortex) based on cortical columns is going to be a high level functional model of that mostly abstracts away what the individual neurons are doing. At this level of abstraction, you don't really care about the detailed behavior of individual neurons, since you're assuming that their aggregate behavior can adequately be described by your higher level model.
The title is very broad and with it's postulate hints at a massive overclaim, but this paper seems to explain something significant and interesting about how synapses work. None of which generalizes to "the brain" or "learning" really.
"What underlies learning in the brain might be actually simpler than previously thought despite the brain being one of the most complex objects in the known universe. A collaboration of Scientists led by the EPFL Blue Brain Project has achieved a major advance in accurately simulating the synaptic changes thought to implement learning in the neocortex, opening the door to greater understanding of learning in the brain."
“is this neuron going to spike and if it does can we say why” is, though.
“if external threats are removed” is an attempt to linearize, and therefore remove chaotic behavior. Sometimes you can pull this off and have a reasonable model at the end; sometimes you cannot. This tells you more about your model than it does about the system you’re studying.
Name me a stable system that does not become destroyed when an overwhelming amount of energy is applied to it. Is a tortoise with a stick of dynamite attached on a random timer chaotic or not? Seems yes. So what are the limits that define chaos?
By that you could say all is chaotic. We lack a suitable definition here.
A triple pendulum at rest now will be at rest 1s from now.
A triple pendulum moving now will likely be moving 1s from now.
That doesn't mean that the triple pendulum isn't a chaotic system. It just shows the chaotic nature of the system doesn't manifest at the level of detail you're considering.
Observing choas, one of the biggest hallmarks is its extreme stability. These systems always evade explanation and predictability at the smallest scales and build wild persistent structure at the largest. Chemical reactions are messy, but this one has been going for 4 billion years. We struggle to explain how even as high a scale as atoms can create the entire planet we see, let alone quantum fields, but understanding what it looks like just needed a few good thoughts about evolution. Life is the pinnacle of choas.
Denying a hypothesis without any proof against it is actual pseudoscience, which is what you’re implying. Biological systems are complex and chaotic, but a single mechanism exists to replicate dna or make protein so what exactly are you supposing here?
Err, I dont think a single mechanism exists either to make DNA or proteins... so I'm not sure where that comes from.
However in the case of "learning", we're talking about the macro and micro physical structure of the brain encoding vast amounts of sensory-motor models of its environment, etc. We're talking about the brain-body-environment as the site of mood, emotion, desire, belief, imagination, etc. We're talking about that thing when given vallium (a simple chemical effect) has extremely complex net effects on mood, desire, belief, etc.
Inasmuch as CO2 has likewise, extremely complex effects.
The statement that the relevant actions of the brain-body system here are complex is scientifically uncontroversial. Yet, pseudoscience continues -- because some researchers ignore (or are ignorant) of the methodological implications.
They're saying "none of those are simple" and also tossing in some examples of how much the brain does and how easily it's disrupted. I don't really know why you're having such a hard time of it.
What does climate have to do with the brain? How are you drawing an inference between how the climate works and how the brain works? I agree with others that you’re propagating pseudoscience. Even if you’re right, and there isn’t a simple rule that explains human learning, you’re right for the wrong reason.
But the brain isn't a complex system because the climate is a complex system. There is no relation between the two systems, other than that the two are emergent phenomena on Earth. Proof that the climate is a complex system doesn't automatically prove the brain is a complex system. Apriori, you can't draw any logical conclusions about the brain from the climate. You're right about one thing: It's amazing the confidence people have
The article is about whether or not human learning is powered by a simple, understandable process. Your response is: “No, human learning is like the climate, it’s complex.” I’m just pointing out what a nonsensical statement that is. The article offers evidence about how synapses work. What is your analogy, exactly? Usually an analogy has some points of comparison that help illustrate a topic. E.g. chess is like war. There are two sides in both war and chess, both war and chess require anticipating your opponents moves etc. Therefore, you can make the argument that you can understand something about war through understanding chess. I’m really failing to see how the climate helps us understand the brain. Other than your claim that both are beyond comprehension.
Ok, but clearly this article is arguing that in fact, human learning is not a complex system, at least in terms of synapse function. What the Wikipedia article says is irrelevant to this new information. Again, there is no connection between the brain and the climate. What if we prove the brain is not a complex system? Does that mean the climate is not a complex system either?
a flexible framework for studying cortical learning algorithms in
silico
What does this refer to? Are they talking about (mechanical) simulation of a brain model in a computer program, biological experimentation with chemical processes on a silicon substrate, or about translating these findings to current machine learning systems?
TLDR; There are lots of types of neurotransmitters and lots of types of neurons, even within the general class of Pyramidal neurons. There seem to be large variations in how and when the types of neurons form and modify synapses (connections) between them. What if there were only a few factors that govern most of that variation? Using computer simulation the authors focus on calcium as a signaling mechanism, the size of synaptic spines, and known behavior of pre and post synaptic structures to create a predictive model of how and when synapses will be modified. They then compare their simulated results and model with results taken from real-world experiments and get good agreement.
Kind of off topic from the article, but here’s another scenario to add to the AI Risk conversation:
Sentience (along with consciousness, sapience, and qualia) turns out to be undefinable woo, leading to a declaration that AI is not sentient and neither are humans, and so rather than treating machines like people, we end up treating people like machines.
Yeah, yeah, we already do treat people like machines etc. etc., but we risk further dehumanization if we decide the human mind is nothing special after all.
I find that the term "woo" suffered a semantic drift in the last few decades. It used to refer to things that pretend to be scientific by using the terminology and aesthetics of science without being grounded in real empiricism. One classical example of "real woo" is homeopathy.
Nowadays, "woo" is being applied to anything that science cannot model or explain. Science cannot explain consciousness? Well it must be woo then! You tell me that you experience your own consciousness, and that you are sure it exists? No, it's an illusion my poor child. An illusion of whom, you ask? That is starting to sound like philosophy, which is not science so it must be woo. Careful.
Woo is becoming synonymous with "heretic" and "science" no longer means what I thought it meant. In my view, the defining aspects of science, the ones that make it such a splendorous endeavor, are modesty and doubt. All scientific theories are under the permanent risk of being falsified, otherwise they are not serious theories. There are many things we do not understand, and it might even be that this will always be the case. Some things defy our current understanding or reality, and consciousness is one of them. That doesn't make it woo, and those that pretend that it does are like the fearful religious people of previous ages, as they shout "woo"/"heretic" at those who have the temerity of harboring doubts and questions.
This question has been addressed more than decade ago by people like Harel Shouval and Nelson Spruston. Notice the resulting form of STDP which is not quite what is seen in experiments (Whether STDP is a fundamental plasticity rule, or if it even happens in vivo is another question that has been debated). There are a lot of plasticity models proposed, this is another one. Good, but we actually need more experimental data in order to tame the beast.
41 comments
[ 3.0 ms ] story [ 71.2 ms ] threadThis model describes how connections between individual neurons (real physical cells, not abstract "voting units") work. Inter-neuron connections (aka synapses) are the basis of long term memories/learning.
The human cortex is a thin sheet (scrunched up in our head) that would stretch out to about the size of a tea towel, that has 6 layers of neurons, with a specific inter-layer connect pattern.
A cortical column is what you'd get if you took a cookie cutter and cut a small cylindrical section, like a cookie, through the cortical sheet. It'll have those 6 layers of neurons, but maybe 100 neurons total (not 6) depending on what size cookie/column you're talking about.
Any model of the brain (cortex) based on cortical columns is going to be a high level functional model of that mostly abstracts away what the individual neurons are doing. At this level of abstraction, you don't really care about the detailed behavior of individual neurons, since you're assuming that their aggregate behavior can adequately be described by your higher level model.
Biological systems are complex and chaotic. Denying this so that "any old model will do" is pseudoscience; and we should start calling this out.
Are they? Seems they are extraordinarily stable. Tortoises can last two centuries, that's not the signature of chaos. What do you mean by that word?
Chaos: When the present determines the future, but the approximate present does not approximately determine the future.
Lorenz
A tortoise alive now is 99.995% likely to be alive tomorrow, if external threats are removed.
Pretty sure that is not the behaviour of chaos under your definition.
“if external threats are removed” is an attempt to linearize, and therefore remove chaotic behavior. Sometimes you can pull this off and have a reasonable model at the end; sometimes you cannot. This tells you more about your model than it does about the system you’re studying.
By that you could say all is chaotic. We lack a suitable definition here.
Sure. So what?
That doesn't mean that the triple pendulum isn't a chaotic system. It just shows the chaotic nature of the system doesn't manifest at the level of detail you're considering.
However in the case of "learning", we're talking about the macro and micro physical structure of the brain encoding vast amounts of sensory-motor models of its environment, etc. We're talking about the brain-body-environment as the site of mood, emotion, desire, belief, imagination, etc. We're talking about that thing when given vallium (a simple chemical effect) has extremely complex net effects on mood, desire, belief, etc.
Inasmuch as CO2 has likewise, extremely complex effects.
The statement that the relevant actions of the brain-body system here are complex is scientifically uncontroversial. Yet, pseudoscience continues -- because some researchers ignore (or are ignorant) of the methodological implications.
It's amazing the confidence people have
The climate was an analogy to illustrate the point
They're both complex because they meet the definition of a complex system
What does this refer to? Are they talking about (mechanical) simulation of a brain model in a computer program, biological experimentation with chemical processes on a silicon substrate, or about translating these findings to current machine learning systems?
That one.
Sentience (along with consciousness, sapience, and qualia) turns out to be undefinable woo, leading to a declaration that AI is not sentient and neither are humans, and so rather than treating machines like people, we end up treating people like machines.
Yeah, yeah, we already do treat people like machines etc. etc., but we risk further dehumanization if we decide the human mind is nothing special after all.
Nowadays, "woo" is being applied to anything that science cannot model or explain. Science cannot explain consciousness? Well it must be woo then! You tell me that you experience your own consciousness, and that you are sure it exists? No, it's an illusion my poor child. An illusion of whom, you ask? That is starting to sound like philosophy, which is not science so it must be woo. Careful.
Woo is becoming synonymous with "heretic" and "science" no longer means what I thought it meant. In my view, the defining aspects of science, the ones that make it such a splendorous endeavor, are modesty and doubt. All scientific theories are under the permanent risk of being falsified, otherwise they are not serious theories. There are many things we do not understand, and it might even be that this will always be the case. Some things defy our current understanding or reality, and consciousness is one of them. That doesn't make it woo, and those that pretend that it does are like the fearful religious people of previous ages, as they shout "woo"/"heretic" at those who have the temerity of harboring doubts and questions.
https://en.wikipedia.org/wiki/Betteridge%27s_law_of_headline...
I'm really interested in the whole history of the Blue Brain project.