Terabyte Hard Drives

[fullerene]

100 billion neurons. Only 10 billion are in the cortex. Each of which has the potential to have an axon going to any of a million of its nearest neighbors, but no more than a few thousand axons total. Let's say 10 thousand of its nearest million neurons. That's a binary decision, whether to have a connection to a given other neuron. In a typical connection set, making full use of capacity, each will then be connected to 10k of 1M neurons and so each neuron carries about 20 bits times 10k connections or 200 kilobits of information. Multiply that by 10 billion and you get 2 petabits.

What makes you think that the information in the brain is stored simply by whether or not a neuron is connected to another one? It seems like an awfully strong claim to make about memory, and, quite frankly, I think you're wrong. It's not the best source, but it's literally the first one that pops up when you google "how does memory work", and I don't want to be bothered to search for actual research papers on the subject (it's a pain in the ass). From that site, though:

First of all, [researchers studying brain patterns of people remembering something] observed a high correlation between the initial cerebral activity pattern (when the participants were shown the images) and the later cerebral activity pattern (when the participant were recalling the images). This brain activity isn't localized in a certain place in the brain, it has a very complex pattern - in other words, a certain memory isn't stored in a certain place, it is a complex combination of reactions

Oops... information storage in the brain isn't such a binary decision after all. If it comes down to memory being the brain "reliving" the experience, I think it would be better modeled by the fact that when a neuron fires (where the memory/experience "begins" in the brain), it could either triggeer any neuron that it's connected to, or it could not--depending on the experience. Using your numbers, a memory could start at any one of those 10 billion neurons in the cortex, but each neuron can trigger any one of the 10k neurons which it's connected to. Supposing that an experience can be relived/remembered by a 2-neuron firing sequence, you're talking about (10 billion)*(10k) = 10^14 "states of memory recall"

But no one is foolish enough to think that an experience only triggers two neurons, so let's extend that to three. now you're looking at (10 billion)*(10k)^2, because any one of those neighbor-neurons can trigger its neighbor neurons. There's no double-counting, because there's no reason to suspect that a neuron can only fire once during a given memory. Now you're up to 10^18 states of memory.

Assuming that any memory can be stored as a single bit (which is a ludicrous underestimate), and each memory only traverses three neurons to recreate an experience (which is another ludicrous underestimate) you're already up to an exabit of storage space.


Scientists with high degrees who talk about subject matter outside of their specialty are like... like people who talk about science without using math. They should really just keep their mouth shut or learn about it before they end up saying something stupid.

(btw, I'm relatively sure flow was exaggerating. Hyperbole is quite common in the way most people speak, so it's probably not helping much by showing how factually inaccurate his exaggeration was)

 

[Latro]

Sandor--there's quite a bit wrong there, but all I'm gonna touch on for now is your Schrodinger's equation comment. Let's estimate what that would actually take. If I recall, each electron you add to a system adds 4 more partial differential equations in as many unknowns to Schrodinger's equation. Let's say a human weighs 60 kg. That's 6*10^4 grams, so that's about 4*10^28 amu. So we have around 4*10^28 electrons. Multiply by 4 and you have 2*10^29 PDEs in 2*10^29 unknowns. That's not gonna fit in anywhere near an exabyte, not to mention that once you have it stored you have a NIGHTMARE of a problem trying to solve it numerically.