Simulating brain function with hardware or software

[Nick R]

I have very limited knowledge of how brains work, and I am being highly speculative, but from what little I know (correct me where I'm wrong) brains are a sort of massive interconnected network of signal receiving/transmitting units. It seems that the "paths" of logic would be crossed - that is a change somewhere could affect the logical path of signals far away from that change, perhaps even all signal paths.

Has there ever been any sort of research into making some sort of special analog computer-like hardware that simulates brain activity or some sort of software that does this? Say by, creating some hardware unit or software construct that mimics the function of a grasshopper brain neuron, and interconnecting them in exactly the same configuration as they are connected in said grasshopper? (perhaps the connections themselves would need to model the connections within the grasshopper brain in speed and function)

I am aware there are such things as "artificial neural networks" but a search on this indicates that they are typically not used to model actual brain function but rather to solve problems such as simple pattern recognition of very limited scope.

 

[potato991]

Yes, the blue brain project is attempting to simulate neocortical columns of rats (running on blue gene supercomputer).

http://www.seedmagazine.com/news/2008/03/out_of_the_blue.php?page=1

I can't help think they should try for something a little less ambitious initially, to improve the chances of success. Like simulating the C-elegans worm, which has a very basic neural structure (or even a grasshopper, as you mentioned).

 

[-Job-]

Yes, the blue brain project is attempting to simulate neocortical columns of rats (running on blue gene supercomputer).

http://www.seedmagazine.com/news/2008/03/out_of_the_blue.php?page=1

I can't help think they should try for something a little less ambitious initially, to improve the chances of success. Like simulating the C-elegans worm, which has a very basic neural structure (or even a grasshopper, as you mentioned).

That's interesting because if they succeed it might go some way towards ruling out the idea that brains make use of quantum effects for solving certain hard problems (which i find very speculative and not really grounded in evidence).

For example, some problems are easily solved by the human brain but are computationally infeasible for traditional computers, image recognition falls in this category. Image recognition as a computational problem is almost provably infeasible in any traditional (non-quantum) computer (solutions are expected to take in the order of 2^n operations, where n is the image size).

Because quantum computers have been proved to be able to quickly solve certain hard problems that we haven't been able to solve in traditional computers, and because the human brain seems to be able to quickly solve certain problems that we're almost convinced can't be solved efficiently has led some people to suggest that the human brain is taking advantage of quantum effects that grant quantum computers their processing capabilities.

Personally, i think this is incorrect for a few reasons. I don't think that human brain can solve all instances of image matching efficiently, especially because certain other problems that are provably equivalent in complexity to the image-matching problem (such as the clique problem) are extremely difficult for humans to solve efficiently for even moderate sized instances.

The second reason is that even quantum computers aren't expected to be able to solve NP-Complete problems, so realistically even Quantum Computers may never be able to solve all instances of image matching efficiently. Because of this i don't see any evidence to expect that the human brain might be acting as quantum computer, since it probably wouldn't help much, at least in solving image matching.

I think that the human brain is just good at performing image matching on instances of very limited size and complexity (in other words the type that traditional computers are able to solve as well).

Anyway, the reason why i find it interesting to have blue gene attempt to simulate a brain is that if the brain does make use of quantum effects (and i wouldn't think this would be exclusive to humans if it is there at all, but i suppose it's not impossible) then Blue Gene must fail. If it succeeds then it proves that at least the mouse brain does not use quantum effects - the reason for this is that it has been proved that simulating the quantum world is a problem that falls in the complexity class of BQP and that blue gene will never be able to solve even if it were thousands or millions of times more powerful.

 

[888eddy]

they will be working on a mouse or rat brain because they are very similar to a human brain (obviously without the capacity of our brains though. the same way a slow 10 year old pc will work on the same principles as a fast modern computer) and therefore will produce useful results. a worm or grasshopper brain would provide no useful information other than yes it can, or no it cant, be done which doesn't justify the expense and time taken for the project.

im not an expert and don't really know much about the level of research on the subject so its time for me to be speculative (since the terms being thrown around a lot on this discussion!):

i think computers as we know them are bad at image recognition etc.. because they are designed using methods invented to simplify them. e.g. binary is used because it makes doing electronic calculations convenient, all the cpu can do is multiply divide and compare different numbers then store the answer or retrieve previous answers to use in new calcs (not quite that basic but its the principle behind it). the cpu does this because it makes it simple enough that we can write algorithms that create the outcomes we desire based on our inputs. the brain however has devoped through natural selection rather than design and is therefore closer to the optimum. this also makes it hard for us to understand its functions since its difficult to simplify into something we can comprehend.

i should imagine half the problem is that we can't understand the brain more than we don't know how it works. however the people doing this research obviously know a lot more about the subject than we do so i should imagine they are headed in the right direction.

on a slightly digressive note i think its interesting to think about how this sort of research might affect computer programing or even their entire design. if the brain computes images so quickly and a super computer takes forever then it seems there is a signifcant performance loss due to the simplification of pc's. i wonder how much optimisation could improve performance by? however i suspect the loss of simplicity between hardware and software programing would make producing software for such a computer near imposible

closer to the subject however if they know enough about a rat brain to simulate it on a super computer then they must have a map of a rat brain and if they have a map of one then surely they have worked out how each microscopic part of the brain processes inputs and outputs and if they've done that then surely they know whether it uses quantum effects or not? which they clearly don't so my point is how can they know enough about a rat brain to simulate it?

 

[-Job-]

888eddy, my guess is that they're emulating neurons in a very simplified manner (no complex interactions) to avoid even more complexity and because that's where it makes the most sense to start.

In some sense the human brain must "make use" of quantum effects because it's a physical system, just probably not in the way a quantum computer does (one argument being that the human brain is too noisy for this) and so can't act as a quantum computer.

People have studied how neurons process signals for a long time so they know to simulate them (and the physical processes that are involved), but whether there are interactions between groups of neurons, components of neurons or the brain medium that make use of quantum effects to achieve the computing capabilities of quantum computers, unlikely as it is at this time, we certainly don't know for sure.

The fact that this is the case i think is why some people speculate about the possibility of the human brain having quantum-processing capabilities and why I'm pointing out that if a typical brain had a component with the computational power of a Quantum Computer, then Blue Gene wouldn't be able to model it.

This is because the Quantum Computer exploits an exponentiality in the number of quantum states, which a traditional computer would need to keep track of in order to simulate it. This would take forever since traditional computers are sequential computers (perform a single operation after the other). Blue Gene is a massive parallel aggregate of sequential CPUs, so though it has some static degree of parallelism it can't match the exponential degree of parallelism in a QC (with respect to its qbits), so in order to simulate even a small QC it would literally take forever. Hence, if Blue Gene succeeds, then at least the mouse brain doesn't use an embedded Quantum Computer for any relevant effect, if it has one at all.

When you say that one of the reasons traditional computers are not as powerful is because of their simplicity, i think there's some truth to that (although in my opinion it has nothing to do with binary, since binary is just a practical encoding of data equivalent to decimal for example, or any other encoding). The sequential nature of our computers are the reason why they're slow, but it's also what makes them predictable (reliable) and easy to program.

The brain is of a completely different architecture, not sequential at all, there's no central CPU, the network is the CPU and each node (neuron) has some processing capabilities, it's a distributed system - but i don't think it has, or needs, the same capabilities of a Quantum Computer.

The human brain has about 100 Billion neurons. Each neuron has the capability to perform an elementary operation (it also has trillions of dendrites, and each contributes to an operation) - so it has a far larger degree of concurrency than Blue Gene has. In addition to this there's the possibility of feedback interaction of these neurons (or groups of neurons) with the activity they generate, which isn't necessarily restricted to their immediate neighbors. The fact that this massive network can be adjusted to some purpose by trial and error (which happens naturally with the death and birth of neurons) means that some percentage of this processing power can be aligned and made to perform a task - so it's not surprise that it's so powerful and of such difficult simulation by our extremely sequential CPUs.

Also, i want to add that each neuron is probably no more powerful than a sequential machine in their computational power (if they're not QCs obviously :P), so if we throw enough sequential CPUs at this we should be able to perfectly match or surpass the human brain's image matching capabilities, for example. Other things like consciousness, then i imagine some weird effect causes it, and i wouldn't be surprised if quantum effects contributed to that, weird as they are themselves.