Is there any software that exists which will simulate a human brain (which is waiting around for the hardware fast enough to make it useful)? If so, what kind of performance would be necessary to make it useful?
Thanks,
Rob
That depends on the context, and the problem you’re trying to solve. For instance, if the problem is “A human brain is thrown upwards at 10 meters per second, from a height of 5 meters. How long until it hits the ground?”, then I can write a program that models the brain quite well.
A whole human brain? No. Not even close. There is a huge amount about the functioning of the brain that is still not understood. Small networks of a few neurons can be computer simulated, but even that involves lots of simplifying assumptions.
There is no software today that even remotely comes close to simulating a human brain.
The closest any one has gotten is to accurately simulate some of the aspects of a 300 neuron worm brain.
This bears repeating. It’s tempting to think that we could, in theory, just scale the simulations up. But in fact the neural network simulations don’t work the same way as real neurons, and there’s still plenty we don’t know about how the brain works even at that low level.
They have recently been able to do far more than that, all the way to complete simulation (not just some aspects) an entire cat brain, and 4.5% of a human brain:
We can simulate neurons in software. If you throw a few of them together they do some interesting things and they are reasonably easy to understand, but they don’t act at all like brains. Throw a few more neurons into the mix and the interconnections very quickly get so complex that no one can understand how they work. But, that’s about the same level where all of these interesting brain-like patterns start to emerge from them.
You can’t just randomly throw in more simulated neurons and expect them all to act like a human brain, though. Exactly how the neurons are all interconnected makes a huge difference, and once you get above a trivially small number of neurons the interconnections are just too complex for anyone to understand at this point. Right now, folks are trying to simulate extremely simple brains, like those in worms and insects, and they aren’t quite there even for those. Getting artificial neurons to start simulating more complex brain functions is several orders of magnitude beyond our abilities right now, and something as complex as the human brain is so far off in the distance that folks can’t even see a clear path from here to there.
You may find this article interesting, and there are links in it to some of the software currently being used. You can read up on the various software packages to get a better idea of where things are at right now.
Wow. They’ve made a lot of progress since I last took a good look at things. I didn’t know they were anywhere near that far along.
They did one that simulated a woman’s brain, but then they did some unspecified thing that upset it and now it’s not speaking to us.
No they didn’t and Scientific American shouldn’t have been one to perpetuate this.
http://www.acceleratingfuture.com/michael/blog/2011/11/more-nonsense-reporting-overblowing-ibms-accomplishments/
If you read the details (I don’t have a link for that but I have read it) of what they simulated you will see that they did not even get close to simulating a single neuron or confirming the output for a given input matched any actual cat as that was not actually their intention.
As previously stated and linked in this post, the flatworm is the only brain they have even got close to simulating. Specifically they were able to get the same output with respect to it’s response to a chemical gradient as a live worm.
No, not really. This is standard science journalism hype. They have a simulation that simulates about as many neurons as there are in a cat brain. That is an impressive achievement technologically, but it is not a simulation of a cat brain. It does not simulate the connectivity or programming of the cat brain (when it comes to brains, those are not really separable issues), and it does not reproduce the behavior of a cat brain (or even try to so far as I can make out). To repeat, we are nowhere near knowing enough about the structure of actual brains, even cat or rat ones, to simulate them meaningfully. Also, although the authors of this study claim their simulated neurons behave more like real ones than previous attempts do, I will bet you dollars to donuts that they are still highly simplified.
Also, there is more to brains and their behavior than neurons. There is electrochemical connectivity through glial cells, there is diffusion of chemicals that affect neuron behavior through the CNS, there are the effects and the control of regional blood flow, and probably other things that I do not know about (and maybe some that no-one knows about yet).
¬¬¬¬¬¬¬¬¬¬¬¬¬¬¬¬¬¬
In future news: Scientists have succeeded in fully simulating the structure and behavior of the feline brain. Unfortunately, they have found that the simulated cat just will not do what they want it to. 
Some other examples of complicating factors:
1 - Glial Cells
As njtt said you can’t ignore these, every month or two I read articles about how much more these cells are involved in computation than previously thought.
There are 10x more of these than there are neurons.
2 - Structure Matters
As engineer_comp_geek mentioned, just throwing a bunch of neurons into a simulation does not get you closer to intelligence. For example, grid cells are physically laid out in a grid-like manner and are activated according to an animals movement within a 3D environment and more recently it’s been discovered that they are used while visually processing a scene. If you don’t capture the details of this section of the brain then either your simulation won’t work or you will need to program that functionality in some other manner.
3 - Other Physical Details
Differences in dendrites (strong vs weak) cause differences in signal amplification and timing of transmission and they believe these differences play a role in learning (as opposed to learning being due to changes at the synapse).
These types of details are important and the scientists are discovering more and more of these all the time.