Slashdot Mirror
Effort to Create Virtual Brain Begins
bryan8m writes "An IBM supercomputer running on 22.8 teraflops of processing power will be involved in an effort to create the first computer simulation of the entire human brain. From the article: 'The hope is that the virtual brain will help shed light on some aspects of human cognition, such as perception, memory and perhaps even consciousness.' It should also help us understand brain malfunctions and 'observe the electrical code our brains use to represent the world.'"
What's interesting about this type of study is the possible philosophical arguments that come up...
Our brains are made of mostly water, carbon, etc.... which form neurons. This is only important in the sense that we are what we are because these neurons are able to take a set structure, where neurons interconnect, and then have a specific function, where they fire.
There's nothing magical about these neurons. Let's say that you could replace these neurons with say, ultra-small marbles, that could take the same structure and perform the same function... It is logical to think that this marble-brain would be an actual brain, the same as any other. It would be a person.
So if they're simulating a brain virtually, but this virtual construct simulates the structure and function correctly, would this virtual brain be aware? Would it be a "person"? I personally, would say that it would. But then, is it moral to ever shut such a simulation off (murder)? Or create it in a virtual world without any other virtual brains to talk to (torture)? Or create it at all for the use of an experiment?
> So what happens if this thing develops a consciousness?
...and it says it's scared. or alone. or just wants a friend.
Yes. That's what has me thinking. Not that I think we should stop, but it's going to be a disturbing moment when the techs running these things get to a point where they ask a simulation brain questions, get it to perform tasks, get it to react like a human does...
While it is true that Moore's Law suggests we will soon have the processing power of the human brain, that doesn't mean we will soon have AI on our hands. If we built this computer and fed into it a "Hello World" program written in Pascal, it isn't going to suddenly become self-aware.
We only have one type of working brain, so it would make sense to replicate this in every way possible in order to create a simulated intelligence. However, this has a great deal of complexity that we neither have the bioloical knowledge to understand nor the technical knowledge to emulate. Literally millions of neurons are connected inside us, forming cortical maps and working at different levels of awareness, from the lower, barely perceptible levels (reflex actions), to the higher, seemingly conscious, levels (deciding whether to order toast or a bagel for brunch).
Anyone who's interested in AI (or indeed the operation of the human brain) should read Steve Grand's book. It is highly enlightening, and very thought-provoking.
One line blog. I hear that they're called Twitters now.
We don't have to simulate the brain function in real time for it to be a valid simulation. And, if you read the article, they are modeling this on many, many slices of mouse brain. As it says, they hope by doing this that they will shed light on perception, memory, and perhaps even consiousness. It doesn't say they will achieve this, just that they hope to.
No sensationalism here. Move along.
As someone who is receiving my PhD in "Computation and Neural Systems" from Caltech this week, and having worked briefly in that lab, I can tell you that the simulation you read about, which is called GENESIS probably simulated the neuron in much greater detail than is ultimately required to create a brain. It simulated the entire physiology and chemistry of the neuron ... every ion flow, trans-membrane voltage, etc. One of the many goals is to explore precisely what information-processing behavior arises from the chemistry and biology.
But, once you determine that information-processing behavior, one should in theory be able to simulate that without a detailed model of the underlying structure. I mean, if I know that impulses from X input synapses cause the voltage at the soma to raise/lower according to a certain time function, and that a certain voltage at the soma causes an action potential to be fired, which will trigger the neuron's own output synapses to fire Y milliseconds later, I should be able to simulate these properties without going to the pain of modelling the ion channels, capacitance, and resistance of every patch of membrane on the whole neuron's surface.
That should buy a few years' worth of Moore's law for your prediction. Consider yours an upper bound, and assume we can make shortcuts to bring it sooner than 2040.
I actually think the top supercomputers are within spitting distance of modelling a human brain - or at least smaller mammalian brains now. The trouble is that despite what TFA leads you to believe, far too little is known yet about the interconnections of those neurons. Even less is known about their learning functions. The state of the art in much of the brain is to stick a few electrodes in, hope you find a couple of neurons that are connected in some way, record for a while and then do statistics on their firing patterns to estimate the strength an type of their pairwise connection. Then by using that they hope to work backwards to deducing the connection patterns of whole clusters of neurons. It's slow, messy work.
The group in TFA uses thin slices of brain where they can more accurately observe which neurons are connected to which, and which neurons they are recording from. It's a useful technique, but since the connections in the brain are three-dimensional, taking thin slices fundamentally alters the structure. It can't tell us anything.
Much of the brain is still a black box, effectively. It will still be a while before we can model an entire brain, regardless of CPU power available. My personal gut feeling is that the understanding of the neuronal network is far more the limiting factor at this point.
I stole this sig from someone cleverer than me.