article and a New Scientist
article offer news of IBM's Blue Brain Project. For the
project, IBM will build another Blue
Gene style, Linux-powered super computer that will be used to
simulate the human brain. The computer will use 8,000 CPUs and should
reach about 23 trillion operations per second. The goal is not to create
AI but to create a simulation of human neuron interactions. The project
will begin with a multi-year attempt to simulate a rat's
column, which has only 60,000 neurons compared to the human
brain's 10 billion. The simulation code will be based on Dr. Philip
Goodman's NCS (NeoCortical
Simulator). NCS is open source software and can be run on any Linux
box, in case you don't have a supercomputer handy.
That sounds like an interesting project. Only one thing, it says that
the biological neurons in their simulation can do more than normal
artificial ones, but arnt they modelling with artificial neurons?
Okay, I couldn't find where they made that claim. Don't sue me, I was
only skimming. However... "normal" artificial neurons are often simple
things summing up their inputs and providing an output based on a simple
mathematical function. When folks talk about Artificial Neural
Networks, that's basically what they mean. There are different ways of
training the networks, different ways of connecting them, and different
ways they can be triggered. ANNs work, and so are in use all over the
place. That's no guarantee that they are an accurate model of a real
neuron, anymore than a robot arm has to be an accurate model of a human arm.
Real neurons in a real brain work in a more complicated way. There are
lots of different neurotransmitting chemicals, lots of connections,
goodness knows about updating and training.
While obviously any neuron in this computer will be artificial, it seems
that the important thing for the researchers is that it is an accurate
model of a biological neuron. For most AI researchers this is not