By Matthew Herper, Forbes Magazine.
IBM has embarked on a quest for the holy grail of neuroscience–the far-off goal of creating a computer simulation of the human brain.
When the first mammals evolved from reptiles 200 million years ago,
one of the biggest changes was inside their heads. Their brain cells were
structured together into columns, an innovation that could be repeated like
a computer chip to make larger and more powerful minds– from mice to cats
and dogs to humans.
“This was the jump from reptiles to mammals,” says Henry Markram,
founder of the Brain/Mind Institute at the Ecole Polytechnique Fédérale in
Lausanne, Switzerland. “It was like discovering a G5 processor or Pentium 4
and just copying it.”
Now, Markram is announcing a collaboration with IBM (nyse: IBM – news
– people ) to create a computer simulation of these fundamental neurological
units, called neocortical columns. The process will involve building a Blue
Gene supercomputer with 8,000 processors that can roar along at 23 trillion
operations per second. Each processor will be used to simulate one or two
neurons. If finished immediately, the machine would be one of the five
fastest supercomputers in the world.
A neurocortical column is a structure half a millimeter in diameter
and 2 millimeters long that contains about 60,000 neurons. (The human brain
is made of 10 billion neurons.) The columns were discovered by Nobel
Prize-winner Torsten Wiesel of Rockefeller University. They remain similar
in different mammals, but the human brain is crammed with more of them. It
was the need to fit in more columns that forced the human brain into its
crinkly, wrinkled shape.
In Switzerland, Markram has put together a large lab dedicated to
studying neurocortical columns in animals. His first effort with IBM will be
to simulate a single rat neurocortical column. That alone is likely to take
several years, as the computer model is rigorously checked in experiments
against neurocortical columns taken from rats.
Once it is clear that one column has been simulated, the project will
move on to simulating several such columns, again verifying its results by
experiments with real brain tissues. Then, it will be possible to create
larger simulations. After a decade or more, it may even be possible to
create a model of the human brain. Markram and IBM both emphasize that the project would not create artificial intelligence but a way to study how
neurons in the brain interact with one another.
“We believe that we will be able to capture the heart of the
information process,” says Markram. “Not just the column but how the
information was formed in memories and retrieved.”
The project could lead to new understandings of various diseases such
as schizophrenia, autism and attention deficit hyperactivity disorder.
Degenerative diseases such as Alzheimer’s may turn out to be more difficult
to model because they involve the failure of more than just brain cells,
Markram says.
For IBM, the project represents one of several initiatives in its Blue
Gene program, which involves building supercomputers based on a powerful new computer architecture. The most powerful of these units, at Lawrence
Livermore National Laboratory, is 16 times more powerful than the one
Markram is using and will be used to simulate the intricate ways that proteins fold–one of biology’s big mysteries. Other efforts exist in astrophysics, atmospheric modeling and financial modeling.
[Thanks to Albert Enayati.]