October 30, 2006

Is IBM's Blue Brain project a precursor to an AI project?

If you want to understand how something works you should model it the best way you can.

This is precisely what IBM and the Brain and Mind Institute (BMI) are trying to do with the brain. Called Blue Brain, it is a project that was initiated in May 2005 with the lofty goal of modeling the mammalian brain. IBM and BMI claim that the aim is not the creation of AI, but a way to study how neurons in the brain interact with one another. Their intention is not to re-create the actual physical structure of the brain, but to simulate it using arrays of supercomputers.

I'm somewhat dubious of their claim that this is not an effort to develop AI. The construction of a simulated brain is awfully close to the development of AI. While this project may not be an immediate attempt to develop AI, I think it is at the very least a precursor to such a project. As the developers themselves admit, they are trying to model how information is formed in memories and how it is retrieved. Yes, it’s a different approach than trying to code for AI, but if it ends up looking like a duck and quacking like a duck…

As evidence that IBM and BMI are thinking about AI, the initial goal of the project is to simulate a neocortical column, which is the smallest functional unit of the neocortex. This is the part of the brain that is responsible for higher functions such as conscious thought. In humans, the neocortical column is 2mm tall, has a diameter of 0.5mm and contains 60,000 neurons. Project developers are initially working to replicate the neocortical column of a rat, which has only 10,000 neurons. It will take 2 years alone to construct a simulated column. Developers hope to model the human brain in about 10 years.

To model these components the developers will use a Blue Gene supercomputer that will run the MPI-based 'Neocortical Simulator' combined with 'NEURON' software. Blue Gene is a computer architecture project that will spawn several next-generation supercomputers -- computers that will reach operating speeds in the petaflops range, and are currently reaching speeds over 280 sustained teraflops! Its 8,000 processors will crunch away at 23 trillion operations per second.

The developers are hoping to use Blue Gene to create biologically realistic models of neurons. In fact, the results of the simulation will be experimentally tested against biological columns.

Once the neocortical column is developed, project developers will attempt to replicate the simulation at the molecular level (does this imply nanotech?) and work to simplify the column such that they will be able to run parallel simulations of large numbers of connected columns. You could say that running parallel simulations is important; the human brain consists of about 1 million cortical columns.

IBM has always thought big. Their dedication to the Deep Blue experiment and their existentially shocking success at defeating a grandmaster at chess shows that they have serious intentions. While Honda toils away on Asimo, IBM is trying to figure out a way to endow such a robot with brains.

Taking into consideration their dedication to these sorts of projects, and considering the resources and expertise at their disposal, I strongly suspect that IBM will be the first group to develop strong AI and possibly greater-than-human artificial intelligence. Blue Brain is certainly a step in this direction.


Forbes: IBM Aims to Simulate a Brain
Blue Brain Project Homepage

No comments: