Henry Markman, a researcher at the Swiss Federal Institute of Technology in Lausanne (Switzerland) is simulating the architecture, morphology and function of the human neocortex by an IBM supercomputer of the Blue Gene family, which is capable of performing simultaneously thousands of thousands of transactions per second. To do so, he created in 2008 a “digital facsimile” of a cylindrical piece of tissue in the rat cortex, using 10,000 neocortical columns of over 10,000 neocortex neurons in three dimensions, of 200 different genetic types, with data from more than 15,000 neurons in culture. In 2011, the team announced it had simulated a “virtual slice” of brain tissue with one million neurons. Markman hopes to emulate brain function, understanding the neocortex as a 'new brain' created by our species, needed for education, interaction with others and higher intellectual functions, such as emotion or thought.
Evolution of the Blue Brain. Copyright http://bluebrain.epfl.ch
To fully replicate a brain would take a computer a million times more powerful than the Blue Gene, but Markman believes that it will be possible to clone our mental functioning someday, and thus the essence of a humang being. For the moment his project, called "The Blue Brain Project" will provide a “unifying principle” for scientists to rally around, gathering data from laboratories around the world in one place. An entire division of the project is devoted to creating a new breed of intelligent robots with “neuromorphic” microchips designed like neurons in the human brain. “The biggest success for me,” Dr. Markram said, “would be if after 10 years we have a new model for neuroscience, where everyone works together. It’s about a new foundation.
Source:
I had an argument with a friend about this project a few years back. There are some significant assumptions being made. The most important is that the problem we face is simply a matter of scale. By implementing our current understanding of neurophysiology with current hardware we feel we may have done something interesting, and given the pace of progress in hardware performance, we will be able to achieve something significant in "decades". This assumes our current reductionist approach to neurobiology scales up to brain size without any significant need for new knowledge en route. I think this is very unlikely.
ReplyDeleteTake simply the question of how to connect the neurons. This scales up exponentially as the system size increases. How will the researchers know what connections to make between the neurons and what strengths to assign to them? The key quote from the NYT piece: “'If I build in enough biological detail,' he reasoned, 'it would behave like a real brain.'” This "if" is what we used to call a nontrivial problem, and the devil is in the detail. So we know how a neuron works, and this fellow learned how connection strength changed in response to a stimulus. That is very nice. But throwing a bunch of neurons together that follow a set of rules and individually behave properly does not produce a brain. The Newtonian equivalent would be a good understanding of the laws of motion, but having no idea what the initial state of the system is. I have yet to hear anyone propose how we might acquire a good understanding of an initial system state for a complex neural network.
With the failure of high level artificial intelligence projects in the 80’s and 90’s, it is understandable that the interest migrated to an approach of mimicking the system design even in the absence of an understanding of the how the system works. There is a term for this practice: http://en.wikipedia.org/wiki/Cargo_cult
Nice comment Armand and I agree with you.
ReplyDeleteThe brain has a huge amount of neurons, each with up to 10,000 connections with the others plus all the synapses in between, firing at different frequencies in each circumstance. So I think that when you try to find a mathematical model to represent this behaviour (like using neural networks) you come up with the conclusion that this is not possible. The brain does not fire the same pattern each time even if seeing the same object (same input), simply because there is always difference in some of the circumstances involved, regarding the environment where the person operates or the person itself. That means you cannot model this with algorithms and microprocessors, the brain simply does not work like a computer.