May 23, 2010
First a disclaimer: The actual research behind this article is way out of my field. Way out. What I understood of the paper I found interesting, but the post is largely about the way it has been portrayed in the press release.
The paper, in Nature Physics, describes the creation of an “organic molecular layer” computer. The molecular layer is capable of parallel computing, that is, performing multiple calculations simultaneously, rather than one after the other as in standard transistor based computers. In the paper, they demonstrate the computer’s capabilities by simulating the growth of cancerous cells and heat diffusion.
The authors make a comparison between the parallel computation of the brain, and the molecular layer. The computer is also capable of “self-healing” to work around any damage in the network. However, in the press release they also have this image showing “a processor that looks like a brain”, generated patters in the layer that look like fMRI scans of the human brain in different mental states.
From what I can tell, the layer has received input specifically to produce the patterns. Whilst this is quite a nice demonstration of the layer, the implication seems to be (at least to me) that the activity in the layer that produces the pattern is somehow imitating the activity in the brain that is producing the pattern of fMRI activity. Even if the layer is performing some sort of similar process (although I’ve got no idea what it could be, and it must surely be vastly simplified) that the activity looks like fMRI activity is surely pretty irrelevant, the location of activity in the brain is just an accident of anatomy. There are ways in which this computer acts like a brain, but this isn’t one of them.
It does, however, seem to be the way the research is being sold, for example, at h+, where the patterns are described as having an “uncanny resemblance ” to fMRI data.
This is an amazing invention, with some properties which mean it could be described as a ‘neural’ computer, but I feel like it is a bit misleading, either on the part of the press officer, or the columnist, to sell the story on the basis of the MRI figure.
If the science is interesting (and this is) it should sell itself!
EDIT: This, from IEEE Spectrum seems to be a more balanced account of what it is/does.
Bandyopadhyay et al. Massively parallel computing on an organic molecular layer (2010) Nature Physics 6, 369 – 375