Neuromorphic Electronics

I previously mentioned about the synapse program which is DARPA's attempt at creating a neuromorphic brain. A neuromorphic brain is basically a computer chip that is designed to replicate the functioning of a real brain. According to the wired defense blog they want to make it at a cats level of intelligence. The government wants to replicate a cat's cortex in a chip.

The follow-on phases of the project will create a technology that functions like the brain of a cat, which comprises 10^8 neurons and 10^12 synapses,"

Why a cat and not a rat? I have absolutely no clue, but I guess they have to start somewhere. I thought it might be easier to attempt to replicate an insect level of intelligence, but our government is somewhat more ambitious than that. There is no mention of creating a human level intelligence yet, but that would be the next logical progression. It probably won't end up getting that far, though.

You can read more about this new project at this website. (Never mind, The press release has actually been retracted, apparently the deal has not been finalized yet, so the company HRL may not be doing this). The press release is mysteriously dissapearing from every website, so I won't link to it. One could say, though, that the cat is out of the bag. Basically, there really isn't much new info that hasn't already been discussed. What will this actually accomplish? The chip could end up being a million dollar paper weight. Maybe I'm a little pessimistic, but it would be interesting if they got it to act in a similar manner to a real brain.

I'll admit, though, I really don't understand too much about this type of technology. Ray Kurzweil had an interesting comment in his book The Singularity is Near about creating an artificial brain that seemed appropriate (pg 177-178).

Modeling human-brain functionality on a nonlinearity-by-nonlinearity and synapse-by-synapse basis is generally not necessary.

He goes on to say;

Preserving the exact shape of every dendrite and the precise squiggle of every interneuronal connection is generally not necessary. We can understand the principles of operation of extensive regions of the brain by examining their dynamics at the appropriate level of detail.

Basically he is saying that we have (or will soon have) detailed enough maps of the brain to start to replicate it's functioning either in a computer or on a chip. We don't need to perfectly model every receptor protein or atom in a neuron (or even synapse?) to get a decent representation of the brain.

Kurzweil really is quite a creative genius and a very interesting futurist, but he does get a heck of a lot of stuff wrong in his predictions. So I'm not so sure that he is necessarily correct. He does give a pretty good argument, though, that this is possible. However there is a lot we don't know about the brain. For one thing, there is the 3-Dimensional arrangement of neurons in specific brain areas and how those regions interact with each other. The interaction of these regions according to the laws of physics is not fully understood even if you know the firing patterns of the underlying neurons and region activation. It is unclear if you can condense that function down to a presumably smaller (or perhaps larger) chip while retaining the essence of those complex interactions, without also replicating the underlying physics, chemistry or any other "level" inherent to this world. Maybe it is possible to drastically simplify things and still get a working model. However, I still have my doubts that this will be possible.