In the past, the research has focused on re-creating non-trivial neural networks. However, the network is nothing without the organism - sure, you can examine the internal oscillations - but that hasn't led us anywhere. (See "Perfect C.Elegans" research paper, 1998). One can't debug without seeing the results, and that's the case with a neural network without the rest of the organism.
Why C.Elegans? It's the simplest organism with a nervous system (302 neurons), and the connectivity has been completely mapped. It's a hermaphrodite, with a fully-sequenced DNA. It's also one of the more studied organisms out there.
Considering that nobody outside the jains seems to think there is an ethical issue with experimenting on invertebrates like insects, and those are significantly more complex than c. elegans...
and that not even the jains have said anything about computer programs symbolically representing a real animal...
I might be asking a dumb question here but can it be possible to eventually simulate the worm's reproduction cycle so that we can fast forward the simulation to millions or even billions of generations to see more complex organisms evolving?
It's not a model of the worm's genome, let alone its full development required to actually produce the worms, let alone of populations of the worms, let alone populations of the worms interacting with a full blown natural environment. Which is what determines what happens in natural selection.
Evolution does not naturally tend toward complexity, it tends toward fit to the environment. In one case that may be specializing in eating eucalyptus leaves. In another case it might be having a big brain. We tend to think that the whole direction of evolution is toward producing us but there is no basis for that in evolutionary theory. We ended up with particular traits because they were available in our population and gave our ancestors a leg up in their specific environments.
Looks like the beginning of Permutation City[1] - won't be long now (relatively) until we won't be able to observe a difference between real c.elegans and simulated c.elegans, simulate its environment, simulate other organisms, etc. Will these worms be _alive_? What happens when computers become advanced enough to simulate humans? What if they already are? o_O
Well, hold on a second. Even if we know the worm's neural network and how it's connected, that doesn't necessarily mean that a computer simulation of it will do exactly what the worm does. The problem is that we're not doing an atom-by-atom simulation of the worm but rather we're hoping we've extracted the important logical features of how the worm works, so that the thing will run on modern computers.
This is actually the more exciting part of the research. Serious ethical questions aside[1], imagine if you could simulate a human's neural network. Probably your first simulation would seem unbelievably stupid. Perhaps you find out that you didn't account for the ways certain neurotransmitter concentrations will "leak" data from one synapse to a nearby one, without which the system becomes radically disconnected because evolution was lazy and connected them without a specific wire. (I don't know; I'm being hypothetical.) So now you get to introduce some sort of adjacency matrix which manages which neurons are "next to" each other and receive these "secondary signals." Then it seems to be able to learn language, but it still can't balance in the world, and working it out, you find out that there is a big failure in the motor regions because they only work when the right signal propagation delays are introduced, and you were propagating them all instantaneously, and so on.
In the distant past, we had hoped that chess was so complicated that it would only be solved with some great insight into human understanding -- but instead it was solved with brute force. This is one of the first cases where I see that the brute force might be finally able to give us a test model by which we might better understand understanding itself.
The only worrying bit is the neural nets themselves. Neural nets are notoriously difficult to interpret and understand. Even the calculus-based approach of "I'm going to make a tiny tweak to the network and see how it changes the output" offers only a little enlightenment.
[1] I do think ethical questions about killing artificial consciousnesses deserve discussion time; I just don't have much to spare at this moment and it's kinda tangential.
In the past, the research has focused on re-creating non-trivial neural networks. However, the network is nothing without the organism - sure, you can examine the internal oscillations - but that hasn't led us anywhere. (See "Perfect C.Elegans" research paper, 1998). One can't debug without seeing the results, and that's the case with a neural network without the rest of the organism.
Why C.Elegans? It's the simplest organism with a nervous system (302 neurons), and the connectivity has been completely mapped. It's a hermaphrodite, with a fully-sequenced DNA. It's also one of the more studied organisms out there.