Not an HTM question - but I recently discovered C. elegans nematode (a transparent roundworm with about 302 neurons) which seems to be widely used by neuroscientists for their research.
However, I was wondering whether we can apply an approach like NeuraLink does - an interface that records the activated neurons of the brain at the current timestep and synthesize a dataset of the same. States of other corresponding parts may also be recorded to further complement the dataset. Since the number of neurons is manageable, I don’t see why the surgery might be complicated.
Then, we could apply any DL technique and try to predict the state of some neurons; if we manage to pre-train on such a specific dataset and accurately predict a reasonable part of the nematoda’s brain given some input then we can replicate that worm as a 3D simulation easily (or more realistically an agent in a simple environment) and investigate whether it can demonstrate simple behavior.
Is an experiment like this scientifically/biologically possible?
Yes, there are many ways to measure the state of neurons inside of living things.
One way to do this is to infect a mouse with a special genetically engineered virus, which is derived from rabies and instead of kill brain cells the virus injects them with a protein. The protein is a “green phosphorescent protein” which was originally taken from jelly-fish DNA and it’s been modified so that it only lights up in the presence of Calcium Ions. Calcium Ions are an important chemical messenger in the brain, and have been implicated in a lot of common cellular functions.
And then you install a high speed camera and a black-light inside of the animal’s skull.
In this video, the lower right window shows the calcium levels. When a dot suddenly lights up it’s likely because that cell made an Action Potential.
Beyond simply monitoring, is there any plausible way to replicate the behaviour of the nervous system of such organisms? If so, does this depend on faithfully replicating every nuance of neuronal activity, or is it possible to identify (and presumably replicate) algorithms independent of the neurons that execute them?
