What are the major technical barriers before we can identify the input and output channels to this insect's brain and start iterating through all possible input values, recording the corresponding output values? And once we can do that, could we use that data to fly a virtual insect around a virtual world?
This article suggests that it's computationally infeasible even for an organism with only 302 neurons, all of which have already been completely mapped.
Honest newbie question: Why would we even want to do that? Exclusively for academic medical purposes? (ie learning how our body works)
It seems to me, that since evolution is highly imperfect, that trying to mimic living beings, might not be the best idea. While we do mimic them a lot, mostly it seems to be for learning until we can do better. But in this case. Since it's so expensive and not viable to mimic them properly. Shouldn't we just try to come up with our own algorithms instead of trying to copy nature's historically bloated algorithms?
That seems to be a lesson we learned from machine learning. I remember when neural networks were first being talked about seriously a decades or so ago. The goal seemed to be to try to mimic neurons just for mimicing neurons sake. Many critics would say we should probably try other learning algorithms that were obviously more efficient. And nowadays we barely see actual neural networks being used seriously in commercial production simply because we have better algorithms that are not trying to emulate nature, just because.
Wouldn't trying to emulate a virtual insect be repeating the same mistake? If all we want is to design a virtual robot that can look for food and control its wings. I sure as hell don't need thousand something emulated neurons to build that.
>Shouldn't we just try to come up with our own algorithms instead of trying to copy nature's historically bloated algorithms?
People are working on this problem from both sides. AI on one and biological simulation on the other. We don't know which will realize its goals first.
You don't do things in science because they are practical (you just write that on the grant applications), in fact you do many things largely because they are impractical.
It's about wonder and lust for knowledge and all that. It just so happens that the theoretical ground work for all technology piggy backed on this motivation.
I wonder why the Blue Brain guys don't work on c. elegans? They claim they can simulate full neocortical columns, a c. elegans should be no problem and they could compare the behavior.
Arguably, even human brains work that way. You can't really draw a line across part of the CNS and say "This is where the brain stops." The retina, for instance, is almost as much a part of your brain as it is your eye.
By the same token, it's not always easy to classify complex movements as reflexes or thought-guided actions. Cut a chicken's head off while it's walking, and it will keep going for a while. My understanding (IANABiologist) is that the same is true for a human.
This was one of my first thoughts as well, sort of. Alternatively, is it possible to model each neuron when there are only around 8k? Maybe estimate the visual, audio, and tactile bandwidth of the wasp and feed the simulated neurons a stream of simulated environmental data and just let the connections evolve.
There are not many neurons, but there are many (too many) possible connections. To find the right ones, nature took several billion years. Even with the fastest computers, it would take a considerable amount of time (note that the lifetime of these wasps is a few weeks at most, so evolution happens very fast).
At small enough scales -- and this is one of them -- air is not just a fluid, it is a viscous fluid. Movement will be more like swimming than flying; there won't be any landing or liftoff, so much as grabbing on and pushing off.
It's a virtual environment; presumably, one could set the "wall clock" rate of the sim to account for inter-room, or even inter-continental comms latencies.
