Why Neuroscientists Need to Study the Crow
nautil.us
nautil.us
It's like saying CPU to overall device size determines processing power, so a 12-core Xeon in a giant box is less powerful than an ARM32 chip in a phone.
My guess would be that this ratio is a proxy for how important brains are to that animal in its evolutionary niche. A small brain vs. body size indicates that brains must not matter as much as other things, therefore no need to waste resources on it. But a large brain vs. body size indicates that brains are important. This might in turn correlate with greater evolutionary pressure to optimize brain function and efficiency, leading to some super-interesting adaptations and a very minimal parsimonious design.
... which supports the idea that studying these small and efficient brains might be deeply revealing. They're so small and powerful that everything in there must be really important.
It would be interesting - not very ethical and possibly deeply creepy, but interesting - to select birds and other animals for intelligence, and see how just how far you could take a selective breeding program.
You see analogs in e.g. "sclerotic" corporations with lots of employees doing nothing. My bio professor said "life doesn't work perfectly... it just works." Evolution isn't "survival of the fittest," but selection for a "surviving subset of the sufficiently fit." Economics is the same, thus the sclerotic corporation/government analogy.
Of course if evolution actually was "survival of the fittest" it would reduce to a greedy hill climbing algorithm and would converge on the first local maximum it encountered and stay stuck there forever. Tolerance for variation is a (probably provable) prerequisite for anything particularly interesting, and diversity implies inefficiency among other things.
When I was playing with alife and genetic algorithms a lot, I found that relaxing selective pressure often improved performance in terms of overall best solution generated. There's a number of papers out there that draw similar conclusions in a variety of systems. Sometimes you can get GA/GP systems that escape local maxima and find much more clever and interesting solutions by setting up a selection function that adjusts its strictness curve based on measured diversity (we jokingly called this affirmative action), or by creating an environmental topology that encourages diversity (many compartments / demes, etc.).
We do it already for dogs.
I'd like to see a species selected for social intelligence—the kind of tribal-affiliation status-dynamic thing that is theorized to have brought about language.
If an animal gets hurt, it's nervous system will pick it up and do the management necessary to direct resources to repair where it's hurt. Each part of the body requires some sort of monitoring system. The bigger the body, the more computational power must be devoted to body maintenance & monitoring.
A small brain vs body size means there's little "computational surplus", it's mostly used up for making sure the body processes are running, making repairs as required, moving hormones around, etc. A large brain vs body size means spare computational capacity is available for observable intelligent behaviour.
Anyway, that's my theory.
A small brain for a large amount of inputs means the brain's representation is much more aliased than perception, and the internal algorithms work despite very fuzzy input. Cause for that ratio probably being that the brain is very high pass and only cares about high input spikes (i.e. very dull sense of tact, but sensing a sting still matters, so it still needs a high skin sensor density).
While a large brain for a relatively smaller sensory input means the internal representation is more detailed, or it's even creating synthetic detail from internal models of perceptions at the same time as it processes inputs, which would be where problem solving and awareness capacities kick in.
1) This tool is no good, I need one with greater length. 2) I identify one over yonder. 3) It is obstructed. 4) I will need weight to remove that obstruction 5) I have identified suitable weights 6) I will need a tool to get those weights 7) My current tool is sufficient for that 8) I can chain these observations to obtain my goal.
That kind of executive functioning and judgment composed from simple tasks learned independently, the first time, and without supervision seems to me well beyond the norm for what we observe in perceptron models of learning and feedback.
Birds have an analogous structure to the mammalian neocortex [1].
edit: punctuation, removed crankiness
the packing here is more about neurons connections (100T) than just about pure neurons number (100B) - high connectivity requires kind of spherical geometry with folds/ridges.
But like individual transistors, the overall pattern is what determines its functionality. My former field was in consciousness, and one of the interesting puzzles was the cerebellum. The cerebellum contains 3x the number of neurons in the neocortex, but in cases where it's been removed (resection for epilepsy or cancer), people have fine-motor control issues, but not much more. By contrast, removing a much smaller area like visual area V1 will wipe out your visual experience and leave you blind!
So, what explains the opposite interaction between size and effect? Structure. The cerebellum is highly regular, like a GPU designed for repetitive graphics operations. The rest of the brain is more like, uh, an FPGA.
I suppose a possible investigation of that question would be, say, comparing our cerebellum to that of another animal with extremely good fine motor control, like the cephalopods.
I think the cerebellum is no more or less efficient than other areas, and I don't know if "computationally intensive" is the right way to look at it. The cerebellum really is similar to GPUs. GPUs use a lot of silicon to do constrained, repetitive tasks efficiently, but they don't support general processing models outside their narrow domain.
It's hard to compare with cephalopods, though they're very interesting, because their neural architecture is so radically different from ours. For starters, ganglia in their arms allow each arm a certain measure of independent activity. E.g., arms that get severed can try to "feed" the mouth for up to an hour after detachment.