________________________________________________________________________________
I'm always interested in understanding animal neurology, especially in the context of trying to reach more general levels of AI. I find it interesting that modern ANNs perform near human level at some difficult tasks, and yet it seems there are some tasks that insects perform that we'd have no idea how to implement. Animal nervous systems certainly have interesting things to teach us.
It also seems like understanding simpler brains will help in progressively understanding the human brain, even if these brains are very different. Developing a less human-focused toolkit might be what we need; sometimes studying a more general problem is what you need to get past blockers in a more concrete problem.
On that topic I really enjoyed _Other minds_ by Peter Godfrey-Smith, which talks about octopus behavior and some neurology, but I'm interested in recommendations for more technical readings in animal cognition / neurology.
> _near human level_
..._of effectiveness_ ;)
what are the tasks "tasks that insects perform that we'd have no idea how to implement" ?
Bees can land and launch from flowers and house flies can land upside down on a vertical surface.
Sure, tiny flying robots exist, but I don't know of anyone who can make them do either of those things using only onboard sensors and computing power.
Wow, this was such a fantastically well written article! Really a pleasant surprise compared with much of the popsci writing I see. Very clear and compelling communication!
Cool progress on a new model organism and toolbox for behavioral analysis. The main advantage appears to be the spatial separation of neurons. (I have not yet read the paper in Cell.)
As to the “mind” of a jellyfish, mouse, or human: much of the core computational activity is molecular, synaptic, and electrotonic—and is not associated with large membrane voltage swings. These lower tiers of processing are orders of magnitude faster and smaller than the network activity patterns linked to behavior.
It is great to understand crude input-output relations—now at neuronal network level dynamics. But how is this much different than what we know about spinal cord circuitry for reflex control in mammals or cortical-cerebellar control of movement initiation and control. Is it fundamentally a better system or is it simply more colorful, “elegant”, and semi-real time? IMHO; the latter.
Not sure these are quite the right metaphors, but imagine trying to understand microprocessor function from either just input and output relations or from timelapse data on CPU transistor currents—but only every few milliseconds.
To understand neuronal computation we must, at a minimum, record synaptic and membrane currents across cells and circuits at the 0.1 millisecond levels over minutes and hours. And preferably across many environmental perturbations (see Ev Marder’s amazing work on the shocking complexity of somatogastric ganglia of crustaceans.)
And to understand the true sources of behavioral variation in a causally rigorous way we will also need gentle genetic perturbation generated by natural DNA variants—not knockouts on one genetic background—that is a misdirected focus of so much reductionist N=1 neuroscience.
Huh. A remarkably simple and effective technique, and somewhat uniquely effective on jellyfish. Very neat.
Now we need to turn all the glowing and actions into labeled data and run in through a deep neural network