Slashdot Mirror
AI Could Get Smarter By Copying the Neural Structure of a Rat Brain (ieee.org)
the_newsbeagle writes: Many of today's fanciest artificial intelligence systems are some type of artificial neural network, but they bear only the roughest resemblance to a biological brain's real networks of neurons. That could change thanks to a $100M program from IARPA. The intelligence agency is funding neuroscience teams to map 1 cubic millimeter of rodent brain, looking at activity in the visual cortex while the rodent is engaged in a complex visual recognition task. By discovering how the neural circuits in that brain cube get activated to process information, IARPA hopes to find inspiration for better artificial neural networks. And an AI that performs better on visual recognition tasks could certainly be useful to intelligence agencies.
C Elegans is am extensively studied nematode with exactly 302 neurons, whose contetome (wiring) is consistent and known.
But how its brain actually works remains a mystery. Neurons are complex, as is their interactions with the input and output.
Not much point looking at mice with many orders of magnitude more first.
Personally, I do not think that mapping neurons in detail will lead to AI. But if you are going to do it, start with something vaguely tractable. C Elegans.