C. elegans is probably the most well-understood organism in the known universe. It has a tractable number of cells, only a few neurons, and pretty much just swims around. But compared to most computer systems that do any sort of machine learning, it's a frickin' marvel of engineering. It builds itself from a single cell, swims around on its own, finds food, reacts to stimuli, manages to reproduce sexually, and never, ever, segfaults. While it can't be taught to play chess, it has sensory modalities that could potentially perceive a chess board, and if it was wet enough it could probably swim across it.

Lots of people are working on different aspects of machine learning and building neural turing machines and what have you, but who's working on designing the less-flashy faux-biochemical emotional circuitry that will keep all the higher level stuff grounded, and keep the combined entity metaphorically swimming forward when it doesn't know what else to do? Forget artificial intelligence, who's working on artificial hungriness? Artificial fear? Artificial boredom? How do you attach even a trivial attempt at artificial intelligence to even a simulated world in such a way that its actions, though almost assuredly not particularly well-selected, can meaningfully be described as actions? And how do you design that mediating layer so that your AI-controlled virtual robo-worm doesn't just keep banging its face against the same wall forever?

The way these lower-level systems work directly determines what the higher-level systems will try to do. The body controls the brain, not the other way around.