[ryanmetz]

Back when I was a neuroscientist I had a fun conversation at SfN (the largest neuro conf in the world) with a big up-and-coming researcher about the cerebellum. Her take was "I hate that thing. If I never have to read another paper about it again that would be great. It's dumb. I wish we could get rid of it." I wonder how she's feeling about it these days...

[Animats]

That attitude is a big problem. The cerebellum is where all the important stuff happens. In the lower mammals, the cerebellum is most of the brain. The cortex is just the back-seat driver sending goals to the cerebellum.

Bio researchers have done work on decorticated cats. "The cats ate, drank and groomed themselves adequately. Adequate maternal and female sexual behaviour was observed. They utilized the visual and haptic senses with respect to external space."[1] The cortex is optional for basic survival.

The cerebellum had been far too neglected in AI research. I used to refer to this as the "hole in the middle" of AI. We had the expert systems guys working on logical abstractions, and the behavior-based people working on near-stateless stimulus-response systems. Not much in the middle. That's what got me interested in legged running for robots. But it turns out that's better approached as a dynamics problem than as an AI problem, so that didn't lead to a cerebellum. Just a lot of banging on differential equations. On most practical problems, it's easier to engineer a special case solution than to develop something cerebellum like. So there's been low pressure to fix this hole.

The really important stuff in life is getting through the next 10 seconds without a major screwup. If that doesn't work well, and consistently well, survival is unlikely. AI remains bad at this. Robots have this problem big-time. For self-driving cars, vast efforts have been required to make it work at all.

The cerebellum evolved first. The cortex is a relatively modern development. If we really knew how it worked and could make a good functional one for a robot, we'd probably have most of the parts needed for a cortex. But we don't.

When you see a paper like this, [2] you realize the extent of our ignorance. This is like cutting an IC into little chunks and doing a chemical analysis on each chunk to figure out what it does.

[1] https://link.springer.com/article/10.1007/BF00234897 [2] https://www.mpg.de/12027342/molecular-atlas-reptile-brain

[georgewfraser]

Decorticated cats have their whole midbrain, it's not the cerebellum that's doing all that. Also, the cortex is more dominant in primates. Decorticate humans can't do much of anything. Also, humans can live without a cerebellum:

https://en.wikipedia.org/wiki/Cerebellar_agenesis

[MAXPOOL]

> The cerebellum had been far too neglected in AI research.

Cerebellar Model Articulation Controller (CMAC) https://en.wikipedia.org/wiki/Cerebellar_model_articulation_...

A Historical Review of Forty Years of Research on CMAC https://arxiv.org/pdf/1702.02277.pdf

[thinkingkong]

My family has a history of cerebellar ataxia. If youve ever gone through knowing someone experiencing that its pretty horrible. The physical symptoms are bad enough but there seem to be common emotional and logical regulations that leave as the cerebellum dissolves.

I wish it upon noone.