A major advantage we have when trying to reverse engineer intelligence from a biological example is that we know that it must have evolved. That the structure of a cortical column must have had some function in a form less complex than it is now. Recently I have started to explore how examples of convergent evolution have resulted in different solutions to intelligence, with the assumption that some sort of HTM must at least be partially present in a primitive non-mammalian species. With the hope that I could glean some new understanding by seeing the same problem solved in different ways. I found these papers interesting:
I then started to look for a more primitive ancestor and started reading up on fish brains, from there I started looking for examples of that genus with above average intelligence in the hopes of finding a “first brick”, on the road to HTM. What I found was that fish are actually far more intelligent than we give them credit for.
Finally I stumbled across something that got all the warning lights in my mind going:
Cross-modal object recognition and dynamic weighting of sensory inputs in a fish
This study found that a very particular species of fish “gnathonemus petersii” has the ability to do Cross-Modal object recognition, something that would require some form of HTM or comparable alternative to accomplish. I then started to look for research on the brain of the fish to see if it had any unique brain structures that might be indicative of its unique mental capabilities and found that this fish has been the subject of countless scientific studies due to an unusually large brain size for a fish and its unique structure of the cerebellum “valvula or gigantocerebellum” that has expanded to cover most of the fishes brain in the same way the cerebral-cortex has in mammals. This got me thinking that this could be a good candidate for a convergent HTM solution.
The Neuronal Organization of a Unique Cerebellar Specialization: The Valvula Cerebelli of a Mormyrid Fish
I am still trying to work out how a valvula ridge would work as an alternative to a cortical column, the density of neurons is far lower. but a single ridge runs the width of the brain. The bigger hurdle would be how a cell enters a predictive state without a pyramidal neuron. I do know that the valvula has some unique cell types “deep stellate cells”.
Anyways not sure why I am sharing my metal ramblings with you guys, other than the fact that you are the only people I know that might find some of this interesting, although I am no biologist and am probably well out my depth and well off mark. I just could not read these papers regarding a fish brain and not think HTM and thought that was worth sharing as the neo-cortex gets all the love:
“In contrast to the cerebellums of mammals and other vertebrates, the organization of both cerebellar efferent neurons and Purkinje cells in the mormyrid valvula seems to optimize the processing of specific parallel fiber input patterns at the expense of mossy fiber as well as climbing fiber input patterns.”