Non-human neocortex, sensory-motor inference

I was found interesting papers about neocortex of marine mammals.

For example, dolphins spontaneously learn associations between sounds and temporally paired events and
demonstrate extensive imitative abilities for sounds and for behaviors.
The cetacean neocortex appears agranular due to a lack of layer IV and different from human necortex.
In “Preliminary details about sensory-motor inference” I see belief is that layer 4 cells are performing sensory-motor inference.
Do you have any plans to use research on non-human neocortex or model it?


Hi Minou,
All mammals have a neocortex and non-mammals don’t. The neocortex of all mammals are remarkable similar in their highly detailed structure. The big difference between mammals of different species is the area of the neocortex and the sensors and motor areas that connect to the neocortex. The vast majority of research on the neocortex is already done with non-humans. Rats and mice are most common, but also cats, monkeys, and lots of others. There are very few invasive studies of humans but the accepted belief is that if you understand how the layers and cells work in a rat neocortex you will be 90%+ of the way to understanding human neocortex.

Speaking of dolphins. In my book On Intelligence I said that dolphins were an exception, that they only had the older style 3-layer cortex. That is not true. Apparently there was some debate about this in the past. Current thinking is their neocortex is the same. If you find new studies that suggest otherwise I would be interested in hearing about them.


Thanks for your reply!

I think saying that non-mammals lack a neocortex is just slightly misleading, though technically correct. Especially because you’ve said that you believe that a neocortex is required for “real intelligence,” whatever exactly that means. I think that “real intelligence” may actually be achievable in different ways too, not just with the exact same method that a neocortex does. There is plenty of variation in how the neocortex works when you look across different mammals, and so as more animals are studied, I believe it will likely blur the line between cortical and non-cortical brains.

For example, birds do not have a neocortex, but they do have areas like the nidopallium, which seem to serve a similar role. Saying that they are not really intelligent seems very absurd when you look at what they can do with their relatively small brains.. I remember reading a few years ago that the main nuclei in the nidopallium are interconnected in a similar way to layers 3, 4, and 5 of the cortex. I’ll have to see if I can find a source again to back that up though.

The way I see it, I think a lot of non-mammals probably do have similar circuits in their brain, but what differs between mammals and non-mammals is the way the circuits are organized. In mammals, the cortex is structured in layers, while in non-mammals it’s structured in nuclei (though I think I remember hearing that layered structures exist in some non-mammals too). If you think about it, a nuclei-based brain is going to be more neuron-dense than one based on laminae. That’s why a raven can have as many as 2 billion neurons in such a small brain.

The issue is scalability. If we assume that their brains follow similar organizational principles to a neocortex, we see a big drawback to nuclei; the square-cube law. If we assume that every pyramidal neuron (or whatever the analogue is in non-mammals) needs to send an axon to a neighboring region, then we run into an issue. If we double the radius of a nuclei (assuming a perfect sphere), the number of neurons (and consequentially the number of efferent axons) increases by a factor of 8, but the amount of surface area we have to fit all the axons only increases by a factor of 4. This means that we can fit a huge number of neurons into a small space, but we run into issues when we try to make very large brains.

Compare that to laminae; if scale a layer of neurons up, you probably won’t be increasing the thickness in most cases. So, you only double the length and width of the layer, meaning that the neuron/axon count scales by a factor of 4, and so does the area to fit all those axons. The only scalability problem is just finding enough space inside the skull to fit everything. That might explain why so many animals with comparable numbers of neurons to humans tend to be very large. Elephants have around ~6 billion neocortical neurons, gorillas have ~9 billion, and dolphins seem to range between 6 and 37 billion depending on the species (the long finned pilot whale was recently found to have more neocortical neurons than humans, as one of minou’s sources mention).


@Charles_Rosenbauer, I’m curious of the context for the quote about “real intelligence”. I have only seen this term used in reference to other machine learning strategies that are not biologically based, not as an argument that only mammals are truly intelligent. Rather, the argument that I typically see is that most of us can agree that mammals are intelligent, and so understanding the neocortex can help us to understand intelligence.

1 Like

Okay. Maybe there was just some misinterpretation on my part.

I recently reviewed papers on similarities/differences between reptilian and avian allocortex (3 layer) and mammalian neocortex (6 layer). There is a good chance that neocortex evolved as essentially the stacking of two regions of allocortex.

As @Charles_Rosenbauer points out, the connections between allocortical regions may have been functionally equivalent to neocortex. So there is some speculation that neocortex was successful more because of the ease of expansion than any superior learning capability (although this is highly speculative). Neocortical progenitor cells produce cells for all six layers. The precursor circuit required separate progenitors for the different regions of allocortex plus establishing cross-region connections. This would be much harder to scale up and neocortex may have been the key break through for creating a scalable version of the functional unit.

Interestingly, the ends of mammalian cortex are still allocortex. Anterially, the piriform cortex and olfactory bulb still use the three layer structure. And on the far other side (which has evolved to wrap underneath the neocortex) the hippocampus also uses the three layer structure.

Perhaps we can learn something about the role of having the upper and lower layers in neocortex by understanding why the hippocampus and olfactory areas need just one copy of this three layer circuit.