Jeff and Niels review an idea proposed by Viviane around thalamocortical circuitry, hierarchical composition of objects and behaviors.
3 Likes
Interesting stuff, fully agree with Jeff when he was talking about how sensory input may be more involved in intelligence than was initially considered. Yes !, In my view sensory input is fundamental to everything we think and do, after all evolution started with sensory cells and then built a brain to make more advantageous use of the sensory signals.
I have just been looking at the sensory feedback signals from a robotic limb I have been designing.
I was trying to figure out how to build a neural network to create the walking motion, thinking that it would need to be some kind of RNN state machine. Then I plotted the sensory data over time and quickly realised that the movements are more like a melody than a state machine.
This theory has been explored in a 2014 paper by Josef P. Rauschecker - How the brain stores and retrieves musical melodies.
This paper supports what Jeff was contemplating, that our motor systems mimic our sensory input
and thoughts, we can choose to override the actual physical movement or not.
It may be that learning to walk or perform other actions is rather like memorising a melody, something I will investigate further. This process could well extend into vision and touch.
Making that pot of coffee is a ballet, not a to-do list 
3 Likes
From 8:00 to 9:30 in the video, there was uncertainty about whether input to V2 can come from both the LGN and higher-order thalamic nuclei. I believe you referred to a specificity of the cat visual cortex compared to the primate visual cortex. In cats, there are dense projections from the LGN to V2, such that “lesions of V1 alone do not impact greatly on visual processing, but such lesions do impair the perception of higher spatial frequency stimul”.
A random reference on this topic (area 17 is V1, and area 18 is V2):
Domestic cats are also special in that, for a time, their visual
system was the most intensively studied. As a result, as many
as 17 cortical visual areas have been proposed for domestic
cats [176]. Cats are highly visual, and their eyes are large
and forward-facing in this semi-nocturnal predator. The X
(P) cells of the retina project almost exclusively to the LGN,
while the Y (M) cells project to both the LGN and the SC, as
in primates [177]. The X cells of the LGN project, as expected,
to V1, and the Y (M) cells project to V1 and V2, while the W
cells project to V1, V2 and the lateral suprasylvian region, LS
and 21a. As in other mammals, the SC projects to a part of
the pulvinar that then projects to temporal cortex, which is
the LS (lateral sylvian) region in cats [178]. As both areas 17
and 18 receive dense inputs from the LGN in cats, lesions of
V1 alone do not impact greatly on visual processing, but
such lesions do impair the perception of higher spatial frequency
stimuli [179]. Lesions that include both areas 17 and
18 do impact on visual acuity and discriminations of stimulus
orientation [180]. However, these impairments are far less than
the cortical blindness produced by V1 lesions in primates.
https://royalsocietypublishing.org/doi/full/10.1098/rstb.2021.0293
3 Likes
Sorry to put admin in here, but is where is the todo chart you were working through?
You can find the chart here: Monty Project Overview - Google Sheets
We have some more explanation around it in the future work section of our documentation. There we also have more detailed writeups of what we imagine for many of these items (although for most of them we would still recommend writing an RFC first before starting an implementation. If you do we are happy to add you to that roadmap as well 
1 Like