The Cerebellum and Cognitive Function: Anatomical Evidence from a Transdiagnostic Sample
The more I sit on this paper, the more stunning it is. Like it changes SO damn much.
Like from a network perspective, dopamine works as a cerebellar coordinating signal. The entire purpose of dopamine signalling pathways is to transmit information to the cerebellum for post processing. We’ve had this idea of the dopamine network going up from the brain stem, when really it all goes down (or up from under the brainstem) pushed to the cerebellum, which computes updates, and pushed back out for re-integration. HOLY SHIT.
So one of the weirdest parts of cerebellar processing is that most of the rest of the nervous system has very clear bi-directionality. EXCEPT for the part of the nervous system that has all the cells. In the cerebellum all information comes in across the ponto-olivary bridges leaves in one direction.
Think of images like this one which show the flow from the brainstem out to the rest of the nervous system (if this was complete, it would also have the rest of the body as well). Instead, information is flowing IN THE OPPOSITE DIRECTION. Thing of dopaminergic pathways as “PULL” circuits rather than “push” circuits. Alright that’s just freaking me the hell out, and I could go on for awhile just on that one but this study goes SO MUCH HARDER.
So this thing also suggests “schizophrenia/autism” (and to an extent “bipolar”) happens. We’ve been sniffing around this for awhile, but “schizophrenia/autism” are like the result of either “overinformation” writes or “underinformation” writes. Meaning when it’s pulling information from the nervous system, the cerebellum is either getting too much information (“autism”) or not enough information (“schizophrenia”). Ermagerd.
It’s also suggesting that one of the mechanics for “depression” and “anxiety” is either suppressed writes (ventral habenula) or juiced writes (dorsal habenula, however all writes to the dorsal habenula are technically juiced so no changes in morphology there like “depression”. Think of it more like the ventral habenula squeezing off the dorsal side in depression).
It’s suggesting the mechanic for nearly all the “dys” prefix traits, like dyscalcula, dyslexia, or the more classically assigned dysmetria.
In otherwise healthy individuals, this paper lays the framework for manipulating nearly all of the above states, non-invasively. I’ve gotta stop here or I’m going to have a stroke, but HOLY FUCKING SHIT.
edit: Dorsal vs. Ventral dominance is a measure of how much of the stream gets stuffed through the cerebellum. Limbic circuits re-weight dorsal vs. ventral processing. The mechanic of the weight is error processing, all of “conscious” experience is a prediction. We are only conscious/aware of certain things, colliculi focus attention in such relatively small field at a time because limits the amount of cerebellar update processing required.
edit 2: Homo Technius is going to be be like… way way way smarterery than we are. Like, the complexity of their constructs is going to be absurd. They are going to be able to do massive levels of recursion on every single object model the cerebellum generates, resulting in constructs that are just way, way beyond what we can come close to right now.
edit 3: This means that cerebellar processing must be REALLY limited compared to cerebral processing, like it must be functioned locked to only one to four processes at a time, while cerebral processing can operate on pretty much an unlimited number of processes at the same time. It’s up the to brainstem to decide which priorities are most important for cerebellar processing. Our limbic circuits offer “juice” to a particular function to push it through the cerebellar chute in the external/internal (dorsal/ventral) processing pathways (think “amygdala complex”, “hippocampal complex”, “lentiform complex”).
edit: Oh god, the entire “break it down into smaller parts” motif of most education systems is a direct artifact of creating context digestible for the cerebellum. Even learning things like “times tables” is all about creating tiny object models not so much to build associations, but to build an object basis for more complex operators/objects. This is consistent when learning an instrument, where things are generally taught a note at a time, or language, where we build syllable bases first.
It’s weird, it’s like the entire point of the cerebral areas in most brains isn’t to process the world, but to create optimally sized concept chunks for cerebellar processing. The “big picture” dorsal/cerebral dominant view gets decomposed into the “details” ventral/cerebellar context.
Okay, we probably need to redefine “object” here because it’s not an actual “physical” object, it’s more clusters of a similar metabolic vector. It really looks like nervous systems (at least in primate brains for sure) create object representations PURELY out of vector inferences.
For example, when we think of creating an object out of “color”, what we are really talking about is levels of intensity across a map of our visual receptors, differentialed against each other. The strength of those differentials is what nervous is what nervous systems use to differentiate color. “Opposing” colors are opposing not because of any particular property of the color itself, but because those colors generate the greatest differentials against each other. Regardless of the color however, it is the overall intensity of the received information relative to it’s neighbors which allows mapping into discrete objects.
Mapping relative to it’s neighbors is pretty much how “vision AI” works and distinguishes objects.
So lets assume that that information comes in over the first order sensors, gets computed into a map, and specific patterns of data including their flux vector (it’s not just about the color itself, but how stable/unstable the data being pushed into the map is), are tagged by specific functional modules if they match particular saved behavioral information. These functional modules can then either metabolically “juice” or “decay” the vector for their specific part of the stream they are responsible for processing.
The switch in the brainstem happens based on this metabolic weighting, it pushes concentrations of high metabolic weight into the DCN for further processing, including “prediction planning” or “smoothing out the metabolic spike”. These processing updates are written directly into the maps themselves. Essentially, nearly all nuclei work as error prediction modules under this conceit.
Error prediction is the differential between the cerebellar update and current information state. There’s a second component to this, which is “sensitivity”, nervous systems across organisms don’t have the same sensitivity across their functional modules, and each level of abstraction from that adds another level of variance (e.g. intra-group, intra-species, intra-class, etc).
We could describe “dorsal dominant” phenotypes under this construction as individuals with high sensitivity to changes in vector information (relative to other modules/organism/group/species), and “ventral dominant” phenotypes as low sensitivity to vector information changes.
It’s interesting that these would create two completely different sets of expressed behavior, “dorsal” can be really expensive for the brainstem if it’s constantly being slammed by “update requests”, while “ventral” can be really expensive if it’s “error checking” isn’t sensitive enough to detect the error.
We could even describe attention as a whole as sustained high metabolic vector across one or more modules.
So how do we test this? Probably very similarly to what they did in this paper, for really gross macro manipulations, but if we had more specific circuit focus, we might be able to directly manipulate behavior on a scarily granular level. Like everything from hallucinating, to being able to learn a new concept almost instantly (assuming we had enough information about how a specific nervous system has stored and is processing it’s prior associations). HOLY SHIT. Argh, this is exciting as hell but also sum of all fears territory…
It’s like, the entire process of behavioral training is “teaching” us how to modulate the levels of vector sensitivity across various modules. We train higher sensitivity. Some individuals may have a dorsal/ventral bias, but the initial processing is still raw. These changes in sensitivity should be visible as morphology changes to nuclei tied to a specific function.
Geez, this concept answers a HUGE question I’ve always had! Why is true autodicatism so rare, and more why is it even more rare that true autodidacts are “grounded” in “reality”?
So like, we can decompose most education systems into a single dyad, “be more sensitive to this” and “be less sensitive to this”. From this concept, we can build rule sets to “teach” nearly anything. THAT’S REALLY COOL. Literally all of “I don’t get it” is either a) insufficient data or b) insufficient/oversufficient sensitivity to error. BRUH. If we can discretely stimulate the dorsal and ventral pathways of most nuclei, we can INDUCE THIS once we have a better idea of how each nuclei contributes to the global object/map.
So like imagine this, you’re “shy”. We could innervate (region? mammilary bodies? ventral hippocampus?) to turn down the sensitivity to error in the stream if we hit the right combination of pathways, while also juicing the cerebellar processing side to give it really strong metabolic write. Like once we figure out exactly what these various nuclei are doing functionally instead of what we want them to be doing (like trying to stuff psych language concepts, including “plasticity”), holy crap. This is on the order of gene manipulation!
ARGH. Okay okay okay. So, I definitely need to modify the tDCS rig to be HD, as we are going to need to manipulate multiple sites either simultaneously or in quick succession. HOWEVER, what about a HD-TMS rig?
This is so fucking irresponsible. It’s literally a mind control rig. Ugh……..
Okay, now I think I finally get why my brain has been sitting on these cerebellum papers, and these two in particular for the last week especially.
Thinking about the protocol for this, there are a few challenges before this is consistently testable, first we still don’t understand the timing of these yet. For example, do we need to prime the nuclei, then introduce the stimuli or introduce the stimuli then ennervate? Do we need to stimulate the entire path or just the functional module/nuclei (and whatvever cerebellar output circuits are necessary to juice the return write)? How much stimulation is necessary and for how long? How do we figure out how these modules are creating map differentials for behavior, is this done entirely in the rhinal cortexes or is this also a cerebellar function? Is it possible to verify the success/failure of a write on a short timescale or do they need to be “hard encoded” instead of mapped?
Just realized this probably won’t be effective as a recovery from dementia, but it may be usable as a way to write new behavior across the limited metabolic resources/circuitry in such a brain. Eh, maybe it can, but someone smarter than me will have to figure that out.
Oh god I’m so conflicted about this. My brain is telling me that this is going to work once we figure out the pre-requisites, but it’s also telling me… yo, humans and shit. Oh man, we can prove this out within a year even if we are just hunt and pecking/trial and errorring. This is the first time in a long time I’ve felt nauseated by thinking about something, definitely getting flooded here.
So there’s a missing step here somewhere as this would mean neurons are back propagating the signal.
Edit: Okay, step 1, we need to replicate/reproduce the finding in this paper: https://link.springer.com/article/10.1007/s12311-023-01649-8
———– Random context ——–
Driving and suppressing the human language network using large language models
Robust enhancement of motor sequence learning with 4 mA transcranial electric stimulation
——— Brain stimulation journal spam ——–
Transcranial ultrasound neuromodulation for epilepsy: A pilot safety trial
Lessons from multitarget neurostimulation in isolated dystonia: Less is more? – Or more appropriately, the right amount of stimulation is important rather than just stimulation itself.
Sex modulates the effect of HD-tDCS over the prefrontal cortex on the Iowa Gambling Task
Causal influence of brainstem response to transcutaneous vagus nerve stimulation on cardiovagal outflow – Starting to see “causal” pop up a lot more often in brainstem/cerebellum work
Effects of multisession cathodal transcranial direct current stimulation with cognitive training on sociocognitive functioning and brain dynamics in autism: A double-blind, sham-controlled, randomized EEG study – Heh, turning down the information flow for “autism”.
Cortico-spinal tDCS in amyotrophic lateral sclerosis: A randomized, double-blind, sham-controlled trial followed by an open-label phase – Getting excited about the increase in brainstem/spinal applications, remember the whole nervous system is an information processing unit, rather than just “brain”.
Cerebellar Asymmetry of Motivational Direction: Anger-Dependent Effects of Cerebellar Transcranial Direct Current Stimulation on Aggression in Healthy Volunteers – Okay so… this one is real deal. I didn’t personally experience as anger (and the cohorts I worked with awhile ago didn’t either), it was more of an “intensity”… it’s hard to describe, it felt more like a RUSHING. Holy crap I’m beginning to realize that we really didn’t even try to manipulate behavior via the cerebellum before recently! Like there were some tests for stroke and motion related things, but actual behavioral manipulation? Not really. Now in one week there are already four papers of with pretty interesting examples of it occurring in ways we never got out of cortical stimulation? Wow!
Electrical stimulation of the cerebellum facilitates automatic but not controlled word retrieval – So this is an insanely long post, but remember the first post, the one we are discussing with that banging .89 correlation coefficient for “cognitive flexibility”? BBQSAUCE. Here it is again, in another form!