predictive coding explains consciousness

[scruffy]

Physics:

In a Hopfield network (which is omniconnected binary), the "state" of the network is determined by the contribution of each element to the Hamiltonian.

(You can easily find the original paper online, Google "Hopfield 1982").

https://www.pnas.org/doi/10.1073/pnas.79.8.2554

Whenever a neuron changes state, the network has by definition also changed its state. Therefore, if we can divide physical time (which in our computer model becomes Monte Carlo time) into "sufficiently small" intervals so we have one neuron changing state every .05 picosecond or so, then the entire network is changing state that fast.

So in essence what we have is a very special case of "phase velocity", where things seem to move faster than the speed of light. If it takes a fraction of a nanosecond for an electromagnetic wave to get from one end of the brain to the other, then the network has changed state 10,000 times in that interval! !!!

What does this mean? What does it mean when the network changes state? To change state, a bit has to flip, which means information determines state. State changes are information driven.

Why is the "speed of light" important, why am I mentioning it at all? Physics. Atoms, molecules, chemical reactions. It's all about the limit as dt => 0. Because the Hawaiian earring is fractal, and when you zoom in on the origin those hoops get mighty mighty small.

Is this even physically possible? Can I get information from one end of the brain to the other "faster than the speed of light"? Yes, I can. The Planck time is 10^-43 seconds, I have 20 orders of magnitude headroom. It's all about the spike times. Could I, say, cause a bit to flip on the opposite side of the brain before my EM signal gets there? Sure I can - I just have to do two two things:

1. set the state so that the desired bit will flip
2. ensure that the desired neuron is the next one to be updated

For point 2, what do you call it when two neurons need to fire together, or close to each other sequentially? "Coherence". Synchrony. You can see it in the spectra, in the EEG, MEG... heck, you can just stick two electrodes anywhere in the brain and you'll get some coherence.

Offered as food for thought...

 

[Hafar1014]

Thank you. I respect your experience in psychology.



That is a provably false statement.

When you're talking mechanisms you're in my domain. Psychologists know very little about synaptic plasticity.



What are the other four?



I can't define "thought". Can you?



That is a very high level view.

In the interest of landing the plane, my response would be something along the lines of:

Bidirectional synapses in the hippocampus cause plateau potentials that rapidly relocate the receptive fields of place cells.

I try not to use words like "thought", all I know about a thought is I experience it. Beyond that, what is it? I can't say. It's certainly not a data pattern, because machines have those and they don't have "thoughts".

As a psychologist you surely know of the Libet experiments. Psychophysics treats a human being like a black box, it seeks to directly measure experience through reporting. However the reporting of adjacent finger pricks is something different from a verbal reporting of stream of consciousness ("free association" let's say).

The point being, all thoughts are not the same. Pain entering your consciousness is not the same as reading a book, they have distinctly different time courses and different brain areas are activated. Did you ever work with fMRI? You can see it quite clearly that way.

Here is a challenge for you: prove that thoughts originate in the limbic system.

The limbic system is the most powerful brain system. Its relatively small. Every brain works the same way as a process. I have total understanding of plasticity. Trauma therapy can change the hard wiring in the brain.
The other brain systems mange homeostasis.
Emotion is the fuel that drives the personality not reason or rational thought.
Do you truly know what you feel or does what you feel determine what you think you know?

Emotions originate in the limbic system. Thoughts in your context explicit thoughts are created in the prefrontal cortex.
Active brains scans show the pathways and power.
The LS doesnt need to sleep or rest and has unlimited power. The PFC must eat and rest.

Suppose you see an animal threatening you. Your LS Is activated and a message based on memory expressed as emotion will go up to the PFC which sets goals and actions.

Even just crossing the street. You look left and right. Every thought starts as an emotion then becomes action.

 

[sparky]

86 billion neurons. Its average dimensions are about 140mm wide, 167mm long, and 93mm high, and it's roughly 1300 cubic centimeters in volume, though sizes vary, with men generally having slightly larger brains than women


so do women have the same amount of neurons in a smaller space....???? ~S~

 

[scruffy]

so do women have the same amount of neurons in a smaller space....???? ~S~

Yes. Women's neuron density is considerably higher, especially in the cerebral cortex. Men's heads are generally a tad bigger, and the difference kinda evens out. The total neuron count is still a little different but a lot of that is because of differences in sex related brain wiring. Women just don't need certain heavily controlled and complex male hormonal behaviors.

 

[sparky]

Yes. Women's neuron density is considerably higher, especially in the cerebral cortex. Men's heads are generally a tad bigger, and the difference kinda evens out. The total neuron count is still a little different but a lot of that is because of differences in sex related brain wiring. Women just don't need certain heavily controlled and complex male hormonal behaviors.

well.......that kinda explains a lot Scruff.........~S~

 

[scruffy]

I have total understanding of plasticity.

No offense meant, but I doubt it.

It's not your fault, no one does.

I don't think psychologists know the most basic things about some of the psych meds that are being prescribed.

What does serotonin actually do? How does it affect synaptic transmission? (Hint: it acts differently in the hippocampus than it does in the amygdala).

Consider dopamine. The same identical D2 receptor can have different postsynaptic effects depending on what other molecules it's attached to. If you simply put Haldol into someone, it affects all these versions at the same time. Some get excited, some get inhibited, some get modulated, and in others nothing happens. No one knows the full scope of the action.

I do some work with transcription factors related to synaptic expression. This is one of the few ways we can actually tease apart the various actions of a prolific neurotransmitter. The lab I work with has a knockout mouse that's missing one of the transcription factors for a glutamate receptor specifically in the amygdala. As you may know glutamate is the most common neurotransmitter in the brain, and the two prevalent receptor types are NMDA and AMPA. We have a mouse model of specifically the amygdalar symptoms of Urbach-Wiethe disease, that has nothing to do with lipoid proteinosis. In other words we accomplish the same thing a different way.

Brains are complicated. The wiring is very specific. A receptor here is not the same thing as a receptor there. It seems to me any serious study should attempt to merge the top-down and bottom-up views, and that is difficult because so much of the bottom-up information is still missing.

AI helps a lot. The best way to understand a system is to build it yourself, and right now the middle ground is at the level of neural networks (both real and artificial). The bottom-up view of molecular and electrical neuroscience is meeting the top-down view of behavioral and clinical psychology, at the level of machines that learn. So far we've learned one essential thing that is currently disseminating through both camps: spike times matter.

And spike times, are way complicated. One can not make simple statements like "serotonin modulates the target", such a statement tells us relatively nothing. We'd like to know what kind of attractor is in play and what it looks like, and exactly how serotonin affects the trajectories. Why is changing antidepressant dosages dangerous? Because there is plasticity attached not only to serotonin receptors, but also to downstream processes that are modulated by them, and astrocytes, and blood vessels.

 

[scruffy]

Let's talk about serotonin receptors. There are 7 (known) families. SSRI's affect three of them, numbered 1, 2, and 3. Types 2 and 3 are usually excitatory, while type 1 is usually inhibitory. They're all found in the same brain areas, like the hippocampus, medial prefrontal cortex (dorsal and ventral), and amygdala.

Major depression results in neuron changes all over the brain. Loss of neurons and synapses and shrinkage of dendrites in the hippocampus and prefrontal cortex, but increased growth and connectivity in the amygdala. At the point of clinical efficacy, these changes are reversed in the hippocampus and cortex, but not in the amygdala.

SSRI's interact directly with the three receptor families, as well as the reuptake transporter (SERT). And, there are indirect effects. Action on 5HT1 autoreceptors in the presynaptic membrane affects SERT regulation. All these actions have different time courses. Clinical efficacy correlates best with dendrite growth in the frontal cortex and new neuron growth in the dentate gyrus of the hippocampus (which is one of the few brain areas supporting such growth). Obviously this is a form of "plasticity" but not the only form.

Multiple serotonin receptor types can exist in the same synapse. What happens when the target is excited and inhibited at the same time? Furthermore, serotonin diffuses directly across cell membranes independently of any synaptic activity. https://royalsocietypublishing.org/...568/All-the-brain-s-a-stage-for-serotonin-the

Psychologists want to help people through treatment, and that's great, so do I. I want more specific drugs with fewer side effects. Remember before SSRI's there was desipramine and before that there was ECT. No one still knows how any of those things work, except in the most general way. But as we get better treatments we can help more people, and that's quite sufficient for now.

 

[scruffy]

Here, I'll describe the central concept of this thread graphically. Start with a layered neural network, like this:

The layers called "Convolution" suffice for an example. You can see the layer called "Input", let's say that's the retina. And let's say the convolution layers are our visual cortex, so from front to back we have V1 => V2 => V3 and etc.

So, you can trace a retinal signal through the network. Each layer has a characteristic delay. The N70 evoked potential occurs primarily in V1, and then it becomes the N100 in V2, and so on. These are discrete processing stages with discrete delays. But now we're going to take the whole set of layers, and embed it into a gigantic Hopfield network - likely a deep network with way more neurons than there are in the source layer.

So in a Venn diagram, the layers are a subset of the whole space. In this diagram the layers are A and the whole space is B.

What will happen is the larger network will come to predict the trajectories between the layers, as if there were additional layers there! !!!

The larger network will make its best guess as to the transformations from one layer to the next, and over time these guesses will become quite accurate. The larger network thus "embeds" the layers, and any fraction of any pattern appearing in the layers will evoke the entire pattern in the larger network.

In this way, the larger network becomes "conscious of" the transitions in the layers. Let's say I previously learned that eating candy is good. Now a piece of candy appears on the retina. The instant the information appears at V1, the whole pattern appears in the larger network and I already know I'm going to eat it. As I eat it I am "conscious of" eating it because every prediction made by my brain is being matched with the consequences.

This process of extrapolation occurs in the limit as dt => 0, which according to previous estimates is in the 50 picosecond range times the number of bit flips representing the distance between the information ("Hamming distance" in the binary case, or more generally Ln norm in the continuous case).

 

[scruffy]

By the way, the most effective treatment site for major depression is the subgenual cingulate cortex, an area which is neither PFC nor amygdala.

And, we can not yet explain the dramatic antidepressant effects of ketamine and scopolamine.

And oddly enough, vagus nerve stimulation sometimes works as a last resort in patients where all other treatments have failed.

Patients Once “Paralyzed by Life” Show Lasting Recovery With Implanted Nerve Therapy

Research shows that vagus nerve stimulation can provide long-lasting benefits for patients who do not respond to conventional treatments Around 20% of adults in the United States will experience major depression at some point in their lives. Many people feel better after trying a few treatments.

scitechdaily.com

 

[scruffy]

This is one of the better videos on predictive coding.

This is Dr Karl Friston from Univ College London, talking about free energy. (Which in machine learning terms equates with prediction error).

It's only about 15 minutes long, and it explains the vocabulary I've been using, like Kalman filtering and Bayesian inference.

 

[Hafar1014]

No offense meant, but I doubt it.

It's not your fault, no one does.

I don't think psychologists know the most basic things about some of the psych meds that are being prescribed.

What does serotonin actually do? How does it affect synaptic transmission? (Hint: it acts differently in the hippocampus than it does in the amygdala).

Consider dopamine. The same identical D2 receptor can have different postsynaptic effects depending on what other molecules it's attached to. If you simply put Haldol into someone, it affects all these versions at the same time. Some get excited, some get inhibited, some get modulated, and in others nothing happens. No one knows the full scope of the action.

I do some work with transcription factors related to synaptic expression. This is one of the few ways we can actually tease apart the various actions of a prolific neurotransmitter. The lab I work with has a knockout mouse that's missing one of the transcription factors for a glutamate receptor specifically in the amygdala. As you may know glutamate is the most common neurotransmitter in the brain, and the two prevalent receptor types are NMDA and AMPA. We have a mouse model of specifically the amygdalar symptoms of Urbach-Wiethe disease, that has nothing to do with lipoid proteinosis. In other words we accomplish the same thing a different way.

Brains are complicated. The wiring is very specific. A receptor here is not the same thing as a receptor there. It seems to me any serious study should attempt to merge the top-down and bottom-up views, and that is difficult because so much of the bottom-up information is still missing.

AI helps a lot. The best way to understand a system is to build it yourself, and right now the middle ground is at the level of neural networks (both real and artificial). The bottom-up view of molecular and electrical neuroscience is meeting the top-down view of behavioral and clinical psychology, at the level of machines that learn. So far we've learned one essential thing that is currently disseminating through both camps: spike times matter.

And spike times, are way complicated. One can not make simple statements like "serotonin modulates the target", such a statement tells us relatively nothing. We'd like to know what kind of attractor is in play and what it looks like, and exactly how serotonin affects the trajectories. Why is changing antidepressant dosages dangerous? Because there is plasticity attached not only to serotonin receptors, but also to downstream processes that are modulated by them, and astrocytes, and blood vessels.

Meds arent the topic. Youre coping and pasting concepts you dont understand as they relate to human psychology. None relate to the topic. I suggest you read Alfred Adler and learn about personality development and goal oriented behavior.

 

[scruffy]

The above is where neuroscience and machine learning meet. For example - the physics of free energy informs us about a synaptic wiring pattern called feed-forward inhibition. (That's not exactly what it is, but that's what it's called).

Feed forward inhibition is ubiquitous in the brain. It looks like this:

Input => E + I
E => I + Output
I => E + I

E is excitatory neurons, and I is inhibitory neurons.

In this pattern the input goes simultaneously to E and I. Since there is a very tiny delay through the inhibitory neuron, what happens is you get one spike out of E and then it gets inhibited.

The inhibition primes the E neurons for plateau potentials, so when they come out of the inhibition they're going to burst. By that time though, the input has been integrated in the dendritic tree, so the E neuron can then generate a burst whose spike count matches the output

So in the network as a whole, you first have time-to-first-spike, which tells you where the hot spots are (the hot spots arrive earlier and because of the I neurons they tend to inhibit the later arrivals). Then after the first spike you have a small silent period during which network integration occurs, and then you get the network output in the form of a series of bursts with rate codes.

You'll note this is halfway to dynamics, and machines are still incapable of this kind of activity ("because" there's no dynamics in them). So neuroscience informs machine learning too. The machines have to be able to do this, and currently can't.

 

[scruffy]

Meds arent the topic. Youre coping and pasting concepts you dont understand as they relate to human psychology. None relate to the topic. I suggest you read Alfred Adler and learn about personality development and goal oriented behavior.

I'm an engineer, not a psychologist.

I don't make statements about personality, that's not my domain.

My domain is physics and computations, which these days necessarily involves topology.

I read Adler long ago, it was required reading in undergrad psych. It's an okay model, there's nothing wrong with it, it's just not the way I choose to look at the world. Personality is data, whereas I'm interested in infrastructure.

Meds are the topic insofar as they alter consciousness. Consciousness has nothing to do with personality, it's a physical process that's common to all of us.

 

[scruffy]

Meds arent the topic. Youre coping and pasting concepts you dont understand as they relate to human psychology. None relate to the topic. I suggest you read Alfred Adler and learn about personality development and goal oriented behavior.

Here's an example of the kind of thing I'm interested in. It turns out, the same identical neural network with the same synapses can hold and process 1000 times as much information when it's operating in chaotic mode.

"Chaos" though, is the knife's edge of instability. When the network becomes unstable you get epilepsy. So how is chaos controlled in a neural network? One answer is the hot spots I mentioned.

And how exactly do computations work in a chaotic environment? Seems kinda counter-intuitive, don't it? But no... you can create the full spectrum of chaotic logic gates, just like you can create the full spectrum of quantum logic gates.

Chaos computing - Wikipedia

en.wikipedia.org

 

[Hafar1014]

I'm an engineer, not a psychologist.

I don't make statements about personality, that's not my domain.

My domain is physics and computations, which these days necessarily involves topology.

I read Adler long ago, it was required reading in undergrad psych. It's an okay model, there's nothing wrong with it, it's just not the way I choose to look at the world. Personality is data, whereas I'm interested in infrastructure.

Meds are the topic insofar as they alter consciousness. Consciousness has nothing to do with personality, it's a physical process that's common to all of us.

Youre talking about neuropsychology. As far as altering consciousnesses thats not quite correct. Let me offer you the psychology involved and together we can see things better.

Consciousness is simply self aware. Lets call that a broad perspective. The human mind doesnt equate to a computer model. Forget under grad psych.
Medications have different specific effects.
Benzodiazepines alter mood by suppressing anxiety.
Anti depressants effect depression and some anxiety as well like Wellbutrin.
Anti psychotics suppress delusional thinking.
Consciousness cant be altered unless we give a general anesthetic. Then youll know and feel nothing. Literally unconscious

Consciousness is a metaphysical process since it exists after the body dies. You cant separate it from the personality. Youre conscious of who you are as an individual.

Modern therapay especially my field which is trauma therapy is based on neuropsychology. I can in therapy rewire the synapse in your limbic system that supports the emotions created by trauma and PTSD.

Personality is life experiences transformed into synapses emotion and memory.
What is your main interest? Your Adlerian goal

 

[scruffy]

.
Consciousness cant be altered unless we give a general anesthetic. Then youll know and feel nothing. Literally unconscious

That's covered in the Toker link. And I added the bit about electrical stimulation of the claustrum, which has been replicated at least half a dozen times.

Youre conscious of who you are as an individual.

On this point we agree. And it's exactly what Friston is talking about.

Modern therapay especially my field which is trauma therapy is based on neuropsychology. I can in therapy rewire the synapse in your limbic system that supports the emotions created by trauma and PTSD.

Excellent. That's important work. I can relate to it personally.

Personality is life experiences transformed into synapses emotion and memory.
What is your main interest? Your Adlerian goal

I think we're looking at the same thing from different angles. We should definitely talk, as professionals vocabulary is important. Consciousness is different from awareness, yes? Babies are aware of "red" long before they're conscious of it.

Right now my goal is to modify TensorFlow to be able to handle dynamics. This is vital for ongoing research in the machine learning field, which right now is informing neuroscience about synaptic mechanisms and vice versa. The array of varieties of neural plasticity is impressive.

For a living (kind of ha ha) I work with molecules, mostly glial cells related to PML. I do the number crunching and visualization, I let others do the mouse killing cause it's too traumatic for me.

 

[Hafar1014]

That's covered in the Toker link. And I added the bit about electrical stimulation of the claustrum, which has been replicated at least half a dozen times.



On this point we agree. And it's exactly what Friston is talking about.



Excellent. That's important work. I can relate to it personally.



I think we're looking at the same thing from different angles. We should definitely talk, as professionals vocabulary is important. Consciousness is different from awareness, yes? Babies are aware of "red" long before they're conscious of it.

Right now my goal is to modify TensorFlow to be able to handle dynamics. This is vital for ongoing research in the machine learning field, which right now is informing neuroscience about synaptic mechanisms and vice versa. The array of varieties of neural plasticity is impressive.

For a living (kind of ha ha) I work with molecules, mostly glial cells related to PML. I do the number crunching and visualization, I let others do the mouse killing cause it's too traumatic for me.

Terminal lucidity
According to the neurological this cant exist but it does. Consciousness exists when the body dies or is near death

Terminal lucidity is a rare, unexpected return of mental clarity, memory, and personality in a person with severe psychiatric or neurological conditions (like dementia or schizophrenia) shortly before death, often lasting minutes to hours, providing a poignant final connection before they pass away, with potential causes linked to shifts in brain chemistry or neural activity as the body shuts down, though the exact mechanism remains unknown.

Signs & Characteristics

• Sudden Clarity: A person who was unresponsive or incoherent becomes alert and communicative.
• Memory & Personality: They may recall past events, recognize loved ones, express needs, or engage in meaningful conversation, acting like their former self
  .
• • Brief Duration: The clarity typically lasts hours, sometimes days, but generally ends with death soon after.
  • Associated Conditions: Observed in patients with dementia, Alzheimer's, brain tumors, stroke, meningitis, schizophrenia, and severe psychiatric disorders.

Consciousness is the soul

 

[scruffy]

Consciousness is the soul

To prove that, you have to establish the physical link.

This is why I say we're looking at the same thing. You hear me talking about weirdo stuff like the speed of light, we're looking for the physical mechanism supporting consciousness. And because consciousness is oh so complex, we start simple with awareness, and even that is impossible to explain at this time.

However we're getting close. The physics of predictive coding is very powerful. It's physics, like electricity and magnetism. At the end of the day it exists at the same level as Schrodinger's equation and Heisenberg's uncertainty principle. And this, we discover and prove by looking at neural networks in the limit as dt => 0 ("spike times matter").

The more your statement above is correct, the more physics matters. The universe works on physics, so if there's life after death, that works on physics too.

 

[Hafar1014]

To prove that, you have to establish the physical link.

This is why I say we're looking at the same thing. You hear me talking about weirdo stuff like the speed of light, we're looking for the physical mechanism supporting consciousness. And because consciousness is oh so complex, we start simple with awareness, and even that is impossible to explain at this time.

However we're getting close. The physics of predictive coding is very powerful. It's physics, like electricity and magnetism. At the end of the day it exists at the same level as Schrodinger's equation and Heisenberg's uncertainty principle. And this, we discover and prove by looking at neural networks in the limit as dt => 0 ("spike times matter").

The more your statement above is correct, the more physics matters. The universe works on physics, so if there's life after death, that works on physics too.

NDE research and I stated terminal lucidity also proves it.
TL involves irreversible damage to the brain resulting in severe loss of memory and personality. Then all of sudden it returns. How? The brain is damaged. Consciousness is energy. Energy cannot be destroyed only changed. People who experience this die 1-2 days after it occurs.

I have had many personal experiences with the after life. We dont end when our bodies die. The energy that holds our emotions memories and personality continues.
Consciousness is energy

 

[scruffy]

Consciousness is energy

And, the physics part links energy and information.

Boltzmann did this for an ideal gas, which is so nicely behaved it's never found in nature.

Then information theory came along, adding more detail to the relationship between information and entropy.

Now there is information geometry, which addresses both energy and entropy. And this is part of what Friston is talking about, when he links free energy with prediction error.

Changing a bit is expensive, it costs energy. You're talking about rewiring synapses, that costs. You have to put energy into it. All that energy goes somewhere, it becomes potential energy in the synaptic matrix of the neural network.

And from there, the potential energy can be converted to other kinds of energy. Neural networks work on an energy "surface" that has a shape, with hills and valleys. Like all things in this universe, the energy seeks a minimum. And there is a Hamiltonian attached to this dynamic system, an equation that describes the energy.

At the point where the brain no longer functions, the energy has to go somewhere else. It doesn't just evaporate, that never happens because there is conservation of energy.