| Author |
Message |
samkane
Joined: 07 Jun 2005
Posts: 5
07-20-05, 06:26 pm |
 Post subject: Motor Cortex |
 |
|
I am a bit confused about the way motor cortex behaves according to the book (p.120)…
I understand that in the sensory cortex input goes up the hierarchy and prediction goes down. In the motor cortex, the motor command goes down but what goes up the hierarchy? The sensory input? |
|
| Back to top |
|
| Author |
Message |
Orion
Joined: 10 Apr 2005
Posts: 42
Location: Reed City, Michigan, U.S.
07-22-05, 01:04 pm |
| Post subject: |
|
|
How about the actual actions of the muscles?
_________________
"The more numerous the laws, the more corrupt the state." --Tacitus |
|
| Back to top |
|
| Author |
Message |
samkane
Joined: 07 Jun 2005
Posts: 5
07-23-05, 02:43 pm |
| Post subject: |
|
|
| Orion wrote: | | How about the actual actions of the muscles? |
Do you mean the information about actual muscles’ position? |
|
| Back to top |
|
| Author |
Message |
Orion
Joined: 10 Apr 2005
Posts: 42
Location: Reed City, Michigan, U.S.
07-24-05, 12:02 pm |
| Post subject: |
|
|
Well, I was thinking the electrical activity. That, of course, is the exact same info that's passed down. The motor cortex has to learn these commands while acting them out; perhaps it flows up simultaneously with down, somehow. You have a good question. I hope someone who knows more replies.
_________________
"The more numerous the laws, the more corrupt the state." --Tacitus |
|
| Back to top |
|
| Author |
Message |
Lawrence Phillia
Joined: 17 Jan 2005
Posts: 67
Location: Canada
07-24-05, 05:38 pm |
| Post subject: |
|
|
| Well i wouldnt say sensory only and we have to define what we mean by "UP" and "DOWN" in more detail. There are also lateral pathways involved that run parallel with commands from the cerebral motor cortex. A prime candidate would be the cerebellum, an organ specially designed for sensor-motor coordination. There must be and most likely is much dialogue between the motor cortex and the cerebellum. So in effect there are complimentary signals going up and down sensory-motor pathways so that adjustments can be made on-the-fly. |
|
| Back to top |
|
| Author |
Message |
csaba_trucza
Joined: 09 Nov 2004
Posts: 5
Location: Romania
08-02-05, 03:16 am |
| Post subject: |
|
|
I was pondering the same question lately, and while I'm completely unqualified to come with explanations, here it comes what I think.
By observing my 3-months old baby, I observed that she moves her legs and hands a lot. The movements are mostly uncoordinated and seemingly random.
However I observed that if I start to talk to her and caress her, she gets very active (like smiling, laughing, "talking back"), and her legs and arms start to move much more intensively.
My hypothesis goes something like this:
1. First, the visual signals (that's me coming to her) and auditory signals (that's me talking) enter her brain, they go up to the associative areas, activating regions after regions in their path.
2. During this activation, some of the higher level regions connected to the motoric regions gets excited, signals start to flow towards the legs and hands (downwards) and they start to move.
3. At the same time proprioperception regions start sending signals back from the legs and arms to the higher areas (up) and they eventually end up in the associative regions.
After some learning, the signals going down (in step 2. above) start to get refined based on the input from step 1. and input from step 3. And the movements start to become more refined.
There are a few interesting questions arising:
- There must be a way to get the incoming signals (step 1) to get down to the limbs. How the regions should be layed out so the signals actually get down there?
- For learning to happen the signals coming up (step 3) must get together with the signals coming in (step 1) and/or signals coming down (step 2). How you make sure the upwards going signals end up in the same regions as the downwards coming signals?
What do you think? Does it make any sense? |
|
| Back to top |
|
| Author |
Message |
Orion
Joined: 10 Apr 2005
Posts: 42
Location: Reed City, Michigan, U.S.
08-03-05, 08:44 am |
| Post subject: |
|
|
Lawrence Phllia said:
"we have to define what we mean by "UP" and "DOWN" in more detail."
According to my understanding:
In the majority of the cortex, that tied to senses, sensory data flows up the heirarchy, and predictions flow down it. Both flows are important to learning, but the upwards flow provides the new information that learning is based on. In the motor cortex, commands flow down to make the muscles move. The critical question is what flows up, to allow learning to take place.
Perhaps nothing flows up, and the motor cortex doesn't actually learn very well; this would explain why people who have no cerabellum must concentrate on every little movement. But even they can probably train their motor cortex somewhat.
If something flows up, I would think it might be some notion of sucess or failure. The lower columns would suceed or fail at their little jobs, and the higher columns would suceed or fail at their more complex tasks. This might not fit with memory-prediction, though... it sounds more like behaviorism, and reward/punishment is supposed to occur lower in the brain, not at the neocortex.
So I'm still unsure.
_________________
"The more numerous the laws, the more corrupt the state." --Tacitus |
|
| Back to top |
|
| Author |
Message |
Lawrence Phillia
Joined: 17 Jan 2005
Posts: 67
Location: Canada
08-03-05, 06:09 pm |
| Post subject: |
|
|
| Quote: | How you make sure the upwards going signals end up in the same regions as the downwards coming signals?
|
Yeah, you'd think there'd be some form of sensory-motor cognitive map of the entire body ( think Voodoo doll  ) that act as a central waystation or switchboard that could integrate sensory and motor signals in such a way that cognitive manipulation can occur .
As a possible example, the Golgi and muscle spindle organs can signal both muscle force/tension and muscle length respectively. By manipulating signals from these two organs, joint angles( muscle length ) and velocities ( length/time ) can be derived . PID controllers in robotics use a similar principle. Could loosely consider this a prediction of a motor command .
| Quote: | This might not fit with memory-prediction, though... it sounds more like behaviorism, and reward/punishment is supposed to occur lower in the brain, not at the neocortex.
|
Not sure either if this fits with the memory-prediction framework but the cerebellum could predict the onset of a motor command/reaction by "storing" the context inwhich a motor command ocurred as well as the tuned motor program itself, so that when this learned context presents itself again the command is no longer needed to initiate the motor program. |
|
| Back to top |
|
| Author |
Message |
samkane
Joined: 07 Jun 2005
Posts: 5
08-11-05, 07:10 pm |
| Post subject: |
|
|
Thank you for sharing your thoughts. I would like to summarize your comments and reformulate my question…
Everybody seems to agree that something does flow up in motor cortex. :
a) Some some notion of success or failure (Orion)
b) Some sensory input from muscles (Orion)
c) Sensory input for other regions of the sensory cortex, which also form some sort of feedback loop. (csaba_trucza)
After some closer reading, it seems that Jeff suggested that:
a) Motor cortex and sensory cortexes are very similar (p.132)
b) Exception can flows up the motor cortex (p.133 Move the troops to Florida analogy)
So my new questions are:
1) It seems that motor context cannot send an exception up the hierarchy without having an accurate prediction of that this action is supposed to accomplish. So is Motor cortex just a different type of sensory cortex in which sensory input goes up the hierarchy and prediction as well as motor command goes down?
2) Suppose motor cortex is NOT a different type of sensory cortex. It is closely tied to the sensory cortex but is distinct from it. What is the relationship between sensory and motor cortexes? |
|
| Back to top |
|
| Author |
Message |
Orion
Joined: 10 Apr 2005
Posts: 42
Location: Reed City, Michigan, U.S.
08-12-05, 06:21 am |
| Post subject: |
|
|
Obviously, we must have some way to decide to take a certain action. This is the basis for the connection between sense and motor commands. Is it possible for cortex to do such things by itself? Probably not.
_________________
"The more numerous the laws, the more corrupt the state." --Tacitus |
|
| Back to top |
|
| Author |
Message |
Don
Joined: 18 Feb 2005
Posts: 33
Location: Dayton, OH
08-27-05, 01:09 pm |
| Post subject: UP and Down |
|
|
Lawrence Phillia makes good points on this thread concerning (a) knowing what we mean by "UP" and "DOWN" and (b) the role of the cerebellum.
I think it is more illuminating to consider that activation in the cerebrum *resonates* -- it is not the *direction* but the *persistence* of activation across resonant subnetworks that (a) drives learning and (b) initiates motor plans and subplans (mostly in prefrontal neocortex).
Sensory neocortex is also resonant circuitry, so in this respect there is no difference between motor and sensory neocortex.
The cerebellum's primary outputs are all inhibitory, so it cannot resonate and it does make sense to talk about UP and DOWN signals to and from the cerebellum. The cerebellum receives UPWARD proprioceptive feedback from the periphery. Hence it can control and coordinate the fine motor execution of the neocortical motor plan. The cerebellum's function is to sculpt and finally stop the currently resonating-and-executing neocortical subplan so that the next subplan in a sequence can activate and execute.
_________________
Don Loritz |
|
| Back to top |
|
| Author |
Message |
eightwings
Joined: 07 Aug 2005
Posts: 29
Location: Miami, Florida
08-28-05, 08:07 am |
| Post subject: |
|
|
| samkane wrote: | So my new questions are:
1) It seems that motor context cannot send an exception up the hierarchy without having an accurate prediction of that this action is supposed to accomplish. So is Motor cortex just a different type of sensory cortex in which sensory input goes up the hierarchy and prediction as well as motor command goes down? |
I doubt that the motor cortex sends any exception (conflict signals?) anywhere. If anything, conflict signals are sent to the motor cortex from the motor command layer (basal ganglia). These signals are used by the motor cortex to decide whether or not to retain newly formed motor command connections. The question is, what constitutes a motor conflict?
In order to understand what a motor conflict is, we must first specify what is meant by motor action and motor command. A motor action is a physical effect on the system's environment which consists of the system's own actuators, i.e., muscles, motors, solenoids, valves, etc... Every action has at least one associated effector which is a motor command neuron that services a given actuator. Each effector has a unique intensity level and may receive multiple command connections from multiple sources. Likewise, an actuator may be controlled by multiple effectors. Every action has duration and thus a beginning and an end. As a result, every effector is controlled by two complementary types of motor commands: start and stop. That is, a command can be used to either start a motor action or stop it. There are two types of motor conflicts: out-of-step and multiple-activation.
Out-of-step Motor Conflict
An out-of-step conflict occurs when when an effector receives multiple concurrent start or stop commands or when a command tries to stop an action that is already stopped or start one that has already started.
Multiple-activation Motor Conflict
A multiple-activation conflict occurs when multiple effectors try to activate the same actuator concurrently.
In my opinion, this is all there is to motor conflicts, actions and commands. This is the primary purpose of the basal ganglia which is involved directly in motor coordination and control. It is all about timing. Conflict resolution simply consists of detecting conflicts as they happen and disconnecting bad command connections (synapses). These connections all have an initial strength which is increased by a small amount every time a command signal is received. However, whenever a conflict is detected, bad connections are strongly weakened. Eventually, only non-conflicting connections survive. But there is more to motor behavior than this. There is also concept formation and behavior selection. These work under a different but related principle.
| Quote: | | 2) Suppose motor cortex is NOT a different type of sensory cortex. It is closely tied to the sensory cortex but is distinct from it. What is the relationship between sensory and motor cortexes? |
I have reasons to think that the motor cortex must be the complement or inverse of the sensory cortex in terms of function. If you understand one, you also understand the other by simple inference.
_________________
Louis Savain
Temporal Intelligence |
|
| Back to top |
|
| Author |
Message |
curtwelch
Joined: 02 Mar 2006
Posts: 14
Location: Vienna, VA
03-02-06, 08:38 pm |
| Post subject: What the motor cortex is really doing |
|
|
| samkane wrote: |
Everybody seems to agree that something does flow up in motor cortex. :
a) Some some notion of success or failure (Orion)
b) Some sensory input from muscles (Orion)
c) Sensory input for other regions of the sensory cortex, which also form some sort of feedback loop. (csaba_trucza)
After some closer reading, it seems that Jeff suggested that:
a) Motor cortex and sensory cortexes are very similar (p.132)
b) Exception can flows up the motor cortex (p.133 Move the troops to Florida analogy)
|
I developed a theory a few years back that answers this question. It came to me as an insight in how to implement AI. I, like Jeff, have had a 30 year fascination with how to build a brain and have in my spare time over the years tried to answer the exact same questions Jeff has been looking for - that is, what is the overall framework that explains what the brain is doing to create intelligent human behavior. I agree a lot with what Jeff says in his book, but there are a few things I think he's got wrong. The motor cortex is one of them. He's got it upside down (as does everyone).
I believe the traditional view of the motor cortex as the output side of the cortex is all wrong. I think Jeff is right about the fact that the entire cortex does the exact same function. And if it's doing the exact same thing, how can it make sense to say that data flows up in the sensor cortex but down in the motor cortex? The answer is that it doesn't. The data flows up in in all parts of the cortex. The motor cortex is performing the exact same function as the sensory cortex. It's feed sensory data to be analyzed. But what is the sensory data it's being feed? It's feed the output data, from the lower brain.
The lower brain is there to receive sensory data, calculate some instinctive behaviors, and then produce the outputs needed.
The neocortex is a device which monitors what the lower brain is doing, and modifies it's behavior as required, though a learning process. So, all the inputs to the lower brain, are passed up to the sensor cortex for analysis, and all the outputs of the lower brain, are passed up to the motor cortex for analysis - just like the sensory side.
The entire cortex, both sensory, and motor, then feed correction signals down to the lower brain to modify it's behavior as required.
So, from this perspective, the sensory cortex is analyzing the external sensory data, and the motor cortex, is analyzing what the brain is currently telling the muscles to do. Both halves of the cortex, are performing the exact same function. The data which is feed to it to be analyzed, is flowing up in the hierarchy.
The sensory cortex forms invariant representations of what is happening in the external world, and the motor cortex is forming invariant representations of what the body is being told to do.
Now, here's the other half of the picture that Jeff missed about the cortex. Jeff for the most part just ignored the most important part of the problem - the problem for which the entire brain was built - output. The brain is there to control the body, yet Jeff just writes off output (behavior as he calls it) as if it weren't important. I think Jeff is dead on when he talks about what the cortex is doing in terms of analyzing the data sent to it, but he never answers the question of how the cortex makes a decision about what output it should generate. He talks about how it does generate output by performing the reverse function in the motor cortex (which it's not really doing per my theory above), but he doesn't answer the very simple question, of how it makes decisions.
In other words, just because the cortex has developed a large and complex "understanding" of the signals sent to it, how does it make the decision of how to reply to the current state of the environment? The cortex for example might understand that if you reach for the wall, you hand will touch it and and stop moving. But now the cortex could either make the hand form a fist, and pound the wall, or it could just make the body turn around and walk away. How does the cortex decide which to do? Jeff doesn't give us no clue about how this happens, yet it's the most fundamental question of brain function. How does the brain make decisions about what to do next?
The answer is that the brain is reinforcement learning machine. And most specific, the neocortex is the part of the brain that implements the reinforcement learning technology. It changes the behavior of the lower brain (its outputs), based on experience with pain and pleasure signals from the lower brain. Behaviors that have led to pleasure in the past, are strengthened, and behaviors which led to pain, are reduced.
This is the job of the neocortex. It monitors the output of the lower brain, and modifies it's behavior based on experience with pain and pleasure.
And to do this, the brain must understand the current context of the environment. It must have internal signals that represents it's best understanding of the state of the environment. It then weakens, and strengthens the connections between these state signals, and the behaviors they control through learning. And the point of all this invariant signal generation which the cortex is busy creating, is that those invariant signals represent the current state of the world. In other words, if the brain needs to learn to run away from dogs, and chase rabbits for food, it needs an internal signal that represents the fact that there is dog, or a rabbit, in the environment. It can then form connections from the dog signal, or the rabbit signal, to trigger the start of a run-away, or a run-towards, behavior. Which one it ends up triggering, is a function of reinforcement learning, not the invariant signal generation. So the point of all this stuff the neocortex is doing as talked about by Jeff, is simply to create the foundation needed, for the reinforcement learning system, which is the primary function of the neocortex.
Now, with all that said about reinforcement learning, I can return to the motor cortex. Why does the system need to analyze the output of the brain using the same system that it uses to analyze the sensory data we might ask. The answer can be seen in what it allows the brain to learn.
If you want the brain to learn a reaction to the dog, you need the sensory cortex to do it's work so it can recognize there is a dog to react to.
But, if you want the brain to generate a fixed output pattern, independent of what is happening in the environment (such as walking), it needs it to understand what it's doing (where it's legs are). This is what the motor cortex does. It forms invariant representations of the patterns that the brain is producing. So that when it produces an action such as stepping forward with the left leg, it will produce an invariant representation of the "step forward left" behavior. And it can then learn the recognize the step-forward-right behavior. And after seeing these together in a left/right behavior enough, it forms an invariant representation of "walking". All these invariant representations can then be wired (though reinforcement learning) to trigger the next change in behavior needed to maintain this constant left-right walking behavior. So, with the help of all the invariant representations learned by the motor cortex, the neocortex and learn to reinforce the sequence of behaviors needed to generate these constant output sequences which are independent of the environment.
At the same time of course, sensory data from the environment is being used to regulate walking to allow you to not bump into walls, and to allow you to reach for something and actually grab it. All that is coming from the sensory side of the cortex to help modify the outputs.
If the sensory side can then see a condition which triggers it to modify the output, to stop the leg motion. The motor cortex sees the leg has stopped moving, and the "walking" invariant signal turns off (it "knows" it is no longer walking), and when it turns off, the motor cortex then will no longer continue to loop producing the walking behavior. So the sensory cortex, just by a momentary modification of the the behavior, can cause the loops in the motor cortex to stop, or start, new behavior sequences.
So, in summary, the motor cortex is performing the _exact_ same algorithm (if you will) as the sensory cortex (not the inverse function that everyone seems to assume). It's just monitoring the outputs of the lower brain, where as the sensory cortex is monitoring the inputs to the lower brain. The entire cortex, through reinforcement learning, is learning what modifications to the outputs are needed. The actual output device of the brain is not the motor cortex, but something in the lower brain (maybe the basal ganglia? I'm a programmer, not a neuroscientist).
The motor cortex with it's modifications to the lower brain outputs allow the brain to produce any fixed sequence of behaviors needed (reaching, walking, the sequence needed to speak a a given work, etc). The sensory cortex, then modifies these behaviors, to kick the motor cortex into stopping, and starting, and switching to do a different sequence, or just simply, making small modifications to it's sequences (keeping balanced while walking, or directing the hand to the object with help of vision and touch sensory data while performing a standard reach sequence).
With this view, the motor cortex is in fact doing much of what Jeff thinks it's doing, but it's not doing it by passing motor commands down the chain. It's doing it by analyzing the raw signals sent to the mussels, and by learning what modifications to the output signals are needed in each different context by reinforcement learning.
_________________
Curt Welch |
|
| Back to top |
|
| Author |
Message |
skilesare
Joined: 15 Feb 2006
Posts: 18
Location: Houston, TX
03-03-06, 07:11 am |
| Post subject: |
|
|
Curt,
First off, I think your ideas are great and I'm going to have to re read a couple of times to get it all. I think it is unfair to say that Jeff ignores these things. I watched a presentation he gave where he specifically mentions that he left motor interactions out of the book to keep it readable. I'd love for him to go into more detail somewhere because I too am interested in what is going on.
The way that I have it in my head at the moment is that the specific predictions of the Motor Cortex project to the lower brain and can over ride the instincts that are there already. The cortex is doing its best to override the lower brain although it can't always do that.
As we try to build an intelligent machine we can get around some of this because we can radically simplify the lower brain. Simply wiring a set of specific predictions to the task of "output an A" and another set to output "B", etc. Eventually the machine learns to output patterns of A B B A(Instant Immortality  ). The fact of whether the A actually got output as expected can be monitored by the motor cortex as well as the Visual(or what ever we create) cortex confirming that it was output.
-SkileSare |
|
| Back to top |
|
| Author |
Message |
curtwelch
Joined: 02 Mar 2006
Posts: 14
Location: Vienna, VA
03-03-06, 04:51 pm |
| Post subject: |
|
|
| skilesare wrote: | Curt,
I think it is unfair to say that Jeff ignores these things. |
Yeah, all I know about Jeff is what he wrote in his book. I've never meet him, nor have I read any other work by him. So when I say he ignored something, I was really just saying it was ignored in the book. He might actually have all sorts of interesting thoughts on those subject (which to some extent I would expect).
For the most part, I think he and I are on the exact same page for all the most important issues about the brain and AI. I thought it was a great book and I'd like to see more like it.
_________________
Curt Welch |
|
| Back to top |
|
 |
Page 1 of 2 |
All times are GMT - 8 Hours
Goto page 1, 2 Next
|
|
You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot vote in polls in this forum
|
Powered by phpBB © 2001, 2002 phpBB Group
Please contact the board administrators if you have any questions regarding the OnIntelligence.org forums. |
| |
