Circuits, Time, and Representations:

Why should motor action develop?  Obviously motion offers many advantages for organism survival, both among single celled organisms and multi-cellular organisms. Movement is a key problem because it represents both opportunities and dangers for an organisms survival.  Even organisms which move in place such as corals must have processes which regulate when to move and when not to move.  These processes are representations.  They do not have to be conscious, but they are representational.  Once an organism itself is a moving object, the organism must develop more complex structures to regulate it's movement.  In the multicellular organisms, movement is regulated by cells.  

Below are diagrams for hypothetical cellular (stigmeric) address structures.  In Figure A we see a single connection from in input to an output.  A single cell connects some input to some output.  In an multi-cellular organism, it would be a single neuron connecting an input like a light sensing cell, or a touch or vibration sensing cell, to an output, a muscle cell for movement.   This simple structure connects an input to an output.  That is a basic representation, because without this connected network, the input and the output (action) have no relation.  In this simple circuit, inputs make outputs. 

It is important to understand how representation happens here.  Ordinarily, the cell is an agent and behaves stigmergicly.  Such that not all inputs will produce outputs.  However, below we treat the agent as a simple automata*.  When the cell receives an input it produces an output.  The representation is the structure itself  input-agent-output.  To connect some input to some output is the representation.  Why should that input and that output be connected.  There is no reason; the connection is arbitrary.  

The connection may serve some survival advantage, but that is something that occurs only after the fact.  The initial connections between inputs to outputs are explicitly arbitrary.  We cannot know some connection serves survival purposes until later, which means the survival purpose cannot back propagate a design process into the connection making.  The process of connecting inputs to outputs is representation making .  

[*As an automata we would describe the behavior of the cell as being driven by it's state.   Using state as a model to describe the agents/cells behavior is a reasonable thing to do.  However, later we will see that state itself is a description for more complex interactions and is not a thing itself.  ie. state is an illusion.]

This is the basic idea 

-figure A

In figure B, we see the same connection, but there are two cells in the chain from input to output.  The obvious affect of this arrangement is that it will take longer for input to generate an output action.  The cells or agents may do different things or may do the same things.  But as a feature of architecture among identical agents, passing through two agents takes longer than passing through one.

-figure B

In figure B1 we see one input, with two paths to the output.  What is the effect of this kind of structure?  Given that each cell signals when it receives a signal, this simple structure will produce two output signals from each input signal. The second output signal is caused simply by the time delay of passing through 3 cells.  Roughly, the longer path will take 3 times as long to generate an output as the single cell path.  This circuit also shows that one input generates two outputs.  The structure alone produces a new representation. 

For instance, assuming the cells all behave in a simple manner, the input will produced an initial output and another output that is followed by some regular interval

-figure B1

In figure B2, two inputs which will both generate the same output.  "Input A" has one cell that connects it to the output.  "input B" has 3 cells that connect to output.   Given the cells all fire at the same rate.  The "Input B" path will take longer to produce the same output.  But it only does this output when a separate input signal is received.  By using different numbers of cells in a chain, connected to different inputs, the organism can respond to different inputs, as different actions to take, over time.   Input B generates the output after an interval.  An input B generates a "waiting period" prior to output.

The other thing of note is that two different structures and inputs produce the same output.  This is an example of a representational difference between Input A and Input B.   This circuit implies that Input B should take longer to produce an output.  The same, but different.   Representationally modeled as  inputA ; output   inputB ; output    InputA and inputB are different.  And the representational process [inputB ; output] is conceals a more complex chain of representational action. 

-figure B2

Figure B3 shows a variation of the B2 model.  Except in this case, the output is only generated if cell A3 receives an input signal from both A2 and B1 in a certain time period.   Because these are stigmergic signals, or driven by systems biology processes in the cells, we should except some degradation of the signals over time.  Thus "Input A" will only generate an output if "Input B" occurs at the right time (or vice versa).  This is a timing mechanism.  It is also a model for a conditional circuit an "if and" statement. 

-figure B3

Figure B4 is an example of an either/or circuit.    So regardless of what input occurs, two outputs are induced.  The outputs are separated by time.  In this sort of circuit, other outputs can be put in place to tell which input induced the outputs.  

-Figure B4

Figure B5 shows a small change.  B1 signals when it receives some aggregate of a signal.  Perhaps input B signals twice or input A and B signal.  But no one input signal induces B1 to signal.  Whereas A1 induces a signal from any input it receives.  Here we see a difference in sensitivity.   Of course A1 and B1 likely have some other input (not shown) which should affect how likely the cell is to signal (excitatory or inhibitory effects). 

-Figure B5

In figure C we see how a single input repeats.  And because the T1-3 series of cells is the same as the A1-3 series, the interval between input and outputs is the same.  Figure C also shows a second output on the repeat. Perhaps this second output is an external signal of the repeat.    This kind of circuit could be made to be perpetual, as in a clock.  Where output A is the movement of the clock, and output B is the "tick" of the clock.  

Notice that "repetition" is an artifact of this structure.  Where in a physical/material universe is repetition?  it is nowhere; repetition is an idea.  We cannot actually say that this circuit shows repetition... although we do, because repetition is a representation that appears to be instantiated by this structure.  

A slightly more complex circuit could be a counting circuit.   And if outputs connected to inputs in some structure, that structure could "stop" the  counting.  If we look at how cells (agents/actors) interact with each other, we will find that we cannot untangle one representation from another.  Looking at the circuitry, and the signaling, we only see the structures and the messages.  We do not see the representations the structures and messages instantiate.  

a corollary example is looking at a portion of an MP3 file and a JPEG file.  They both contain binary data.  It's the representational context that treats the binary data differently.  

-figure C

Figure C1, shows

In a longer circuit, the messages from cells would not have to be concurrent.  It is the association of an output signal and an input signal that matters.  This is a simple structure to show that representation of inputs to outputs can be mediated by some structure.    If M1 retains some state from receiving a signal from A1, and then at some point in time receives an input signal, then M1 would send on it's own signal, and the duration is irrelevant.  M1 also can show that some action to output is associated to some input.   That is, output (actions) should produce input (feelings).  

For instance, if the signal to M1 from either A1 or InputA are equivalent.  Whenever A1 signals to OutputA, it also signals M1.  If M1 is the sensation of InputA,  M1 cannot distinguish between InputA or A1 - thus A1 "feels like" inputA .  This circuit shows the recognition of that relationship.   This kind of structure seems necessary for movement.  We must feel not only inputs to move, but as we move, the movement itself generates inputs.   A modern corollary would the phantom buzzing of a cellphone in pocket. 

-Figure C1

In figure C2,  we see the I/O circuit joined to a predictive circuit.  The  key relationship shown here is the function of M3.  Does it excite M2 or inhibit M2?  Input C signals M3 that expresses some (external) condition.  M3 takes the signal of an output action (M1) and  joins it with the signal data (or lack thereof) from Input C and produces some excitatory or inhibitory marker to M2.  M3 acts to modify the signaling of M2.

Input C may be an actual input, or it may be a collection of circuits.  From each cells point of view, the received signals are all input signals.   Thus in a more complex circuit M3 may receive signals or the absence of signals that mean "too early"  "too late" or a signal that means "one time".

We should expect to see lots of inputs to an M2 from various modifying cells that constantly time the signal firing of M2.

Note that A2 sends a signal to M2 and in some initial state, M2 will or will not fire.  M3 adjusts the condition of M2, but once adjusted, that is the new state of M2 and it will send a signal or not.  It may be that M3 will adjust M2 and that adjustment has a gradient effect so that M2 will send a signal, or will be initially insufficiently stimulated to send a signal.  Or it may be that M3 will modify M2 so that it will send a signal when it receives the next signal from A2... or it may inhibit M2's signal sending capacity entirely.  Whatever the condition is, M3 can initiate or inhibit a current signal or it can set M2 for it's behavior on receiving the next signal from A1. this functionality correlates with the actor model of computation.

-Figure C2

To be clear, inputs and outputs may simply be other cells, or long chains of cells.   Adjusting a cells signaling with excitatory or inhibitory signals from other cells implies that those other cells carry some value, some signal about what the affected cell should be doing. And this is one example of where a representation comes into being.  It is helpful to see these network structures as outcome oriented developments that are arbitrarily created and then kept around because they survive a natural, even random process of network trimming. [Daniel Bushey, Giulio Tononi, Chiara Cirelli. Sleep and Synaptic Homeostasis: Structural Evidence in Drosophila. Science, 2011; 332 (6037): 1576-1581 DOI:10.1126/science.1202839]  []

Until figure C2 we have only one kind of signal.  With C2 we add 3 more.  These alterations affect the power of the circuitry to represent relations of inputs to outputs, and to represent relations of inputs and outputs over time.  They obviously can represent inputs and outputs over space.  

In a computer system, the problem with cells is how they connect in a virtual space.  How do cells "find each other" without a space to explore?  

This exploratory activity is driven by internal molecular processes of the cell itself.  It is a side-effect of the systems biology of the cell.  The cells can signal each other locally and thus "reach out" to each other.   But is not necessary to think of cells signaling each other so much as there is an unequal distribution of "signal" molecules in one area over another.  This distribution difference directs (synaptic) growth not unlike how molecules affect the rotation of flagella or the development of microtuble structures to "grow" in certain directions.  

The cells in a tissue act according to the signals they receive in terms of motion and shape change, because they occupy a space. Signals exist in a space and the cells expand and function in that space according to those signals. **

We have here looked at how cells, connected to other cells, signal each other and thus embody representations of inputs and outputs.  But the growth of cells and the connection of cells to each other can also form a structure of representation.  

In C2 we see how the signaling of cells is modified.  The growth of cell connections can also be modified.  Cells receiving an input from a cell can respond not by signaling another cell, but by growing connections to other cells.  This growth function can be excited or inhibited by signals from connected cells.  So that one cell may reach out to other cells continually looking for the "right" connection.  That "right" connection gets finalized when the reaching cells stops receiving excitatory growth signals from some other connected cell(s).  

Thus growth of a cell may be affected by receiving signals from nearby cells to connect with them (a local signal) or by signals at it's base that it has not yet made connections to the right cells to stop growing.   In all likelihood there is an interplay of signaling at both ends, and that novel learning represents both a local condition and an impetus to make connections.  This impetus though is a result of structures of cells producing that impetus.  Which means there is some relation of inputs and outputs that impel the learning to take place.  There is also cell excitation of the production of growth signals by local cells.

Thus "looking for a connection" is a kind of structure that is signaled back and forth in the development of connections between cells.  

Do we see this sort of growth phenomena in representation?  Yes, we do.  But it's oddly not a feature of representation.  We do not grasp the representations we do not see or comprehend, only the ones we do understand.  And thus comprehension, and apprehension comes to us spontaneously, from below our representational space.  Such a model of connection is one that would lie beneath representation, until a connection is made.  That novel connection is the new representation, the new learning, the apprehended idea, the novel realization, the imaginative creation.  

When we experience novel learning, we often have an "I get it" experience which feels like a "confirmation signal".   Confirmation signals should be signals that tend to restrain connection growth.  This is the model suggested here. Notice that the feeling of "I get it" doesn't come from the representation learned itself, but from some representation about the representation learned.  That is, a second circuit confirms learning has taken place and then dampens the grow new connections signals of other neurons.  

The modification of existing connections, as we see in figure C2 is a kind learning.  But it is the adjustment of an existing connection and not the creation of new connections between inputs and outputs.  Most learning is going to involve a range of growth of new connections and cells, and modification of cell connections to achieve optimums.    And that is the primary distinction.  Growth of connections, the development of representations, is different from the optimization of signaling once a connection exists.  

**(Cancer cells grow and expand without regard to the space they occupy and the normal signals they receive.  In a basic way, cancer cells become abnormally stigmergic and respond only to their internal function.  We can think of cancer cells more as machines and less as stigmergic actors. The cancer cells stop responding to "messages" from other cells (no change of actor state produced by normal signals and hence do not respond to signals that would initiate apoptosis.)

Figure C3 shows us how the signaling may work.  In reality, all the connections between inputs and outputs and cellular circuits must be created.  The M2 connection to A1, being "grown" here, implies that the "Output Action Signal" from M1 to M3 has already been grown.   To be clear, it isn't that M2 is looking for A1, as labeled here.  But rather M2 will "stop" growing when the signal from M1 from from inputC is received by M3 inducing M3 to stop sending an excitatory "grow" signal to M2.

As we see in stigmergic behavior in all agents, it is signaling which induces the side-effect of behavior.  As in ants looking for food, the chemical trail is what makes the circuit for other ants to follow.  It could be M2 creates a multicellular connection to A1.  This may not be optimal, and because of timing (decay of the signal as in ant trails) or because the timing of signals "means" something not quite right, the signal from M3 to initiate growth may fluctuate; the growth signal itself can be optimized.  Growth signal optimization would give us source directed growth.  But this source directed growth will look very like bacterial movements because the cell itself does not have representations of space or objects or directions.  

How neurons form synaptic connections, and drop synaptic connections can be completely, or nearly completely random.  But the impetus to continue to form and drop connections is the "learning imperative".  The organism must have some representational reason to learn and to stop learning.  And this representational reason most likely exist as neural structures that continually signal growth activity until those structures themselves are satisfied (recognized learning has occurred).  But at the cellular level, where synaptic connections are made, the cells are responding in typical systems biology fashion where the molecular networks in the cell produce side-effect behavior to grow synapses in arbitrary ways that are continually being modified by the steady state behavior the molecular (systems biology) networks. 

-Figure C3

A suggested set of signals:

generic signal
grow connections
spawn/divide a new cell/agent

likely random processes:
? make connections, drop connections
? weaken, strengthen connections

Figure D begins to show what representation "as awareness" look likes.  Figure D is an extension of C1 showing an association between an input and an output(action).  The state of that association (C1) is reflected in the signals from the R cells to the circuit cloud.  The simple connections from R(1,2,3) REFLECT the state of the C1 circuit between an input and an output.  Moreover, the cells are structured to signal information AS IT HAPPENS.   

Notice that there is a delay from Input A, A1, or M1 to the circuit-cloud, equivalent to the time it takes to pass through each reflecting R cell.  Also, R1 and R3 are duplicate signalers for A1 and M1 respectively.  Why should we need to duplicate these cells?  Because we would expect that R1 and R3 are connecting to cells in another location away from the output terminals of A1 and M1.***

For example, the R cells reflecting the state of some output to an input, should delay the "recognition" or "being aware" of the input to ouput association and state.  Thus a suggested neurological delay of "intention" would be a natural outcome when the "intentional representations" are done across some group of neurological structures and then signaled to some other group of structures for action or this process may be concurrent, where "cognition"  is concurrent to intention to act.  These features of representation in terms of cognition and cognition would show up as delays.  The structure of circuitry and the delay in signal processing would be the simple explanation for Libet's experiments showing a time discrepancy between conscious awareness(attention) and action.  Meaning that the consciousness and action are functionally concurrent from a representational standpoint, but show a physical differential because of the circuitry of the representational substrate.  

A variation on this D figure where the Input A and A1 represent intentional signals (versus an external input), and recognition of an intentional signal (output to M1)  are reflected elsewhere, like in attention (self directed representation) structures of the circuit clould.   Explained more simply: the delay of signaling brought by the R cells may be an example of the delay demonstrated by Libet and others, where the C1 structures (input A, and A1) represent a start of intentional action and the R structures are the reflected recognition of that action. 

Figure D, which we could call C1+R evokes meaning in the signals it conveys via the R cells.  But what do the R cells convey?  Even with a simple signal from the R cells, it's that the R cells are different because they convey the state of C1, as if C1 is a thing itself.  The R cells reflecting the state of C1 is REPRESENTATIONAL.   The R cells do not explicitly reflect C1's state.  They reflect it "associationally" or as a representation.    eg.  state of C1; R2   This connection of (state of C1) ; R2 is the signal R2 makes into the circuit cloud, what else could the signal of R2 be?  

In reality, we have lots of interconnecting cells, and the structures between them all are each slightly different (incredible complex connections and variety).  It seems clear that with greater structural variety and signal pathways, the greater nuance and diversity we see in experience and representation.  We cannot tell what  representations some set of circuits manifest, although we may be able to guess.  What matters is that the circuit structures and the signaling form a simulation.  That simulation is a representation of relations of inputs to outputs.  It is this whole simulation that is experience.  

Figure D

***  In biology this duplication may also serve a redundant survival function.  For instance, if A1 is lost, R1 could develop connections to substitute for A1's signaling role.   

This is the basic model of representation as cells (agents) signaling other cells as shown in these circuit diagrams.  Representations do not "emerge" from the structure and signaling of cells.  Rather, representations ARE the cell structures.  And the signaling of cells is, loosely, the awareness of those representations.  If the whole system is distorted, either by changing signaling or by damage, or by altering the kinds of signals being made, or the frequency or syncopation of signaling, there should be some change of "awareness", and some changes in representation, in experience.    

Explicitly though, this is not emergence.  When inputs are connected to outputs and when cells connect to other cells, that is MAKING.    Cells connecting to other cells is the making of a representation.  Cells signaling to cells via those connections is the "having" of representation.  Representations only make sense in the context of the representational simulation (experience) as instantiated by the cellular stigmergic system.  That is, representations only make sense in the situation of a representational environment, where representations are the products of structures created via stigmergy.  

In this context we may think of representations as emergent, except that we always make recourse to the efficacy of the emergent property or to some other representational fact that is not a fact of the whole cellular system.  The whole cellular system is, explicitly not representational.  Emergence is a representational way to describe what is not an emergent property at all, but is a stigmergic phenomena.  the cellular stigmergic system instantiates representations and representational action.   The representations do not emerge because the representations exist independently of their instantiation. 

A note.  cell connectivity and dis-connectivity should be driven by the local molecular environment of the cell.  Which means altering the local environment would affect whatever learning the whole organism is engaged in.  And this is what we see with drug use.  Cannaboids increase the likelihood of the loss or impairment of connections.  which would be an example of how structure itself is affected by the underlying molecular interactions.  
[Acute Cannabinoids Impair Working Memory through Astroglial CB1 Receptor Modulation of Hippocampal LTD Jing Han+others 2012 ]

B note:  Sleep corresponds to reductions of the connectome.  synaptic connections must be created to form representations but synaptic connections must also be uncreated to forget.  The fact of forgetting is crucial to any kind of representational simulation development. Representation making requires a process of representation unmaking.   

For instance, Initial experiences are often seen as unique, but those same experiences over time must not continue to be unique, but must become standard, similar, even identical to operate in a simulation. The initial development of connections as demonstrated in sleep and synaptic homeostasis study [ [Daniel Bushey, Giulio Tononi, Chiara Cirelli. Sleep and Synaptic Homeostasis: Structural Evidence in Drosophila. Science, 2011; 332 (6037): 1576-1581 DOI:10.1126/science.1202839] shows us when this forgetting process happens.  

If synaptic connections were reduced during wakefulness, that would have obvious deleterious effects for moving organisms.  A predator may forget why it was chasing prey, or the prey may forget why it was fleeing a predator.  A bee may forget it's route from the hive and never return.  But if predators and prey and bees do not forget about prior routes, about prior chases, then these experiences would intrude into current experience, so that a bee spontaneously starts following yesterday's route home or a rabbit spontaneously starts running away from yesterday's fox.  thus the process of unmaking connections will inevitably be done in a way that enhances the survival of an organism.  Spontaneous forgetting while moving around in an environment is such an obvious source of danger for an organism that the unmaking of synaptic connections must occur at a time when the danger of forgetting is very low.  

The period of synaptic connection reduction, of unmaking of synaptic connections, and thus of  representations, is sleep.