Efference copycats

Came whiffling through the tulgey wood,

Pitfalls and traps
Cognition research at even a moderately advanced (post-graduate) level is fraught with pitfalls and traps for the unwary. The embarrassing truth is that mainstream scientists don't even know how the brain moves the simplest of muscles, never mind the more esoteric features of the brain like consciousness and self. This page 'sets the record straight' by providing the definitive answer about basic motor ennervation. The solution found by TGT is confirmed independently by Feldman, Markham and others.

There is a problem which is much less known than Libet's Paradox, but whose impact is just as serious. This problem is efference copy (EX*). Most researchers, even advanced ones, just read the standard texts on efference copy, and accept it 'as is' without really trying to understanding it. These are the efference copy-cats. The theory of EX matches the known observations (ie neural signals detected), to a fair degree of accuracy. The only problem is, the paradigm as they would have it narrated, simply doesn't make good sense.

'efference copy-cats' accept the concept 'as is', in spite of its lack of causative agency
For an explanation of a system to be acceptable and satisfactory, it must not violate one of the basic rules of the game. The most important of these rules concerns causes & effect. Firstly, causes MUST precede effects. Secondly- EVERYTHING must have a cause**. Consider two identical situations in which the same subject either (a) performs an action, or (b) doesn't perform it. In case (a) we correctly infer that it is the subject's motivated mental commands that cause the subject to act. The cause of the behavior is the subject's mind state at that time, an intentional*** mental state which has been generated by the aggregated influences of the subject's biological drives and their intensities.

The terminology in this topic is potentially confusing, since the two major names differ in only one letter. Afference and Efference are adjectives used to indicate the sensory input and motor output flows, respectively, of information in living organisms. Plain 'input' and 'output' are often used for machines, but it is customary, though admittedly somewhat old-fashioned, to use Afference and Efference to refer to the same channels in living creatures.

Risk of sensory overload from reafference? Wrong, wrong and wrong

Figure 33 depicts a simple organism interacting with its surroundings. The information it receives as sensory input can come from two sources only, namely self (called reafference) and non-self (called exafference). The basic idea behind the theory of EX is that each organism risks sensory overload from reafference. This is the case because the self is its own closest information producer. Exafferent sources are, ceteris paribis, generally of lower intensity since they are farther away. In human terms, when we shout, we don't listen.

In fact, far from being undesirable, reafference is needed. Feldman (who calls the UGP 'referent control') puts it thus...

"Note that in the case of referent control, posture-stabilizing mechanisms are not suppressed. When the common threshold position is shifted, it is the initial arm position that becomes deviated from the new stable EP. As a result, the same posture-stabilizing mechanisms, instead of resisting intentional motion, actively drive the arm to the final position, without any posture-movement problem. In other words, referent control allows the nervous system to convert posture-stabilizing to movement-producing mechanisms. It also automatically transfers stability from the initial position to the final one due to the shift in the torque-angle characteristic. Contrary to von Holst's assumption (Fig. 1), suppression of reafference would be detrimental to motor action: reafference is essential both for postural and movement stability. Nor is preprogramming of motor commands necessary in the referent control strategy: these commands emerge automatically, depending on the difference between the actual and the velocity-dependent threshold arm positions."

Two modes of behavior production - open loop and closed loop

There are two modes of behavior production - open loop and closed loop. Another term for open loop is feedforward. The organism produces the action without waiting for, or caring, about the consequences. Similarly, an alternative label for closed loop is feedback. The organism produces the action after waiting for, and taking account of, the sensory consequences.

Latency / error trade-off

In the open loop/ feedforward mode, the aim is to produce the action much more quickly, ie with less latency (delay), but at the cost of (potentially) less accuracy, or greater error. In the closed loop/ feedback mode, the aim is to produce the action much more accurately, ie with less error, but at the cost of (potentially) greater latency, or delay. Living creatures are at times able to achieve BOTH minimal error AND minimum latency, combining these modes while retaining state values between command invocations.

Timing of EX signal is wrong

So far, so good. Or so everyone believes. The flaw in this thread of reasoning is hard to see. There is little doubt that neural circuits are indeed activated in ways that correspond to EX. The error comes in the interpretation of this neural signal. The first problem we notice is that the timing of this signal is not quite as one would expect.

Interpret the EX signal as a command

This error in timing disappears if we interpret the so-called 'EX signal' as a command. Commands are one of the basic elements needed to satisfy the cybernetic drive states that generate all agency and new behavior. Commands are notably absent in behaviorist theory, which is based on the (unstated) assumption of a passive organism which reacts to environmental stimuli. This assumption (the everyday word for 'axiom') leads inexorably to the well known Behaviorist reflex paradigm of 'stimulus-precedes-response'. Nowhere in the Behaviorist weltanschung (worldview) are found the concepts of internal drive states, or feedback information flows.

Even when alone, all organisms experience some level of internal (cybernetic) drive-states

Organisms of all types, from the very simple to very complex ones like us, are just as often motivated / commanded to act by internal (cybernetic) drive-states (archetypically, hunger, thirst or sex), rather than be goaded to react by external inputs, such as the actions of other organisms. In fact, even with no external inputs, such as when the organism is alone in its surrounds, it is always experiencing internal drive state differentials at some level or another, be they lower level ones (hunger, thirst, satiety, discomfort) or those which operate at a higher level (appreciation, attraction, curiousity, boredom etc).

EX theory is wrong

A strong indication that the theoretical basis of EX is shaky comes, paradoxically, from its long history in cognitive science****. Scientists have been talking in a roundabout manner about EX without confirmation since the early 20th Century. There is a very good reason why EX theory will never achieve widespread scientific acceptance- it is just plain wrong.

The cybernetic theory upon which TGT is based claims that TGT is a special case of the Universal Governance Paradigm (UGP), in which feedforward commands and feedback control share the tasks of motion governance. First, the feedforward information flow delivers the sub-goal, if the applicable programming paradigm is declarative, or the sub-routine, in the case of procedural programming, to the target sub-system. Then the sub-system executes it. In the case of open loop actuation, nothing more need be considered. In the case of closed-loop actuation, there is/are human user/s and/or the environment whose inputs must be taken into account - see Figure 33(b).

Just like 'efference copy', drive-based (goal-state generated) commands are also feedforward - the saccade signal (+L) and the feedback correction (T-P) add together to form the total governance signal (called referent control by Feldman). This addition process occurs whether the system is a simple analog cybermaton, as in figure 1(a), or a much more complex organism, as in figures 33(a) and (b). It is simply the way that situation-specific tasks (which require sub-system to sub-system commands for effective coordination) are integrated into the declarative (feedback-controlled, integrity-based, identity-conserving) architecture of living things.

Phylogenetically, very few plants needed to move so a purely feedback-based basic I/O system (BIOS) was sufficient. The idea of the individual plant is not present in plant DNA- rather, a population will spread seeds, thus acquiring greener pastures by a process of 'moving through reproduction'. If you like, the seeds are the 'restless youth', the 'mobile form' of most terrestrial plants. When animals started to develop from plants they evolved the technique of motion, from which the concept of individual variation arose. Eventually, the time-scale built into the animal's motion control system grew shorter and shorter until consciousness was needed- ie the individual organism itself was the only sub-system with access to information needed to make crucial decisions fast enough. It could no longer rely on genetics or epigenetics to react to important situation change. Hence centralized nervous circuits first formed brains, from which consciousness (self-aware software) then evolved.

*the 'X' connotes 'Xerox', a name which is virtually synonymous with copying. The 'X' is used because there are already too many acronyms with 'C' in them - it is one of the most common letters in English.

**This rule should NOT be confused with the 'pop philosophy' idea that there must be an unbroken, endless and BEGINNINGLESS chain of causes and effects leading back into the distant past of infinity.

***the word 'intentional' is used here in the philosophical sense - referring to a state that is 'about' some object, process or situation in the world. The everyday meaning, that of an action which is deliberately intended, rather than casual, random, reactive or accidental, is still rather too close to the philosophical meaning for comfort.

****An example of the one-sided way that some researchers consider the issue is Buntain, C. (2012) Efference Copy - Did I do that?

The muscle maintains the limb it controls in the same position

Figure 34 (below) depicts the UGP as it applies to a simple bio-effector ( eg a somatic muscle, like a biceps, or triceps). The muscle by itself has one task- to maintain the limb it controls in one position (the position it is currently at)*. This is the thematic opposite of what most of us suppose the task of a muscle is, namely to move the limb. For the muscle to also move the limb, something extra is required- the neural signals that are part of its 'keep the same position' (KSP) feedback loop must be added to, or biased, DIRECTLY (ie as a part of a scalar addition) by top-down command signals**. A possibly helpful way to think of this function (remember, though, it is just an aide memoire) is that the KSP maintenance loop is the 'plant' part of its control system, while the top-down bias (TDB) signal is the 'animal' part (plants don't have to move, whereas animals do).

*The usual picture that is used - where a candle is burning the skin on the hand, and the 'snatch' reflex pulls the hand out of danger- actually represents the exceptional case. Most nerve impulses to somatic muscles serve to maintain position (individually) and posture (collectively). The REAL reason that the usual case is not depicted is because even the experts cannot TO THIS DAY agree on the cause and effect mechanism of somatic motion. This is DISINGENUOUS and intellectually dishonest.

**Although TDE/GOLEM Theory, or TGT, is still the only current 'cog' design which has ALL the problems around the control of motion solved, the idea of the TDB directly augmenting the homeostatic KSP part has been discovered by at least one other person - by Anatol Feldman in 2015/6. He has settled on 'Equilibrium Point Theory' for his version. Feldman's Equilibrium Point (EP) seems to be identical to the point marked 'TDB' in figure 34(a). Please note that TDB stands for Top-Down Bias, while TDE stands for Trilevel Differential Engine. Feldman, A.G. (2016) Active Sensing without Efference Copy : Referent Control of Perception. J. Neurophysiol 116: 960-976.

Identity-preserving feedback loop is biased by a top-down task-completion signal
There is a repeating pattern at work here- that of a basic identity-preserving feedback loop, biased by a supervening (ie top-down) task-completion signal. This pattern is none other than a variant on the TDE pattern itself, as in figure 34(b)(ii). It is even present in the way that visual perception is composed of two complementary streams*- a 'what are we looking at' stream, together with a 'where is it located in subjective space' stream. The human brain represents cybernetics on steroids, if you will. It is quite amazing how much nature has achieved with multiple variations on this same simple four-part canon.

*as demonstrated in the phenomenon of blindsight, where the subject reports no conscious perception of seeing an object, but intuits its spatial location better than 50:50

Cerebrum controls movements in space, while cerebellum controls bodily motions over time
We are now in a position to present a neuroanatomical overview. Hughlings-Jackson understood as far back as 1821 that the cerebrum controlled the organism's movements in space, while the cerebellum controlled its bodily motions over time. But this is precisely the same relationship that computers have with their software, in the following way: it is the cerebrum which constructs the subject's planned spatial trajectory, given the subject's intended destination, by a process of keyframe interpolation. The cerebellum then uses this spatial movement plan to select a matching set of interpolating steps guaranteed to produce the intended spatial trajectory. Figure 35 depicts this scenario.

The problem of sensorimotor representation was called 'common coding' by James in the 19th Century
There is a big, BIG problem lurking in the epistemological bushes, waiting to exploit the unwary hunter. William James understood it well- he called the problem topic 'common coding'. The problem is this - imagine that you were some kind of God. Given the task of designing animals, how would you code sensory inputs compared to motor outputs? Which one would be fundamental, and which one would be derived from the fundamental one. Many people, when asked this hypothetical, choose the motor channel as the fundamental one. After all, motor outputs are causative agents, and causes are more fundamental than effects, right?

Wrong, as it happens. It turns out that finding the right physiological law which governs the link between muscular force and limb movement is simply too computationally difficult. What is worse, IF you decide to do any force computing at all, you must compute both forward (muscle force --> limb movement) and inverse (movement --> force) functions to have a complete solution that covers all situations.

Isotonic means same force is applied at different positions.
This point is a rather subtle one, but the Russian anatomist Anatol Feldman makes the general principle clear with a simple experiment. A subject is asked to resist a force 'isotonically', with elbow resting on a platform, and forearm loaded with a weight and pulley. The term 'Isotonic' means that exactly the same force is used to resist the weight at a series of different positions. The gravitational pull in all situations is unchanged, due to the platform, and so are the muscular activation levels, if E(lectro)M(yo)G(ram) readings are taken.

There is no invariant which consistently matches forces to positions
Therefore, the very idea that there is, or indeed, EVER COULD BE a 'pre-programmed' relationship between muscular force and limb position is one that is terminally flawed. Imagine an astronaut floating in space, or a SCUBA diver floating underwater- these subjects control their limb positions and postures easily, without any extra training at all, yet the muscle forces involved must be radically different because of the move from gravity to no gravity. Clearly, the mind controls its limbs using SOME form of computation, which in turn requires the existence of a system of internal state variables. Groups of these variables must act in concert to enact reliable and repeatable principles. No one is debating that point. What we do know is that those functional groups CANNOT use motor forces as guiding parameters, because there is no invariant which consistently matches forces to positions.

But this seems to defy logic. How can we adjust limb positions and body postures if not by varying the output of the muscles involved? The answer is as subtle as it is semantic- of course we vary muscle pull to change positions and postures, but since we cannot use invariants to guide us, we cannot use feedforward, open-loop mathematical models to pre-compute the forces needed for each desired configuration. Instead, we have no option but must use the sole remaining possibility, closed-loop feedback control. Instead of using learned invariants to predict output, we use variation itself to compute it. In practice, a mixture of the two paradigms is best, (a) using feedforward (computational) methods at lower system levels for the speed advantage they confer (incurring the cost of low-level learning), and (b) feedback methods at higher system levels for the extra control advantage they confer, (avoiding the costs of learning at these higher, more individualistic and consciousness-based levels).

Desired positions are used as inputs, then required forces computed automatically by neural thresholds
Instead, all known animals code ONLY positions, displacements or velocities- ie EFFECTS, NOT CAUSES. So, you may wonder, how are the right force magnitudes computed? The answer is, they are not, because in a cybernetic system, they don't have to be. By means of a whole panoply of feedback loops, all the force levels in the various nested neural loops actually SET THEIR OWN VALUES FOR THEMSELVES. All the system does is provide desired positions in subjective space (or required joint angles in proprioceptive coordinates), and the force levels automatically follow*.

*This procedure is not unlike the Artificial Neural Network method training called backpropagation - no one knows (or cares) what the final network weights are, so long as the transfer function obtained is satisfactory.