The Main Types of Information Machines

Two Types of Information Machines#

A categorization is without any explanation for the rational, given, though it might seem “obvious”. Namely, there are mainly two types of information machines,

• Electronic calculation and reasoning machine - that which process information, belonging in the field of informatics rather than of cybernetics.
• Feedback and self-regulation machines, that belong properly to cybernetics.

Obviously Ruyer is trying to say that processing is different from transmitting, though in the sense of cybernetic transmission is rather strange, in that it feeds back and self regulate. Bu none of these are explained, so we shall defer.

The electronic variety of the first type of machines - calculating machines - while similar to those cogwheel machines, Ruyer implicitly claims that, are completely different in terms of their designing principle. These include ENIAC, BINAC, EDVAC, etc. They operate not by mechanical movements - not analog - but by opening or closing circuits - digital. Then he sets out to rephrase the characteristics of digital machines. And he contrasts the functioning of digital circuits with that of physiological “circuits”(it's really interesting that nowadays humans are metaphors of computers), by e.g. citing William James. Interesting move, but since Peirce also was closewith and corresponded extensively with William James, not really that interesting, anyways:

When we think with “and,” “or,” “if . . . then,” “but . . . ,” it is very likely that we do not simply put nervous circuits in series or in parallel. We have, 
as William James says, the “feeling of but” or the “feeling of because.” 

And he occasionally makes the move of comparing what is nowadays called “unconscious” (but caution since, e.g. Cormac McCarthy never calls that thing, that which is somatic, unconscious) to what machines do:

If I brake suddenly in front of an obstacle, my fear, which often arrives late, has nothing to do with the assembly that allowed the prompt and appropriate response.

But at the same time, he mentions the atmosphere of meaning, which is vaguely similar to what I'm trying to express with the word “semantic sphere” - those without the need of this atmosphere of meaning can be replaced effectively by an automatic assembly. He mentions inducing machine and says that this machine (that is at that time) at the early stage of of planning seems to be an enriching operation, unlike deduction. But the fact nowadays is that induction (and interpolation, extrapolation, etc.) can be quite effectively solved by Bayesian learning programs, but deduction need always to be supervised, since the machines do not know what to deduce. However the important thing here is that what's under the investigation of Ruyer is the class of machines which are made to do induction by exploiting the mechanism of feedback, “education of correlates”. By arguing that induction can in principle be done, he claims that cyberneticians “may have some hope of making inventing machines” - as in Gestalt Theory, invention represent neither a transcendent action nor a process of pure trial and error without direction but a spontaneous reorganization of the perceptual field, carried out according to the principle of least action or of least tension - in short, optimization. Feedback is a way to achieve optimization, and optimization is always done by adhering to the principle of least action.

Optimization = Entropy Increasing#

Here's an excellent observation:

We know that in macroscopic physics, the principle of least action has the closest relation to the principle of evolution toward maximum entropy. To bring invention, creation, or the restoration of information back to the principle of least action, is in fact to bring information back to what, according to cybernetics itself, is precisely its opposite, entropy.

“It is the increase in entropy, the march toward symmetry, rounding, and equilibrium” - and nowadays people will say that it is finding the local equilibrium. Optimization is done in such a way that the equilibria themselves are pre-defined by conscious beings. By local equilibrium, what I'm saying is that a local minimum of a potential:

Occam's razor principle as “proved” by the theoy of Kolmogorov complexity, versus, “economy of thought”, in the sense of finding a local equilibria - by maximizing (locally) some entropy.

Then straightforwardly, Bayesian approach is refuted:

The test of completing words in a sentence only seems to resemble number completion or the extension of a curve. The machine for detecting errors (or pseudo-errors) in a digital series is in reality only a machine for detecting singular points, which may or may not have the character of an error, depending on the meaning given to it by a conscious observer. On the contrary, it can be an interesting “residue,” in the sense that logicians of induction understand the word.

But Bayesians may still claim that what is thought of as meaning is in fact just what “emerges after a certain equilibrium is approached”.

Since feedback mechanism and self-regulation are precisely those used by inducing machines, I don't see the point of writing the next two sections Ruyer wrote. By the way, plants and those “unconscious” (maybe we should divide the unconscious further into somatic one and a nonsomatic one) beings while are self-regulating does seem to operate by means of feedback mechanism, and it is defined, by Ruyer, as

a cyclical functioning with a regulating loop through which a current of information flows, automatically compared to an “ideal”.

Interested Sensitivity and Perception#

What is now called “pattern recognition” is, in the language of cyberneticians, the recognition of forms as universals. The standard problem in pattern recognition appears, but as I've expected, Ruyer's discussion isn't really on the point.

It's essentially a problem of metaphysics, for now. The class of problems stated by Ruyer, that is, say

When I recognize a plate, even when viewed from the side, the plate remains an elongated ellipse or trapezium, although I know it is round. According to the hypothesis, the final consciousness—after the intervention of the group scanning mechanisms and the resonance with the cerebral model—should be that of a round plate. 

What Ruyer is pointing out is, rather, a matter of perspective, or point of view: a plate is certainly not a round one, it is only round from a certain perspective. When we see a plate and recognize it as round, we do not transform it, we simply grasp it as it is seen, but at the same time we know it is round. 

We can admit that the emotional atmosphere, or even the atmosphere of meaning, is not always essential in behavior. We can also admit that even the psychological interpretations of gnosis or agnosia are sometimes superficial. But in the perception of the plate, triangle, or hutch, it is difficult to reject an immediate intuition. We see the particular triangle, at the same time and indissolubly, as triangle and as particular.

The point, I think, should be that, moreover, we expect it to be round, since, well, plates are round. And only in a particular semantic sphere an object can be seen as a plate. Collapse of final causality with efficient causality is not the only thing that cyberneticians commit, but they, essentially, want to eliminate all forms of final causality.