Brain and Perception

Holonomy and Structure in Figural Processing

Karl H. Pribram

• Publisher: Psychology Press
• Year: 2013
• ISBN:
  9781135832919
• Edition: 1
• Topic: Psicología
• Subtopic: Historia y teoría en psicología

Presented as a series of lectures, this important volume achieves four major goals: 1) It integrates the results of the author's research as applied to pattern perception -- reviewing current brain research and showing how several lines of inquiry have been converging to produce a paradigm shift in our understanding of the neural basis of figural perception. 2) It updates the holographic hypothesis of brain function in perception. 3) It emphasizes the fact that both distributed (holistic) and localized (structural) processes characterize brain function. 4) It portrays a neural systems analysis of brain organization in figural perception by computational models -- describing processing in terms of formalisms found useful in ordering data in 20th-century physical and engineering sciences. The lectures are divided into three parts: a Prolegomenon outlining a theoretical framework for the presentation; Part I dealing with the configural aspects of perception; and Part II presenting its cognitive aspects. The appendices were developed in a collaborative effort by the author, Kunio Yasue, and Mari Jibu (both of Notre Dame Seishin University of Okayama, Japan).

I	Configural Aspects

Objects are not fleeting and fugitive appearances [images] because they are not only groups of sensations, but groups cemented by a constant bond. It is this bond alone, which is the object in itself and this bond is a relation.

—(Poincare, 1905b)

3	Transformational Realism: The Optic Array, the Optical Image and the Retinal Process

Instead of postulating that the brain constructs information from the input of a sensory nerve, we can suppose that the centers of the nervous system, including the brain, resonate to information. (Gibson, p.267)

Introduction: The What of Perceptual Processing

This lecture concerns the initial sensory mechanism, which in the eye consists of the optics of pupil and lens and the receptor processes of the retina. In vision, I hope to show that making a distinction between the optic array, optical images, or flows, and retinal process (and specifying the transformations involved), resolves several hitherto intractible issues. These issues are: (a) the grain problem, that is, determining the scale of the origins of sensory configurations; (b) the existence or nonexistence of a retinal image; and (c) the dimensionality of the initial sensory process.

Interpretations regarding this initial level of processing are critical to understanding what it is that is being perceived. As a result, the interpretations have been the subject of considerable controversy. Of primary concern, has been “the grain problem,” because the input to receptors is described at two very different scales: One of these scales is microphysical and based on analysis by physical instruments, the other is macropsychological and based on reports of introspection. Eddington described this issue in terms of two tables: one is composed of atoms and subatomic particles; the other is used to serve a meal. Psychophysics describes the relation between these modes but does not deal with the fact that physical descriptions are made in microterms of patterns of energy, entropy, and information (probability amplitude-modulated frequencies), whereas reports of introspection yield macroconfigurations such as tables and chairs. The gap in scale or “grain” of the psychophysical pair is immense and has hitherto seemed insurmountable.

The holonomic brain theory addresses the grain problem by specifying the transformations that occur between processing stages in the same terms as those used in physical measurement (energy, entropy, information) without neglecting the configurai properties of the percepts being described. The key to achieving such specification lies in the transformational nature of the Fourier and Gabor relationships enfolding and unfolding configuration into and out of the spectral domain.

As suspected by Gibson in the epigram introducing this lecture, the grain problem has been addressed in two ways: Either (a) percepts are constructed from simpler elements or (b) regions in the brain directly resonate to configurations already present in the input to the senses. The resolution to the problem presented in the holonomic brain theory is orthogonal to these views and therefore somewhat unexpected: The initial sensory process is neither constructive nor direct. As detailed shortly, sensory processes are transformational. The initial processing stage is not a construction from elements but consists of reordering a distributed, enfolded array that forms the content of the sensory input. Neither are the initial levels of processing—the image and object-form levels—direct. Rather, they involve transformation. However, construction becomes a part of the perceptual process as a result of cognitive operations. These operations inform the image by preprocessing the input channels via the brain’s corticofugal pathways.

For vision, the controversy regarding grain has centered on the existence of a “retinal image” and other such representations in more centrally located brain regions. Resolution of the controversy comes in the holonomic brain theory by unpacking the concept “retinal image.” This is done by showing that an optical image is formed from the optic array by the pupil/lens system, whereas, a retinal process performs operations much more complex than those that register an image in ordinary photography.

Optic array, optical image, and retinal receptor process all differ in their configuration. Each processing step transforms, changes the form of what is being processed. How then can one ascertain what is being processed, the critical features that characterize a percept? The answer to this question is: By ascertaining that which remains invariant across processing stages. Such invariants are discovered by a procedure that is an extension of that used in psychophysics: Constraints are placed on processing by manipulating a situation that can be described in the computational language of physics or engineering. The limits within which the subjective report of the ensuing perceptual experience remains unchanged provide an indication of the invariant features of that situation.

Such experiments have shown that relations between perceived parts of images and object-forms are the grist of the perceptual mill. The classical experiments by Stratton (1896, 1897) in which he continuously wore lenses that inverted the optical image, only to find that within a week everything was once more experienced as right side up, demonstrated this beyond doubt. The experiments by Richard Held (1968) and Ivo Kohler (1964) showed that moving about was critical to this adaptation of phenomenal experience.

Given that relations are basic to what is being processed, the next question becomes, “Relations among what?” Mach concluded, on the basis of his observations (see review by Ratliff, 1965) that what is being processed are the spatial and temporal relations among the magnitudes of infinitesimally small points of radiant electromagnetic energy, the spatial and temporal derivatives of luminance. As described later in this lecture, a more comprehensive but related formulation of this approach has been used by Sejnowsky and Lehky (1987) to compute configuration using “the second derivative of the tangent vector to the surface along a line” (p. 18).

The Optic Array

What must be taken into account when describing the initial stages of processing can be illustrated by examples taken from research on visual perception. An ecological view of perception was developed by James Gibson on the basis of his extensive research, which clearly showed that visual form perception is initially three-dimensional (or even four, that is, space and time dimensional) and not composed of elementary lines and two-dimensional planes. This led Gibson to view perception as direct or immediate rather than constructional (with the exception of the cognitive influences on perception).

Although the holonomic brain theory supports Gibson’s intuitions, the tenets of the theory were not acceptable to Gibson. Both he and his colleagues (as well as many others) suspected that its constructional aspects implied construction from elements (elementarism). These suspicions become groundless once it is understood that the initial perceptual processes are transformational and not elementaristic. In accord with Gibson’s view, the constructional aspects of perception are top-down cognitive operations, such as those involved in learning: for example, the progressive differentiation of the sensory input (see e.g., Gibson & Gibson, 1955, and Lectures 7–10).

Some philosophers (e.g., Putnam, 1973) and scientists, such as Gibson and Turvey, are puzzled by the fact that whereas “perceivers are primarily sensitive to higher order variables of stimulation, light which lacks macroscopic structure provides no information to a visual system.” (Fowler and Turvey, 1982). As noted earlier, this is known in philosophy as “the grain problem.” The choice must clearly be made for a perceptual system responsive to higher order variables. These are, in fact, provided by reflected and diffracted patterns of radiant energy. Gibson and Turvey as well as most scientists fail to distinguish between radiant energy and “light” that is produced when patterns of such energy stimulate appropriate receptors. Nor do Gibson or Turvey realize that reflected radiant energy provides such patterns. This is because patterns are not discernable in the ambient diffracted “scatter” produced by reflection from objects.

A simple demonstration suggested by C. A. Taylor (1978) in a small volume on Imaging, prepared for instruction at the 6th form level in England, illustrates what is involved. Taylor placed a slide in a slide projector and projected it after removing the lens. “Technically the pattern on the screen with no lens in the projector is called a hologram. … the term simply means that each point on the screen.… is receiving information from every point on the object” (p. 2).

Taylor went on to demonstrate that indeed each section of the screen receives information from every point on the object (the slide) by performing the following experiment: The projector is placed a few feet from the screen and then a converging (magnifying) lens is used to form a reduced image of the slide on the screen. The reduced image of the whole slide can be produced with the lens at any position within the patch of light, demonstrating that all sections of the illuminated screen contain information about all points on the slide (Fig 3.1).

Fig. 3.1. a) The hologram relationship: each point of the slide contributes scattered light to each point of the screen, b) The hologram relationship: wherever the lens is placed a complete image of the slide is produced. From: Taylor, C. A. (1978). Images: A unified view of diffraction and image formation with all kinds of radiation. New York: Wykeham Publications (London) Ltd.

This demonstration makes clear that incident radiation becomes diffracted, “scattered,” by an object—scatter being defined as an organized bouncing of incident radiation off the object so that the organization of the radiation becomes distributed. It takes a lens to transform this organized scatter into what we are able to recognize as a (space-time) image. Taylor rightly pointed out that “the simple ray diagrams of geometrical optics hide a great deal of the complexity of this operation” (p. 3). In short, image formation depends on the recombination of incident radiation “scattered” by reflection and diffraction from surfaces and objects.

Thus, the dilemma posed by ecological optics is not really a dilemma after all. With respect to vision, reflected patterns of radiant energy that enter the pupil appear to lack macroscopic structure, but appearances are deceiving: The structure is hidden because it becomes diffracted: enfolded and spread, distributed, into a form displaying “nonlocality.” Just as in a hologram, or in the placement of a radio receiver tuned to a broadcast, every location potentially contains the essential information necessary to reconstruct the macroscopic structure. Pupil and lens then unfold this potential into a recognizable optical image that interfaces with the retinal process (Fig 3.2).

Fig. 3.2. A simplified drawing of the path of light rays through the eye’s optical system, ignoring refraction by the lens. Note that due to reflection and diffraction, A and B would not represent isolated points on a figure but nodes in a distributed array. From: DeValois, R. L. and DeValois, K. K. (1988). Spatial Vision. New York: Oxford University Press.

As with Gabor’s insights that led to the invention of optical information-processing systems such as holography, the holonomic brain theory holds that ambient patterns of energy that appear as “scatter” actually enfold and distribute macroscopic structure into a new order or organization (Fourier transforms entail sets of operations called “point spread functions”). This order serves as a potential to be re-transformed into a space-time image by the optics of the eye. Configurations hidden by the distributed nonlocal reordering of the input (as in a hologram) can be unfolded by the pupil and lens performing an inverse transformation.

The Optical Image and Optical Flow

How do these arguments regarding image formulation relate to the issue of the existance of a retinal image? Gibson (1979) claimed that no such representation of the object world need be involved in the perceptual process. The holonomic brain theory on the other hand, because it necessarily incorporates all the stages of processing as performed by the organism, begins with the observation that, in the stationary eye, the optics do in fact create recognizable images of objects. The fact that this is so tells us something about the system and cannot just be dismissed.

There is a sense in which even Gibson would admit such images:

If we could think of an image in the derived sense as a complex of relations, as the invariant structure of an arrangement, in short as information, there would be no great harm in extending the original meaning of the term. But this is hard to do, for it carries too much weight of history. It is better not to try. It would surely be false to say that there is a phonograph record in the ear, and the same error tempts us when we say that there is an image in the eye.

Thus, in ecological optics, the optic array, the optical image (or flow), and the retinal process become confounded. Gibson argued rightly that the retinal process differs from a photograph and that the optic array external to the eye conveys the complexity necessary for perception to occur. However, he ignored the transformational steps that characterize the optics of the eye and the distinction between the resulting optical image and the retinal process.

A realist stance toward both the momentary optical (moving, informative) image and the optic array identifies their difference. Thus, as noted, the optic array is considered to consist of a spectral manifold (an enfolded, distributed form of radiant energy), which is a transform of the patterns defining objects. The optic array thus resembles a holographic film—or a cross section of the waves carrying radio and television programs that have been broadcast (cast-broadly). By contrast the optical image is a three or more dimensional flow in the unfolded, space-time sense. To an observer of an excised eye it appears much as does a photograph taken by a camera; under natural conditions, flow patterns constitute an occulocentric space.

A further argument has been made to the effect that although the eye is a camera that focuses an image on the retina, it is we, with our visual apparatus who “see” that image, that we cannot know the actual design that the optical apparatus projects onto the retina. But the same considerations hold for a photographic print. Are we to deny the reality of the photograph as we perceive it to be? The holonomic brain theory considers optical images (flow) as real but distinguishes it clearly from the retinal process that transduces that flow by virtue of its neurochemical and neuroelectric properties.

The medium that operates on (records) the optical flow is different from that which records a photograph. The photographic medium is sensitive only to the intensities (amplitudes) of energy at any given point. By contrast, as Selig Hecht (1934) and others (Sakitt, 1972) have shown, retinal receptors are sensitive to single quanta of electromagnetic energy. Therefore, the quantal aspects of incident radiation must be considered in any representation of the receptor process. Mathematically, the holonomic brain theory therefore represents the receptor process not only by a real number representing the amplitude of the energy at a particular location but by a complex number that takes into account attributes of the quantum, that is, a number with direction. Thus, the retinal receptor processing of the optical flow is sufficiently multidimensional to allow visual experience to be three and even four or more dimensional.

There is an interpretation of the function of the pupil/lens system, presented in several texts on physiological optics (e.g., Hecht & Zajac, 1974) that differs completely from either Gibson’s or the position taken here. Although the transformational description of the aperture/lens system is the same in this approach as in that taken in the holonomic brain theory, what differs is the interpretation of what is being transformed. In most interpretations other than those used in ecological optics or formalized by the holonomic brain theory, the events occurring on both sides of the aperture/lens system are described in complementary fashion, that is, in either space-time or spectral terms depending on which description is the most convenient. Such approaches do, of course, distinguish between holograms (a spectral representation) and ordinary photographs (a space-time image), but only with respect to how each is formed and not in relation to stages of perceptual processing. Complementarity in psychophysics is akin to complementarity in quantum physics (the Copenhagen solution to the Heisenberg duality). “Meaning,” interpretation, are eschewed.

Co...