8.6 Ecological vs. Empirical Approaches

Instead of ... [assuming that] perception is some kind of internal operation of the brain (the seat of the mind) on the signals from the world, e.g., interpretation, addition, supplementation, or organization, but in any case a "processing" of the input ... I suggest that the act of perceiving is one of becoming aware of the environment, or picking up of information about the environment, but that nothing like a representation of the environment exists in the brain or the mind which could be in greater or lesser correspondence with it--no "phenomenal" world which reflects or parallels the "physical" world Q. J. Gibson, 1974/1982, pp. 371-72).

Gibson (1979) found himself at odds with both the fading metaphors of behaviorists, who often likened the brain to a mechanical device, and the emergent metaphors of the cognitivists, who frequently spoke of the brain as a computer. One of his important insights was that actions involved in detecting and selecting information are, like orienteering (the use of a map and compass to navigate between checkpoints along an unfamiliar course) just as important to subsequent understanding of what is perceived as the processing of sensory stimuli.

Gibson's ideas about the importance of orientation led him to question the mind-body dualism of behaviorists and cognitivists , which dualismassumes the mind is a mechanical device or a computer and is therefore separable from mental phenomena (see 2.1, 7.2). Essentially, Gibson converted this ontological dualism into a useful methodological distinction:

The aim of a perceptual experiment was no longer to ascertain what hypothetical processes converted inadequate inputs into percepts, but rather what stimulus variables specified what perceptions (J. J. Gibson, 1979, p. 25).

This methodological innovation regarding stimulus variables led Gibson to the

... novel distinction between literal and schematic perception. When psychophysical experiments are arranged so that a subject "will make the best observations of which he is capable," perception is usually accurate and veridical, yielding what Gibson called the literal visual world. When one's experimental method involves "impoverished, ambiguous, or equivocal stimulation" with constraints on observation, such as brief exposure time, one obtains schematic perception of the visual world. Schematic perception is often inaccurate and erroneous, but, Gibson insisted, "the alterations and distortions might have been eliminated if the conditions of observation had been different" (Gibson, 1950, p. 217). The effects of perceptual habits and social custom on perceiving were assimilated by Gibson to the concept of schematic perception: The origin of perceptual error lies, not in mistaken processes of imagination, but in a mixture of inadequate stimulus conditions and the human tendency to adopt the customs of the group, even when they are less than fully adaptive (Reed, 1988, pp. 184-185).

Perhaps Gibson's (1979) most serious doubt about information-processing models was that such models focus on the organism's analysis of stimulus information at the expense of the organism's activities in detecting and selecting stimulus information. Thus, information-processing models tend to minimize the context of stimuli-their locality, temporality, and relatedness to other factors in the environment and in the organism.

8.6.1 Direct Perception, Context Sensitivity, and Mechanicalism

The modern theory of automata based on computers... has the virtue of rejecting mentalism, but it is still preoccupied with the brain instead of the whole observer in his environment. Its approach is not ecological. The metaphor of inputs, storage, and consulting of memory still lingers on. No computer has yet been designed which could learn about the affordances of its surroundings Q. J. Gibson, 1974/1982, p. 373).

Despite significant improvements in sensor and computing technologies, artificial-intelligence (AI) systems have been unable to emulate everyday tasks performed by animals and people, because Al technologies lack sufficient means for selecting and encoding contextual and situational variables (Winograd & Flores, 1986) and because artificial intelligence is not embodied (Joh9son, 1987). As McCabe (1986) notes, " 'Context sensitivity' is the bane of all associationistic models" (p. 26).

In the process of reinventing the concept of retinal imagery that underlies his radical theoretical postulates concerning perception, Gibson (1966) implicitly relied on the context and situatedness of ambulatory vision. In his empirical research, he paid particular attention to the boundary conditions that affect and constrain visual perception in everyday living. This investigatory focus led Gibson to findings that be could not explain within the paradigms of the positivist tradition that convention had imposed on his discipline. Thus, Gibson was forced to rethink much of what psychologists bad previously supposed about perception and to propose a new approach as well as new theoretical concepts and definitions.

The problem is that positivism relies almost exclusively on the traditional physicist's characterization of reality as matter in motion in a space-time continuum. This "mechanicalism" of Newtonian physics and engineering is allied with sensationalism, a set of assumptions permeating philosophy, psychology, and physiology since the modem era. Roughly speaking, sensationalism maintains that only that which comes through the senses can serve as the basis for objective scientific knowledge. Sensations, however, as Gibson (1966) consistently argued, are not specific to the environment: They are specific to sensory receptors. Thus, sensations are internal states that cannot be used to ensure the objectivity of mechanistic descriptions.

Conventional psychology relies on sensationalism and mechanicalism to treat perception as a mental process applied to sensory inputs from the real world. This treatment of perception, however, fails to bridge the gap between (a) incomplete data about limited physical properties such as location, color, texture, and form, and (b) the wider, more meaningful "ecological awareness" characterized by perception of opportunities for action.

The grand irony of the Cartesian tradition in psychology is that it forces on proponents of the mechanistic schools of thought concepts from the mentalist tradition, and vice versa. Recently this Cartesianism has cloaked itself in the computer metaphor of information processing; nonetheless, many of the most prominent neuroscientists of this century have been forced by their mechanistic account of perception into an outright mind-body dualism. What has been left out of the picture altogether in 20th-century psychology, as Gibson makes clear, is the active self observing its surroundings (Reed, 1988, p. 201).

Ecological psychologists employ "geodesics" (Kugler, Shaw, Vincente & Kinsella-Shaw, 1991, p. 414) to complement mechanistic systems of description based on Cartesian metrics. Examples of geodesics are least work, least time, least distance, least action, and least resistance. Ecological psychology conceives of these action pathways as "streamlines" through the organism's niche structure and environmental layout rather than simple traversals of Cartesian space.

Geodesics are constrained by factors such as gravity, vectors associated with the are of an organism's appendages or sensory organs, and energy available for exertion. For a simple example of geodesics, consider how cowpaths are created by animals avoiding unnecessary ascents and descents on an undulating landscape: In addition to serving as records of travel through Cartesian space, the paths reflect cow energy expenditure and the ability of the cows to detect constraints imposed by gravity.

Geodesics are essentially a thermodynamic construct, and as such they can be applied to human activity in media environments. Optimal perceiving and acting in mediated environments does not necessarily follow boxes, frames, or other contrivances based on arbitrary grids imposed in the Cartesian tradition-pages, tables, rules, keyboards, screens, and the like. True optimums for action and perception must be measured in terms of cognitive and corporal ergonomics rather than the metrical efficacy assumed by a one-grid-fits-all-organisms approach. Designing keyboards to conform to a grid may simplify circuitry and manufacture, but such keyboards may strain the human wrist.

Media designers and researchers can use geodesic analysis to study how users interact with print and computer-based media by, for example, tracking the extent to which users recognize opportunities for action afforded by features such as headers, indexes, icons, "hot buttons," and modal dialog boxes. In terms of thermodynamic efficiency, skilled use of shortcuts and navigational aids to wend one's way through a media environment is similar to the challenge faced by the cows: What pathway of action yields the desired result with the least expenditure of energy?

8.6.2 Situation and Selectivity

In place of a sensation-based theory of perception, Gibson (1974/1982) proposed a theory based on situations and selectivity: Perception entails the detecting of information, not the having of sensations. Rather than assuming a hypothetical perceiver, Gibson opted for a real, everyday perceiver, with all the possibilities and limitations implied by the ordinary. He situated this perceiver in an environment populated by ordinary, everyday people, living organisms, and natural as well as artificial affordances, rather than imagining the perceiver in an objectively accessible world defined and measured by conventional, mechanistic physics.

Gibson also appropriated familiar terms to create a new ecological vocabulary designed to complement the lexicon of physics (Reed, 1988):

1. Substances, surfaces, and media as complements for matter
2. Persistence and change as complements for space and time
3. Locomotion as a complement for motion
4. Situatedness in a niche as a complement for location in space and time

Gibson's (1979) development of ecological theory began with studies of the properties of surfaces. He identified several issues that have since proved important to designers of virtual realities and simulations.

First, a surface is not discrete like a detached object, and thus surfaces are not denumerable. Instead, a surface is nested within superordinate surfaces. Second, a surface does not have a location as an object does, a locus in space. Instead, it is part of what I call the environmental layout; it is situated relative to the other surfaces of the habitat underlaid by the ground, the surface of support (J. J. Gibson, 1979, p. 351).

8.6.3 Alternatives to Traditional Empiricism

The idea of environmental layouts serves as a useful example of the tension between the mechanistic approach and the ecological approach. "Environmental layout" reflects a persistent concern expressed in the writings of ecological psychologists: The very successful systems of formal description and analysis employed by classical physics have been misapplied in describing the fields of action and perception available to organisms. There is little doubt that descriptions derived from classical physics are well suited to disciplines such as mechanical engineering and even biomechanics. Nevertheless, if we infer from Thermodynamic principles that opportunities for action are ultimately determined by complexity of organization rather than space and time per se (see earlier discussion), then the usefulness of space-time grid maps for analyzing and explaining organic behavior is only partial. More useful are environmental layout maps that indicate opportunities and pathways for action and perception.

Critics such as Fodor and Pylyshyn (1981) have questioned the empirical foundations of ecological psychology, demanding that its new lexicon be verified within the conventions of laboratory-bound experimentalism. Yet, ecological psychologists such as Koffka (1935), Johansson (1950), Lashly (1951), McCabe (1986), and Turvey, Shaw, Reed, and Mace (1981) share with field biologists and anthropologists doubts about excessive reliance on laboratory experiments for gathering data relevant to the study of complex interactions between organisms and their environments.

Experimental psychologists often seem to feel that context effects are to be controlled and eliminated from an experiment if at all possible. This, we would argue, is a mistake. One can indeed suggest that some of the most serious conceptual errors in the history of psychology--errors that misled researchers for decades--began with naive attempts to remove phenomena from their natural contexts. We would argue rather that context effects are impossible to eliminate and that we should not wish to eliminate them totally, but only to study them. There is no zero point in the flow of contexts. They are not incidental phenomena that confound our careful experiments: They are quintessential in psychology. There is no experience without context (Barrs, 1988, p. 176).

Like many other life scientists, Gibson (1979) had to defend his ideas against some fairly vociferous opponents. Many of his defenses were polemical, and in our reading of his work we have learned to tolerate an imprecision in terminology and syntax that unfortunately left his ideas and arguments open to misunderstanding and marginal criticism. Without, then, either defending or exonerating his rhetoric, we offer our summary of Gibson's views on empiricism:

1. Empiricism can be distinguished from objectivism.
2. Eschewing objectivist theories of description need not imply abandonment of the scientific method, only rejection of unwarranted extensions in which the observations of a hypothetical observer are elevated to the status of a "God's-eye view" (Putnam, 1981).
3. The risks of misunderstanding inherent in cultural relativism, objectivism, and scientism can be ameliorated if reports of empirical observations are taken as instructions to others about how to replicate or verify findings and experiences rather than as veridical descriptions of reality (Winograd & Flores, 1986). We would add that when the authenticity of mediated representations is doubtful, the most ethical policy is to ensure that users can obtain instructions about how to replicate or verify represented objects and events. Lacking such instructions, users should be able to access information about the provenance of the representations.