Is Perception Based on Unconscious Inferences
Perception and awareness of our surroundings are things we continually take for granted and without these abilities it would be impossible to navigate our way safely round our environment. Roth (1986) described perception as, “The means by which information acquired `via the sense organs is transformed into experiences of objects, events etc” This raises the question of how the sensory stimulation detected by our sense receptors is converted into conscious perceptual experience?
Many theories of perception suggest that while the sensory pattern is dependent on the stimulus, it is often supplemented by other sources of information such as our knowledge and previous experience of the world and that inferences are made about the real world based on this information. This seemingly simple and effortless process has actually proved to be much more complex than first thought Inferences are logical judgements “made on the basis of a sample of evidence, previous judgements, prior conclusions, etc. rather than on direct observation.
Inferences made unconsciously are defined as, “judgements made on the basis of a limited amount of evidence or data and made without awareness” (Reber and Reber, 2001) and were first articulated by Hermann Von Helmholtz (1867), who considered them to be a result of associations and experience. Through these ideas he pioneered the constructivist approach of perception which considered information from the senses to be insufficient in forming perceptual experience which, according to Helmholtz, required various cognitive processes.
Unconscious inferences were thought to be the main processes involved in perception, adding meaning to the sensory input by unconsciously combining the information from the stimulus with other information to derive the perceived event. These inferences were described as being unconscious because we are not usually aware of this process and we cannot change our perceptions at will. This introduced the idea that perception was not a passive process but required intelligent problem solving based on knowledge.
Although Helmholtz did not fully develop this idea of unconscious inferences it was extremely influential and gave subsequent theorists a concept by which to explain the processes at work in perception. A modern and more detailed version of the constructivist approach is provided by Gregory (1973, 1978) who explained perception in terms of hypotheses forming and testing, predicting unsensed characteristics of objects.
He proposed that signals received by sensory receptors triggered neural events and interacted with appropriate prior knowledge and it was on this combination of information that he believed hypotheses were made to predict events in the world. This concept of unconscious inference has been demonstrated by perceptual constancies and illusions. The fact that we perceive a 3dimensional world from a 2dimensional image on the retina can be explained by the constructivist approach, using concept driven processes to fill in the gaps of information in the retinal image.
Unconscious inferences allow behaviour to be appropriate to objects even though all the characteristics of that object may not be available in the stimulus. For example, a table can still be perceived as a table even if one of its legs is occluded and therefore not present in the proximal stimulus. The visual system ‘fills in’ the occluded portion rapidly and automatically inferring that the object is in fact still a table and behave towards it in the same way we would to any other table ( Gerbino and Salmaso, 1987).
The size constancy mechanism works to maintain the perceived size of an object regardless of its reduction in size on the retina due to it being at a greater distance from the observer. This mechanism can be seen in a study by Ittleson (1951), where observers were presented with three playing cards in a darkened room, with depth cues removed. One of the cards was of normal size, another was half the size and the other was double the normal size. The largest card was judged to be closest to the observer and the smallest card was judged to be furthest away, even though they were all at the same distance.
This suggests that as well as depth cues, familiar size information is used in conjunction with change in retinal image size to gain an impression of relative distance. The Ames room illusion (Ames, 1949) shows how perceptual hypotheses may be very inaccurate if the stimulus information appears familiar but is actually completely new. The room is an irregular shape with a sloping floor and one end of the rear wall much further from the viewer than the other end. However, when viewed through a peephole, the room creates the same retinal image as a normal rectangular room.
The cues suggesting that the rear wall is at right angles to the viewer are so strong that observers tend to perceive the person inside the room as growing and shrinking as they walk across the room, rather than perceiving the room as being an unusual shape. Gregory suggested that this is because prior experience is of rooms of a rectangular shape and therefore hypotheses are formed based on the expected shape of the room. Although this explanation has been challenged (e. g. Day and Power, 1965), this remains a powerful demonstration of the vulnerability of perception and its inferential nature.
Shape constancy ensures that we still perceive the correct shape of an object regardless of the shape produced on the retina due to the object being viewed from a particular angle. For example, however much you tilt your head you still perceive trees and telegraph poles as being vertical. Brightness constancy works, for example, to maintain the perception of coal as being black even in bright light, and paper as white, even in dark shadow. This is similar to colour constancy which allows familiar objects to retain their colour under a variety of lighting conditions.
When and how a system breaks down is often used to give insight into how it normally operates which is why illusions have been used to give clues into the underlying processes in perception. Gregory’s theory accounts for a variety of visual illusions, of which he says there are four types; Illusions of ambiguity, paradox, distortion or fiction. Ambiguous illusions demonstrate how perception can be ambiguous since the same pattern of stimulation creates two different percepts through switch of attention.
This is evidence that perception cannot be derived directly from the stimulus in a one to one manner. An example is The Rubin Vase (Rubin, 1915), a picture where two black faces in profile and a white vase are seen alternately but cannot usually be seen at the same time. The size constancy mechanism can also be misapplied to produce distortion illusions such as the famous, Muller-Lyer illusion where two vertical lines are depicted side by side, one with fins pointing inwards at each end, the other with fins facing outwards.
Although both vertical lines are the same length, the one with the fins pointing inwards is perceived as shorter than the one with the fins pointing outwards. Gregory explains this in terms of depth cues being used automatically as if these figures were 3dimensional. In the 3dimensional world inward facing fins would suggest an edge jutting towards them, whereas outward facing fins would suggest an edge pointing away from them, therefore revealing how far away the object is.
Receding, more distant objects are opposed by the size constancy mechanism, which enables us to see objects at their real size despite changes in distance, therefore the line with the outward facing fins is perceived as smaller because it has been scaled down based on what are thought to be cues of depth, suggesting that this line is more distant than the other line. Another type of illusion demonstrates how we use our existing knowledge and expectations to make inferences about external stimuli, leading to distortions of reality.
The hollow face illusion shows that there is a strong bias to see faces, or all complex objects, as convex despite monocular depth cues suggesting otherwise, which seems to be an effect that can only be explained by prior experience effecting perception. Even when we know the face is hollow it is still perceived as solid, indicating that it is not conscious knowledge that is influencing perception. The Kanizsa triangle is an example of an illusion of fiction, where we see what is actually not there, showing that we can actually perceive what is not given in the stimulus.
All these examples suggest that perception involves construction based on inferences and hypotheses. What we perceive is not in the data but in our perception of it and our attempt to construe the stimulus in the way we normally perceive the world can mislead us. However, this constructivist approach and the idea of perception being based on unconscious inferences suggest that perception will often be in error, but perception is typically very accurate.
Experiments carried out to support this approach tend to involve static, degraded and artificial stimuli and presentation of these stimuli is usually very brief which reduces the impact of bottom-up processing, causing greater reliance on top-down processes such as hypotheses and inference. The retinal image may not be so impoverished in the real world , which rarely contains projections of single, isolated objects, but consists of more complex images with surrounding and background objects.
Movement is also a vital part of perception which allows multiple perspectives of the environment but is lacking in most constructivist experiments. Perhaps it is possible that these theorists underestimated the richness of sensory information available in the real world, therefore relying too heavily on inferences in the explanation of perception. It is also difficult to see where inferring begins, since if one inferential belief causes another, surely some properties must be perceived directly initially.
Gibson (1979) introduced a direct theory of perception, which conflicted with the popular constructivist approach. He assumed that all the information that was needed to perceive the environment was available directly in the stimulus, without any internal processes of inference being required. He thought that constructivists had been wrong in considering the stimulus to be a static image on the retina, which he suggested was only the input for a single receptor.
Instead the input for a perceiver is the pattern of light extended over time and space, referred to as the optic array, made up of several million light rays of different wavelengths and intensities, showing that light from the distal stimulus can be a rich source of information. Not all light comes directly to the eye, many light rays are first reflected off other objects, and therefore the eye receives complex interactions of light rays moving in different directions, which also change dramatically as the observer moves around their environment.
However, it also contains invariant information which remains constant in the changing environment. Perception does not involve adding up the features of objects and comparing them to stored knowledge but, involves ‘picking up, the invariants which provide information about the layout of objects in the environment. He claimed that this was done in an automatic way, by a process of resonance involving receptors being attuned to these invariants. They have become attuned to appropriate aspects of the environment by perceptual learning that has occurred in the history of evolution, rather than by experience in the person’s lifetime.
There are two types of invariants; transformational invariants which are patterns of change, and structural invariants, which are ratios of differences between objects. Gibson uses this notion of invariants to explain perceptual constancies without using previous knowledge to make inferences. For example, when a piece of coal is under intense illumination it may reflect a great deal of light, but it will still reflect less light than will more reflective objects in the same scene, such as a piece of white paper.
The ratio of reflected light between objects is a structural invariant which will remain the same regardless of the intensity of the light shinning on them. It is not the total retinal illumination that matters but the ratio of intensities of the two patches of light (Wallach, 1972). An example of a transformational invariant is the proportionally diminishing object as we move away from it. Departures from this invariant pattern would indicate that either there is a change in the rate of movement or the object is changing size.
Therefore Gibson concludes that there is enough information in the light from the visual scene to perceive the ‘true’ properties of the stimuli. Gibson took an ecological approach to perception, in that he considered its primary function to be the facilitation of interactions between the individual and their environment. He did not think that meaning had to be extracted from long term memory, but that all potential uses of objects are directly perceivable in the pattern of stimulation arising from that object is self, which he referred to as the objects affordances.
The affordances of a particular object do not require experience and are perceived directly in the stimulus. Evidence supporting this comes from studies showing that objects of a graspable size are responded to differently than non-graspable objects by infants as young as three months old (Gibson, 1982). However, it is necessary to learn which affordances achieve which goals, during an individual’s lifetime since most objects give rise to more than one affordance and the particular affordance that influences behaviour depends on the current psychological state.
However, it is very difficult for direct theories of perception to explain illusions and they are usually dismissed as being artificial and unrealistic situations but this is not always the case. Some visual illusions produce effects similar to those found in normal perception. The hollow face illusion does not involve static observers or monocular viewing yet observers persistently see this face as solid despite useful information to the contrary (Bruce and Green, 1990).
There seems to be little room for possibilities of explanations for this illusion that do not involve making inferences based on expectations stored in memory . Such demonstrations make it clear that perception is not perfect, as suggested by Gibson’s theory, which claims that there is no difference between sensation and perception. Processes involved in identifying invariants, discovering affordances and resonance are much more complicated than implied by Gibson.
When workers in the field of artificial intelligence attempted to create a model that could extract invariants from the stimuli it proved a very difficult task which led them to believe that Gibson greatly underestimated the difficulty of the detection of physical invariants (e. g. Marr, 1982). Gibson’s theory does not adequately constrain the notion of invariants in the optic array that can be directly ‘picked-up’ or what it is in the nature of the stimulus that makes affordances directly available to the perceiver.
Nor does his theory adequately constrain the process involved in ‘picking-up’ these invariants, which he refers to as resonance but provides no possible operational definition. If sufficient constraints are not imposed then any property in the optic array could be an invariant and any psychological process could be used to explain the ‘picking-up’ of these invariants, which tells us nothing about the processes involved in perception.
Fodor and Pylyshyn (1981) argue that invariants and the process used to detect them can only be constrained appropriately if it is assumed that perception is mediated by inferences, as argued by the constructivist theorists. If invariants are constrained to a certain class of properties which can be directly detected, inferences must surely be made about the whole object or event from these invariant properties. They used the example of attempting to detect a fake Da Vinci painting from a genuine one to illustrate how in some cases information from knowledge must be used to perceive such a difference.
Only an expert of Da Vinci paintings would be able to perceive this difference, based on a great deal of previous experience, mediated by inferences. Neisser (1967, 1976) proposed the cyclic model of perception, which suggested that top-down and bottom-up processes, may interact with each other. He thought that schemas, collections of knowledge based on past experience, were used to guide the perceiver’s exploration towards relevant stimuli in the available data.
If the information retrieved from the stimuli does not match the information in the relevant schemas then the hypotheses is modified accordingly. There is an initial bottom-up analysis of the cues in the environment leading to an inference being made about what the particular object is. This sets in motion a top-down process involving a search for the particular features of the inferred object based on schemata of that object. If features from the information in the stimuli do not conform to the original hypotheses then new hypotheses are proposed and tested.
Gibson’s theory may apply to some aspects of perception more than others. This theory emphasises bottom-up processing, where as the constructivist approach puts more emphasis on top-down processes but the relative importance of either of these processes depends on various factors (Gordon, 1989). Gibson focused on visual perception under optimum viewing conditions and is concerned with perception for action, where as constructivist theories are based on sub-optimal viewing conditions and are concerned with perception for recognition.
Cognitive neuropsychology has found evidence to suggest separate visual systems; the ventral processing stream which is more involved in perception for recognition and the dorsal stream which is more involved in perception for action (Goodale and Humphrey, 1998). This implies that perception may operate differently in different circumstances and in this way, both direct and constructive theories of perception may be correct depending on the situation. Different processes might be involved in perception in higher animals, such as mammals than in other species of animals.
In higher animals, experience may play a larger role in perception compared to simpler animals whose perception may be innate. For example, pigeons do not behave towards partially occluded objects in the same way that they behave to the completely visible object (DiPietro et al, 2002). Nobody is yet in a position to make a final judgement between constructivist and rival approaches to perception. It is possible that we perceive constructively, basing our perception on unconscious inferences whenever the situation is in any way unnatural or when available data is degraded.
However, when we are in good light and in natural surroundings, it is quite possible that perception occurs directly and reflexively, without any unconscious inferences being necessary. Many theorists suggest an interaction between direct and indirect perception, (e. g. Fodor and Pylyshyn, 1981, Neisser, 1967, Marr, 1982) suggesting that although perception of some properties may not involve unconscious inferences, they remain an important part of the perceptual process. Gibson’s suggestion of a theory of perception without the involvement of unconscious inferences is insightful but has flaws when it comes to explaining misperception.
Therefore, there is as yet no satisfactory alternative account of perception which leads to the conclusion that unconscious inferences are important in at least some aspects of perception. Present evidence suggests that perception is not based solely on unconscious inferences. The stimulus may have been underestimated in the amount of information that it contains and the influence it has directly on perception. However, unconscious inferences so far seem to be an integral part of the process of perception.
Get access to
Guarantee No Hidden