Gregory (1973) made use of visual illusions, such as the 'Ponzo perspective illusion' to show how top-down processes affect perception. His view was that perceptions are hypothesis which are tested against the sensory data. In the Ponzo illusion, the upper of 2 lines is perceived as longer even when the viewer knows that the 2 lines are the same size. Even when we understand that the illusion contains angles (indicating distance ) which we are wrongly interpreting as depth cues, we can't make the illusion go away. At this stage the perception is not conforming to the perceiver's expectation of seeing the lines as they really are, of equal length. This indicates that the top-down processing in this case is unconscious, and therefore not in our control, which does not match the description of the first mode of processing which is top-down and consciously applied to high level cognitive tasks such as problem solving and reasoning.
Perceptual hypotheses have been shown to exist where general principles like 'size constancy' are involved but it is not so certain that specific hypotheses are made about all sensory input. From the evolutionary perspective, bottom-up processing would be more likely to be selected for, as it is faster. We would be more likely to survive if we took action directly after perceiving a wild animal from sensory data, than if we were to make a hypothesis about what was being seen and then test it. Gibson's (1979) theory of direct perception states that as the information we receive from the outside world is so rich and complex that we have no need for top-down processing.
This theory accounts for the fast responses we show in racquet games and in fleeing from danger and also illustrates the 2nd mode of cognitive processing identified by the course. It does not , however, explain visual illusions, so perhaps perception is a complex combination of both theories.
There is a lot of evidence that people use a variety of top-down processing in reading. The general use of 'context' when reading allows us to disambiguate meaning; heterophonic homographs such as 'tear' and 'bank' are interpreted by context. Words have also been shown to have a priming effect, that is, words activate semantically related words in the internal lexicon. Meyer & Schvaneveldt (1971) found that people were faster to recognise a word if they had already recently processed a related word. We also use inferences to make sense of text and our stored knowledge can result in expectations of the text which could result in false memories of it. Bower, Black and Turner (1979) found that people 'remembered' facts not explicitly stated in a story they had read. In reading, then, bottom-up processing of the words from the page is interlinked with top-down processing providing inference, expectation and interpretation. The bottom-up processing is serial rather than parallel (one word at a time) and is conscious although some words are read automatically. A.F. Healy (1976) showed with her automatic recognition tests that subjects who were asked to circle all the letter 'T's on a page, often missed out the 'T' in the word 'The". The reading of the word 'The' was so automatic (it is one of the most common words in our language) that the reading of it could not be modified into single letter recognition. The 2 modes of cognitive processing are not then, distinct, when performing a reading task.
The 2 modes of processing have been accounted for at levels of explanation higher than the topic-related theories discussed so far, three cognitive architectures have also been introduced, namely the Schema based architecture, the Production system architecture ACT* and the PDP architecture. An architecture is midway between a framework which is a general set of ideas at a high level, and a theory which is specific to an area within the framework and is both predictive and falsifiable. An architecture can also make predictions but is broader than a theory and tries to provide theoretical integration.
Whereas both schema based and production system architectures are rule based, PDP reacts to stimuli by activating a pattern of strength across connections. This allows parallel processing to be truly simultaneous unlike the sequential control exercised in hierarchically and herterarchically organised systems. This capability poses difficulties in the design of PDP systems as the processes being performed have to be synchronised where there are logical dependencies between processes. This may be the case where an expert was automatically accessing say, patterns of chess pieces from previous games, and consciously thinking about the game currently being played, at the same time. There may be a logical dependency between the automatic processing of his expert knowledge and the decision he takes re. his next move. However, it is notoriously difficult to obtain verbal protocols giving insight into expert knowledge processing, as this process is unconscious and therefore unavailable to introspection. Chase & Simon (1973) ran experiments with chess players and found that experts store 'chunks' of information representing typical move sequences and board positions in chess, which gives them their advantage over novices. In schema theory terms, the experts had stored considerable knowledge in highly organised schemas within the domain of chess. Playing chess, then, or any problem solving involving expert knowledge, seems to involve a combination of the 2 modes of cognitive processing. The bottom-up stimulus of the problem presented, results in automatic processing of expert knowledge in parallel with conscious top-down serial thought about the solutions to the problem.
Both modes of processing are also accounted for in the Production system architecture, ACT*, (Anderson (1983) ). In this theory, accumulated domain relevant facts in declarative memory can become proceduralised, that is, become part of the production memory. In this theory, LTM holds both statements of knowledge in declarative memory (which consists of semantic networks) and production systems in production memory which can be processed automatically when needed. After proceduralisation, declarative knowledge becomes embedded in procedures, this speeds up performance as it is no longer necessary to search for and retrieve the information from declarative memory. Anderson explained that Working Memory activates nodes in the declarative network and this activation controls the matching and firing of productions. Working memory then consists of the active units in declarative memory and information derived from production firing and perception. This architecture accounts for the 1st mode of cognition in the activated nodes of working memory and the 2nd. in the automated, fast procedures of the production memory. It does not have a central executive but exercises control by being rule based and through 'conflict resolution'. But as It spends about 95% of it's time 'pattern matching' or deciding what to do, whereas humans are good at pattern-matching and seem to do it instantaneously, the problem of the control of human cognition has not really been addressed by ACT*.
The attention paid to any set of tasks is said to be controlled by a central executive about which not very much is known. There is a limit to the amount of processing we can cope with at any one time, Kahneman's capacity model is helpful in understanding this. It shows a central processor which deals with the allocation of resources according to the capacity available. The capacity available varies according to our level of arousal; we can process more sensory input when alert, than when tired or sleepy. The processor evaluates the mental effort currently required and if it is more than there is capacity for, it takes account of our current goals (momentary intentions) and our survival instincts (enduring dispositions) in allocating the available resources. That is, it prioritises the tasks according to our goals, but will divert attention to an emergency situation. The working memory model (Baddeley 1990) also shows a central executive which controls the conscious aspects of cognition. This central executive controls the 3 slave systems called the 'articulatory control system' the phonological store' and the Visio-spatial scratchpad. Each has been shown in various experiments to have it's own capacity and to work independently of the others. The central executive is characterised by its limited capacity but as yet this capacity has not been measured. Also it does not attempt to explain how actions become automatic with practice, as ACT* does.
R.M. Shiffrin & W. Schneider (1977) drew up a list of the characteristics of both automatic and attentional processes. They found that skills which take some training to acquire, such as driving and swimming, become automatic. Attentional processing is used during the learning period; it is necessary to devote all of our attention in order to master the inter-related actions involved in say, driving a car. But once the skill is acquired, it becomes automatic, and then it is possible to perform some other tasks simultaneously. Attentional tasks on the other hand, do not involve a great deal of training, it would not take long to show someone how to count out all the coins from a jar and arrange them in equal piles, but it would not be possible to perform any other tasks at the same time. The last characteristic described by Shiffrin & Schneider was that of consciousness. Although there are several differences between automatic and attentional processing, this is the characteristic that is probably the most easily identified. Attentional tasks are those we are consciously aware of doing. We are conscious of paying attention to that task and if our attention is attracted away, then processing of that task stops. Automatic processing is largely unconscious; we can drive a familiar route and arrive without having been conscious of the journey.
The 2 modes of processing then, have been shown to exist, but they are not always distinct, rather they exist at either end of a dimension with lots of variation in between. It has been useful, however, to draw the distinction between the 2 extreme types of cognitive processing for clarity. Mode 1 describes the typical characteristics of conscious processing and mode 2 describes the typical characteristics of unconscious processing but in reality there is a complex relationship and interaction between the two.
