The history of the study of learning
LEARNING AS- Stages in Skill Acquisition (Norman, 1982)
Norman described the learning process from within the sphere of skill acquisition. As such, he proposed that there are four ways of learning a new skill: skill accretion, learning by analogy, skill structuring and skill tuning.
Firstly, skill accretion is similar to Piaget's concept of assimilation- encoding new information in relation to pre-existing memory schemata. In effect, humans are taking information and adding it to a large body of knowledge that they already have (adding information to an existing store.)
The second stage is learning by analogy which concerns adapting existing schemata to new information. As a scenario brings with it a unique segment of knowledge and experience, this is then added to overall knowledge. It is the expansion of the schemata to blanket the numerous experiences a human will encounter.
The third, skill structuring relates more closely to Piaget's concept of accommodation - learning a new skill requires the creation of new conceptual structures or new ways of storing novel information. Skill structuring like this is performed more by children than adults, as children are continuously reconfiguring the way that they see the world. Rather than the 'stretch' of schemata which occurs during learning by analogy, skill structuring involves a 'restructuring' of schemata. Furthermore, it is important to note that there is often a big drop of performance during this stage as there is a shift from one way of doing something to a better way; it takes time to adapt to the new way of performing the skill.
Finally, skill tuning is simply the process of practice in order to master a skill, or the process of finely adjusting knowledge of the skill, e.g. taxi drivers having to learn routes through London in order to pass a test. This is probably the slowest, as it can take many hours of practice for a skill to become automatic, or to go from being a novice to an expert.
What are the characteristics of completing the stages- reaching expertise?
Norman believed there were distinct characteristics of experts, those that had proceeded through the varying stages:
Experts' performance is automatic; they do not require a great deal of attention to fulfil the task.
Experts also have low levels of stress when carrying out the task, as a result of the relatively lower level of required effort.
Experts' points of view are also different from novices, they process a problem from numerous angles.
Experts exert less mental effort on a task due to the increased and refined knowledge they have over novices, and they have no need to monitor their performance carefully.
Experts' perform with apparent ease, in an unhurried approach which is deceptively fast.
For an expert the task just becomes routine - we just 'drive' in the same way we just 'walk'.
Learning/ Skill Acquisition as a product of the environment
Semantic knowledge and procedural skills distinguish the expert from the novice and the adult from the child and together form a major source of individual differences. Humans, unlike other species are born with relatively few 'hard-wired' skills (reflexes). Through extensive practice any skill can be automatised and will appear as if it were originally hard-wired. This gives humans an evolutionary advantage. Humans are born with reflexes, but build skills up over time; the brain isn't committed to any set of actions when humans are born, therefore allowing us to build up skills related to our environment. In effect this can be described as a process involved in human learning; response to the surroundings. This is the centre of Psychology, and it is the process by which high cognitive functions, such as memory, occur.
Learning is not just the brain adapting to the enviroment, but also us adapting the enviroment to us. Therefore we are learning in two different ways when it comes to the enviroment! We just also remember that we learn by example. This involves the social learning approach to development, where we learn from the people around us and the more a role model can teach us, the more reinforcement of learning that occurs.
LEARNING AS Adaptive Control of Thought , Anderson (1982, 1983, 1987)
Anderson incorporated learning into his ACT* model. It comprises of three stages, named after Fitts and Posner's (1964) analysis of motor skill:
Stage one: The Cognitive (declarative) Stage
- A basic description of a particular procedure is learned; a vague outline of 'what' should be done.
- Weak (domain independent) problem solving is present which involves scraping together aspects of general knowledge to solve this particular problem.
- At this point, one is using their own general efforts to complete the task.
Stage two: The Associative (procedural) Stage
- In order to carry out the task, one must move convert from the "knowledge that"(declarative) to the "knowledge how"(procedural).
- A method needed for performing the skill is worked out.
- Strong (domain dependent) problem solving; specific knowledge tailored to the problem.
- Errors in our initial (general) understanding are detected and gradually eliminated, then connections between elements that actually are needed for successful completion are strengthened
Stage Three: The Autonomous (automated) Stage
- The skill becomes more and more rapid and automatic; more general clutter has been removed.
- Speed and accuracy improve.
- Verbal mediation is often lost.
- Can be linked to Norman's idea of "fine tuning"
Anderson claims that these stages can be applied to acquisition of motor and cognitive skills alike. This, however, was viewed by many researchers, as being a major weakness of the ACT*. Recent evidence now suggests though, that the cerebellum (a key structure in motor skill automatisation) is also centrally involved in acquisition of cognitive skill.
Describes how we are able to move from stage 1 (the cognitive/declarative stage), to stage 2 (the associative/procedural stage)
- Success in compiling knowledge requires us to take a 'snapshot' of all relevant circumstances
- Knowledge compilation is the process that creates efficient domain-specific productions from this trace of the problem-solving episode
- Have to have performed the skill once.
- The weak-method production is matched to declarative knowledge. This is then built into the new domain-specific production
- Following this, declarative knowledge no longer needs to be held in the working memory
- A sequence of productions is collapsed into a single production, which does the work of the whole sequence e.g. the process of chunking information (Miller, 1956).
- This can involve several processes:
Strengthening: moving the production further up the hierarchy as more recent productions will be used instead of original ones (even though the original ones are still available).
Generalisation: if there are two productions with a very similar format, a more general production may be created. This will replace each specific instance with a more general 'variable name'.
Discrimination: restricts over-generalisation.
ACT* Implications for Instruction
1. Amount of positive transfer between skills depends on the extent that they share the same production. Supported by research by Singley and Anderson (1986).
2. Use specificity of knowledge. McKendree and Anderson (1987) found intense practice in evaluation led to improved evaluation performance, but not improvement in the speed of the task. This suggests practice needs to be very specific to improve performance.
3. Immediate feedback is crucial to skill compilation, because it allows for a snapshot of the process we've just undertaken to be made (prerequisite in knowledge compilation) (Lewis and Anderson, 1985).
4. How (procedural) v Why (declarative). How is more effective in learning (Pirolli and Anderson, 1984). This has implications because textbooks often only involve information that stimulates declarative learning.
Cortical plasticity is where our brain changes physically as a result of experience. for example professional guitarists use their hands a lot more than most people, and as such the areas of the motor cortex that control the hands are often larger.
Recanzone, Schreiner and Merzenich (1993) Trained an owl monkey to discriminate differences in sound. Practice was shown to improve the monkey's ability, but only with frequencies specific to the range it practiced (around 2.5 Hz). Brain scans revealed that the monkey's primary auditory cortex had adapted with practice to improve skill; the 2.5khz frequency area had got larger.
JR Anderson's ACT* (Adaptive Control of Thought) Architecture
Level 1 of the Adaptive Control of Thought (ACT*) - a framework of cognition that can be applied to the process of learning. This model was proposed by Anderson in 1983. The framework makes 3 claims.
The first claim is that higher level cognition involves a unitary system. This means the whole of cognition can be explained by one system rather than being a multi-factorial approach suggested by Chomsky in 1980. So, it combines all of the different aspects of cognition (language, maths, reasoning, problem solving, memory) into one system.
The second claim is that cognition is based on a production system whereby everything can be modelled using principles which are more flexible than schema representations. These production systems are powerful enough to be computationally universal and model anything using productions but not too powerful so it's possible to make predictions and test them empirically.
The final claim is that the architecture of cognition is made up of 3 memory systems, these are working memory, production memory and declarative memory. It could be considered to be a more elaborate representation of the memory system than both the Multi Store Model and the Working Memory Model. Declarative memory is the equivalent of LTM in these alternative models and accounts for semantic knowledge (facts). There is also a production store which contains the mechanics of carrying out tasks and is executed in working memory.
Level 2 of the Adaptive Control of Thought (ACT*) - level 2 explains 3 knowledge structures that are used to store information:
A) Declarative memory - Built up out of "cognitive units" each comprising <5 elements/chunks. Cognitive units can be in 3 forms, either propositions representing meaning, temporal strings representing order, or spatial images representing spatial configuration. These cognitive units are then combined in a network which normally forms a "tangled hierarchy". This hierarchy is not in fact hierarchical, it simply refers to the mixing of these data types. A weakness of the ACT* model exists here as there is actually no explanation for how information in declarative memory is built up.
B)Working memory - Contains a variety of different sources: currently active productions, information deposited by production activity, information deposited by sensory encoding, and goals. Goals have a special status in the working memory store and are protected from decay through the creation of a "goal stack". The capacity of this store is more like 20 items as opposed to the traditional view of 7+/-2.
C) Production Memory - Encoded in a use specific way, this carries with it a general format of IF (conditions matched in Working Memory), THEN (Execute action). Examples will be explained in the Level 3 section of ACT*.
Level 3 of the Adaptive Control of Thought (ACT*) ACT Models- Anderson's approach is to write a specific 'model' for each domain, represented by a set of productions empirically tested by exposing it to stimuli and seeing how it changed using general learning architecture he has evolved. An application of the 'Adaptive Control of Thought' theory to a specific problem involves aspects of learning such as language acquisition, letter recognition and subtraction.
Looking at level 3 in more detail, an example of a model used in this level could be to arrange blocks in size order (smallest left, tallest right). Anderson describes how the model for performing this task is exposed to "stimuli" which can alter the next action or decision that is made. For example, if the smallest block is already on the left, this stimulus position means the action of moving it to the left can be removed from the action sequence - as it is not necessary. Anderson created these models as "general learning architecture", meaning that the same basic principles can be applied to a variety of tasks to achieve a particular goal. This construction of basic principles demonstrates how an artificial system, such as a computer, could process information and carry out actions accordingly.
The control of Cognition in ACT*
What is the central executive? Three linked processes:
a) Pattern matching underlies all varieties of cognition providing its data-driven quality. ie a specific production will apply if there is a pattern in the WM corresponding to its conditions. Partial matching is also possible.
b) Goal-directed processing provides top-down component where a specific production will apply if there is a pattern of WM that corresponds to its conditions resposible for the serial nature of cognition since only one goal can be achieved at a time.
c) Conflict resolution determines production chosen, in the event that there are several activating the WM, based on: degree of match, production strength, refractoriness, specificity and goal dominance.
ACT* as a model for Theory Construction
Low and high level theorising- e.g. concept of automaticity (low level) and Anderson's theory of learning (intermediate level). BUT Anderson's model was derived from ACT framework (high level theory).
=Not plausible in terms of neural processing,
=Production system architecture is not plausible
=It is not clear how we acquire skills before we are able to use declarative knowledge.
=A relatively simple account as it is single system it means the theory can simply be applied to the area of interest eg. learning how to drive a car. Much less complicated than multi-factorial accounts of thought such as the theory proposed by Chomsky (1980).
=Can be used to explain how skills become automatic.
=It is probably the best example of how to go about scientific theory construction in any domain.