Richard Griffiths - Lecture Notes

Learning 

LEARNING, acquiring knowledge or developing the ability to perform new behaviours.  [MS Encarta Online]

Thorndike's Law of Effect: An action that leads to a desirable outcome is likely to be repeated in similar circumstances.

Stimulus response model

 
Similar circumstances: 
 stimulus
Action: 
 response
Desirable outcome: 
 reinforcement
 

Schematic diagram showing the elements: stimulus, organism, response, of basic learning theory

Organism as a black box: Associative learning.

Organism as a glass box: Conceptual learning.

These two approaches are not necessarily mutually exclusive and may represent mechanisms that can operate simultaneously in human behaviour.  Whether or not conceptual learning occurs in other species is a matter of controversy.

Difference between radical and methodological behaviourism

When studying learning from an associationist or behaviourist standpoint, there are two philosophical positions that that may be adopted.  They relate to your belief in the existence of and explainable underlying mechanism.  These are the radical stance and the methodological stance.

The radical stance asserts that all that can sensibly be said about learning mechanisms is the objectively observable external behaviour.  Anything said about any presumed mental process is meaningless nonsense.  This is caricatured by the joke about the two behaviourist making love.  When they had finished, one rolled over and said: "That was obviously good for you, how was it for me?"  I.e., subjective experience is irrelevant to an understanding of the phenomena.

The methodological stance simply asserts that confining explanations to objectively observable phenomena is a useful way of making scientific progress, but does not deny the possibility of there being subsequent explanations in terms of internal processes.

During the first half of the 20th Century, radical behaviourism held sway in academic and applied psychology.  This dominance was broken in the late 1960's with the rise of cognitive psychology.  However, whatever stance you take, the learning regimes discovered and recorded by the behaviourists are real phenomena, and can be applied in appropriate situations.  For example, a search of the web using the term 'operant conditioning' will bring up a large number of dog training sites.

Procedural verses declarative knowledge

In the literature on learning a distinction is often made between two types of knowledge: procedural and declarative.

Procedural knowledge consists of those things that you can be said to know, but that you can only express by performing them.  A paradigm case is the knowledge required to ride a bicycle.  Typically, what a rider can say about the process is extremely limited, but they can demonstrate quite deep knowledge when they actually ride.

Declarative knowledge on the other hand is the classic text book knowledge, which can be talked about and recited meaningfully.

Associative learning

This involves an organism making a connection (or ‘association’) between a particular situation and a particular response.  In the psychological literature, this usually means an external and objectively observable situation and a behavioural response.

Two major types of learning come under this heading:

  • Classical or Pavlovian conditioning
  • Instrumental or Operant Conditioning
  • Classical or Pavlovian Conditioning

    Associative learning in which there is no contingency between response and reinforcer.
    Contingency: production of the reinforcer is dependent on the response being made.
    This was first described in the psychological literature of the 1920s by Pavlov.  He conducted a series of experiments in which dogs were trained to associate a ringing bell with the presentation of food.  Increase in saliva production by the dogs was measured.

    Initially the stimulus that is to be conditioned, (known as the conditioned stimulus, CS, e.g. a bell ringing), provokes no, or a week response in the organism, but a measurable unconditioned response (UR, e.g. saliva production) to an unconditioned stimulus (US, e.g. food) is detected.  During training, the CS is repeatedly presented together with the US; eventually the subject forms an association between the US and the CS.  In a subsequent testifies, the organism will show the conditioned response (CR, e.g. saliva production) to the CS alone.

    Schematic diagram illustrating the phases of classical conditioning.

    Such ‘Pavlovian’ conditioning is contrasted with instrumental or ‘operant conditioning’, where producing a CR controls the US presentations.

    Instrumental or Operant Conditioning

    Associative learning in which there is a contingency between the response and the presentation of the reinforcer.

    This was most thoroughly studied by B.F. Skinner and his students under the title of ‘Behaviourism’, in experiments in which rats and pigeons were trained to press a lever in order to obtain a food reward using the ‘Skinner-Box’.   You can try this out on the computer by downloading 'Sniffy the Virtual Rat'.

    In such experiments, the organism is able to generate certain motor output, (the response R, e.g. running around, cleaning, resting, pressing a lever).  The experimenter chooses an appropriate R (e.g. pressing a lever) to pair with an unconditioned stimulus (US, e.g. a food reward).  Often a discriminative stimulus (DS, e.g. a light) is present, when the R-US contingency is true.  After a training period, the subject will show a conditioned response (CR, e.g. pressing the lever) even in the absence of the US, if the R-US association has been memorized.

    Schematic diagram illustrating the phases of operant conditioning.

    More complex behaviours (e.g., pigeons playing Ping-Pong) were also developed by the technique of ‘shaping’ in which successive approximations to the desired behaviour are reinforced.

    Regimes of reinforcement and rates of extinction of conditioned responses were examined exhaustively to produce mathematical descriptions of performance.  Amongst the significant results found were; reward is better than punishment for shaping behaviour (or in more behaviourist language, positive or negative reinforcement is more effective than punishment), and conditioning by the use of sporadic reinforcement will result in a longer period to extinguish the behaviour once the reward is withdrawn.

    Conceptual learning

    Where an organism learns a set of internal representations, that subsequently modifies its behaviour in particular situations.

    This idea is controversial!  Hard line behaviourists or associationists would dispute that straightforward conditioning is insufficient to explain the most complex of learned behaviour.

    However, cognitive psychology is premised on this approach.  It also has a ‘common sense’ (or folk psychology) appeal.  E.g., if I tell you that a piece of electrical equipment is dangerously live, you would probably modify your behaviour with respect to it without having to be conditioned by receiving an electric shock first!  I.e., you hold concepts of ‘electricity’ and ‘danger’, and the new information I have given you links into these pre-existing ‘ideas’.

    How are these ideas or representations held in your head?  A number of candidate mechanisms have been proposed, including frames, scripts, prototypes, and even a sort of ‘mental logic’.

    Are the mechanisms that build and manipulate these representations innate and fixed, or can they also be modified or learned?  Early work in this area was carried out by Piaget, who studied the development of conceptual thinking in children. (e.g., the tall and fat tumblers experiment.)

    Dismissing mental logic as a representation method (for the detailed argument on this, see Johnson-Laird), the other mechanisms have been described as special cases of the more general concept of ‘mental models’.

    Mental Models

    The concept of mental models has been most clearly defined by Johnson-Laird. "Mental models emerged as theoretical entities from my attempts to make sense of inferences, both explicit and implicit. They replaced the formal rules of a hypothetical mental logic.  Subsequently, I was able to give a better explanation of meaning, comprehension and discourse, by postulating mental models in place of other forms of semantic representation.  The theory developed piecemeal, but was extended and corroborated by computer implementations and experimental investigations.  It is now plausible to suppose that mental models play a central and unifying role in representing objects, states of affairs, sequences of events, the way the world is, and the social and psychological actions of daily life.  They enable individuals to make inferences and predictions, to understand phenomena, to decide what action to take and to control its execution, and above all to experience events by proxy; they allow language to be used to create representations comparable to those deriving from direct acquaintance with the world; and they relate words to the world by way of conception and perception." [Johnson-Laird 1983, p. 397.] Essentially, mental models must subscribe to the following three constraints: "1. The principle of computability: Mental models, and the machinery for constructing and interpreting them, are computable.

    2. The principle of finitism: A mental model must be finite in size and cannot directly represent an infinite domain.

    3. The principle of constructivism: A mental model is constructed from tokens arranged in a particular structure to represent a state of affairs." [Johnson-Laird 1983, p. 398.]
    How this might be physically realized in a brain is a separate issue, just as the way that a data structure is physically represented in the memory of a computer is independent of its representation in a programming language. (Though the actual implementation may effect efficiency of manipulation and the type of errors that malfunction of the memory might generate.)

    If this view of acquired behaviour is taken, then learning is seen as the acquisition and modification of these mental models, and behaviour seen as being generated from their inspection.
     
     
     

    This page is maintained by Richard Griffiths and does not necessarily reflect the official position of the University of Brighton.


    Index