1. The act or process of induction of a lasting alteration in behaviour or in the behavioural potential, due to the individual's behavioural experience.

2. The *acquisition of information, or the reorganization of information that results in new knowledge.

3. Experience-dependent generation of enduring *internal representations, or lasting modifications in such representations.

The above definitions apply to smart inanimate "systems as well (Moravec 1988; Weng et al. 2001). We limit our discussion, however, to learning in biological organisms with nervous systems. Definition 1 is of the "classical, 'behavioural' type (e.g. Bower and Hilgard 1981). The term 'behavioural experience' in it refers to the wide gamut of sensory, motor, emotional, and cognitive events that take place in a lifetime; the modifier 'behavioural' is introduced to eliminate the need to exclude specifically the experience of disease, injury, and poisoning, which is not traditionally considered to result in learning. Definitions 2 and 3 are variants of the 'informational' type, which refers to the behavioural "performance and behavioural capacity of the organism in terms of 'knowledge' (Plato; James 1890; Squire 1987). Definition 3 (Dudai 1989, 1992) expresses information in terms of internal representations, which are neuronally encoded structured "models of the world that could potentially guide behaviour. This is the preferred definition of learning in this book. This definition implies that all learning, be it in "Aplysia or human, is alteration in an internal representation of some type or another.

The pursuit of internal representations and their modification by experience is thus identified as the crux of learning research, at all the "levels of analysis of learning. In the "reductive analysis of learning, the focus on representational properties is meant to guide the investigator to identify those changes in one level, e.g. the cellular, that cause or reflect the representational alterations in another level, e.g. the circuit. Changes that do not contribute to the representational alteration are irrelevant to learning per se, although, of course, they may still be critical for other functions of the nervous system, such as "homeostasis. Further, the assumption in this book is that internal representations are encoded in the spatiotemporal activity of neuronal circuits. Hence, molecules, isolated "synapses, and in many cases even individual nerve cells, are not expected to encode independently appreciable chunks of behaviourally meaningful models of the world. In order to gain behavioural meaning, the contribution of the molecular and cellular change must be construed within the "context of the circuit (Dudai 1989,1994&).1

A caveat is appropriate here. Molecular states within an individual nerve cell clearly have a meaning as well. But this meaning is at a level of organization that does not suffice to guide directly behaviour and cognition. 'Meaning' is level-dependent, and levels transmit only limited information to other levels (Simon and Ando 1961).2 Therefore, although states at level Li could embody unique meaning at level Li, these states only provide elementary building blocks, or terms, that are used to construct a variety of meanings at a higher level

Lj. For example, suppose a "neurotransmitter activates its "receptor in a synapse. The downstream cascade culminates in modification of an "ion channel. The identity of the modified channel, determined by the "stimulus and its context, conveys a specific meaning to the synaptic state, e.g. modified channel X=+Ay synaptic excitability. But this altered synaptic state could be employed in different ways to construct meaning at the circuit level; for example, it means something very different if the synapse is inhibitory or excitatory, or, at a higher level, if the circuit enhances or suppresses the behaviour. The contribution ofthe molecular change to the representational meanings ('semantics') hence depends on the synaptic context, and that of the synapse, on the circuit context.

Learning has multiple "dimensions. Here are a selected few:

1. Innateness. Some types of learning involve information that is constrained "a priori by innate predispositions. These types of learning are termed 'prepared learning'. They could be ubiquitous in the animal kingdom, for example, "conditioned taste aversion, or species-specific, for example, filial "imprinting, "bird song. Imprinting and bird-song are good examples for the role of "development in learning. Some behavioural definitions of learning explicitly exclude the role ofrigid, autonomous developmental programmes, which do not require interaction with the environment, in the modification ofbehav-iour. But it is doubtful whether genuine use-independent programs exist in real-life. The demarcation line between 'development' and 'learning' is inherently blurred. The two types of processes share molecular and cellular hardware (e.g. "immediate early genes), and it is possible to consider learning as an extension of brain development. Still, the position of different types of learning on the 'deterministic', 'preparedness' or 'developmental' axes vary. An example is provided by "classical conditioning. Some instances ofclassical conditioning involve only augmentation by experience of the response to the conditioned stimulus. This is called a-conditioning ("Aplysia). In other instances there is no significant pre-conditioning response to the conditioned stimulus. This is bona fide classical conditioning. These two types of classical conditioning are hence separable on the axis of'preparedness'.

2. Strategy. There are two major strategies by which a 'teacher' stimulus could modify internal representations (Young 1979; Changeux 1985; Edelman 1987; Dudai 1989). First, the teacher could impose new order in the system by directly instructing it to modify in a certain way. Secondly, it could induce the new structure by selecting an internal representation among multiple endogenous variations, i.e. existing 'pre-representations'. The instructive and the selective mechanisms of learning could coexist.

3. Domain. Learning may involve the acquisition of motor, sensory, emotional, or cognitive information, or to all of the above.

4. Associativity. Certain types of learning are governed solely by the parameters of the unconditioned stimulus. These types of learning therefore do not result in the association of the unconditioned stimulus with other stimuli. Examples are provided by "habituation and "sensitization.3 Most types of learning involve the formation of association among stimuli or among stimuli and actions. Examples are provided by classical and "instrumental conditioning.

5. Specificity. Types and instances of learning differ in the specificity of the acquired information ("generalization, "transfer).

6. Intention. We learn about the world either incidentally or intentionally. The term 'incidental learning' has come with time to acquire multiple meanings (Hilgard and Marquis 1940; Spence et al. 1950; Morton 1967; Hyde and Jenkins 1973; Craik and Tulving 1975; Glass and Holyoak 1986; Rugg et al. 1997; Berman et al. 1998). These are: (a) learning that occurs unintentionally as a by-product of a sensory, motor, or cognitive process; (b) learning in the absence of "attention; (c) learning in the absence of an identified "reinforcer; and (d) an experimental situation in which the "subject is not told that memory would be tested later. Note that in (c) the lack of an identified "reinforcer is used as a "criterion; however, in real life, the reinforcer is always there, only hidden, either in the context or in the endogenous activity of brain circuits.

7. Awareness. The presence or absence of "conscious awareness is a major criterion in the "zeitgeist "taxonomy of learning and memory. Suffice it to note in the present context that conscious awareness in learning does not entail conscious awareness in retention and "retrieval, and vice versa ("declarative memory). When information is acquired in an incidental manner, without awareness of what has been learned, the process is termed 'implicit learning', as opposed to 'explicit learning' (Seger 1994; Whittlesea and Wright 1997). The distinction between 'implicit' and 'explicit' learning has been used extensively in tasks involving rule learning in humans, e.g. grammar learning. In these experiments 'explicit' came to mean that deliberate instructions are given to search for rules that underlie the presented material, whereas 'implicit' is when the subject learns without such instructions but acquires information about the underlying rules nevertheless (Reber 1967; Berry and Broadbent 1988). The term 'latent learning' is occasionally applied to either incidental or implicit learning (Stevenson 1954), but this is not recommended, because 'latency' does not necessarily imply neither incidentality nor implicitness of learning (Dudai 1989; "insight).

8. Novelty. Certain types or tokens of information are unexpected, others are. A useful rule of thumb is that the more "surprising the information, the better it is learned ("algorithm). The novelty dimension of the stimulus to be learned should not, however, be confused with naivete of the subject. Even when the role of innateness is recognized (1 above), many investigators still err to think that the subject's brain enters the new experimental situation as a blank surface (tabula rasa, Locke 1690). This rarely is the case; almost always the subject brings to the task knowledge and expectations ("a priori). This is now evident even at the cellular level. For example, whether a modest input induces a long-term change in the target neuron ("long-term potentiation) depends on what the same cell has experienced 2-3 h before (Frey and Morris 1998). It is hence appropriate to consider even new learning experiments as manipulations of an already opinionated brain ("palimpsest).

9. Rate. Certain types or instances of learning occur in a single trial, as a step function ("flashbulb memory). Others are incremental and require repetitive training. An example for the latter is rote learning (Hebb 1949; Irion 1959), manifested in the acquisition of "skill. The kinetics of learning is commonly depicted as a learning curve. This is the representation of performance, which itself is taken to represent learning, as a function of the amount of experience (e.g. Figure 41, p. 144).

10. Fate. Depending on its type and on the parameters of acquisition, learning could result in "engrams that last from seconds up to engrams that last for a lifetime ("percept, "consolidation, "taxonomy). Furthermore, in some types of learning the information is a priori intended to subserve only the transient task and then be "forgotten ("working memory).

The multiplicity of dimensions clearly hints at the richness of the computational theories, neuronal algorithms and their cellular and molecular implementation, which one should expect to identify in neuronal circuits that learn. It also implies that a master solution to the mechanisms of learning is unlikely to exist.

Selected associations: Acquisition, Engram, Development, Internal representation, Memory

''Although there is no doubt that the contribution of individual neurons to a representation must be evaluated in the context of the entire circuit, the role of single neurons in encoding representations is still unsettled. In any case, this role is circuit and task dependent. Single neurons may execute meaningful computations, but it is unlikely that they encode meaningful parts of complex representations. For further discussion, including estimates of the number of neurons that encode a representation, see *cell assembly.

theoretically, one might envisage a computational system with practically unlimited 'capacity that could read simultaneously all the information in all the levels of a biological system. Such a system is however impractical, and, most importantly, disposes of the advantage of being able to save computational resources by using only the important information extracted at each level. 3It is doubtful, however, whether pure nonassociativity exists in nature; see discussion in 'habituation.

Was this article helpful?

0 0

Post a comment