Where does learning collide?

Machine learning

Learning as improvement on a task with experience

A program learns from experience E with respect to a task T and a performance measure P if its performance on T, as measured by P, improves with E. Mitchell fixed this operational form in 1997; Valiant had already made it precise in 1984 by asking what can be learned from examples in polynomial time to within a chosen error, with a chosen confidence. The learner is an algorithm, the thing learned is a function, and the criterion is error on data it has not seen.[1][2]

Unit: An algorithm, on a task, over a data distribution
Recovery means: Not applicable, a model that degrades is retrained, not recovered
Measured by: Held-out error or loss; sample complexity; PAC bounds

Behavioural psychology

Learning as change in behaviour from contingencies

A relatively durable change in behaviour that results from experience with the contingencies between stimuli, responses, and outcomes. Pavlov's dogs learned that a tone predicted food; Skinner's rats learned that a lever press produced it, and Skinner insisted that the rate of the response was the datum, not any inner state. Rescorla and Wagner made the classical case quantitative: the associative strength of a cue changes in proportion to how surprising the outcome was, the error term that reinforcement learning later borrowed.[3][4][5]

Unit: An organism, over trials
Recovery means: Spontaneous recovery, a conditioned response returns after extinction without further training
Measured by: Response rate or probability across trials; conditioned-response magnitude; acquisition and extinction curves

Neuroscience

a lasting change in the strength of connections between neurons, produced by their activity.

A lasting change in the strength of connections between neurons, produced by their activity. Hebb's 1949 postulate (cells that fire together wire together, in its later slogan form) gave learning a physical mechanism before one could be observed. Kandel's work on Aplysia showed that short-term learning modifies existing synapses while long-term learning requires gene expression and new synaptic growth. Behaviour is the readout; the construct is the change in the tissue.[6][7]

Unit: A synapse, circuit, or organism
Recovery means: Reconsolidation or relearning after lesion; savings on retraining
Measured by: Long-term potentiation; synaptic strength; memory retention in a behavioural assay

Education & learning sciences

a person's construction of understanding usable in new situations, achieved in a social setting and tested by transfer.

The construction of understanding that can be used in new situations, achieved by a person in a social and cultural setting, not by a passive receiver of instruction. Vygotsky located it in the zone of proximal development, what a learner can do with help today and alone tomorrow. The National Research Council's How People Learn synthesized the cognitive account: learning with understanding means organized knowledge, metacognition, and transfer. Transfer, not recall, is the test, which puts this meaning at odds with the behaviourist one.[8][9]

Unit: A learner, in a setting, over time
Recovery means: Not applicable, remediation and re-teaching, not recovery
Measured by: Transfer tasks; conceptual assessments; gains over instruction; the ZPD is observed, not scored

Organizational learning

Learning as change in an organization's routines and knowledge

An organization learns when the processing of information changes the range of its potential behaviours, Huber's 1991 definition, which does not require that the organization behave better, or even differently. Argyris and Schön separated single-loop learning, which corrects errors within existing goals, from double-loop learning, which revises the goals. March showed the trade-off that follows: exploiting what the organization already knows drives out exploration of what it does not, and adaptive systems drift toward the former.[10][11][12]

Unit: An organization, its routines and memory
Recovery means: Unlearning, discarding obsolete routines; recovery from competency traps
Measured by: Changes in routines, procedures, and shared mental models; survey scales of learning orientation; mostly qualitative

Economics & operations

Learning as falling unit cost with cumulative output

The regularity that unit cost or labour hours fall by a constant percentage every time cumulative output doubles. Wright observed it in airframe assembly in 1936 and wrote it as a power law; an 80 percent curve means the fourth unit costs 80 percent of the second. Arrow made it a theory of growth in 1962: productivity rises as a by-product of production itself, so investment teaches. Nobody in this meaning asks what anyone came to know; learning is the slope of the cost curve.[13][14]

Unit: A plant, firm, or industry, over cumulative units
Recovery means: Not applicable, forgetting curves model the reverse after interruptions
Measured by: Progress ratio; elasticity of cost with respect to cumulative output

References

Mitchell, T. M. (1997). Machine learning. McGraw-Hill. ISBN 9780070428072
Valiant, L. G. (1984). A theory of the learnable. Communications of the ACM, 27(11), 1134–1142. https://doi.org/10.1145/1968.1972
Pavlov, I. P. (1927). Conditioned reflexes: An investigation of the physiological activity of the cerebral cortex (G. V. Anrep, Trans.). Oxford University Press.
Skinner, B. F. (1938). The behavior of organisms: An experimental analysis. Appleton-Century.
Rescorla, R. A., & Wagner, A. R. (1972). A theory of Pavlovian conditioning: Variations in the effectiveness of reinforcement and nonreinforcement. In A. H. Black & W. F. Prokasy (Eds.), Classical conditioning II: Current research and theory (pp. 64–99). Appleton-Century-Crofts.
Hebb, D. O. (1949). The organization of behavior: A neuropsychological theory. Wiley.
Kandel, E. R. (2001). The molecular biology of memory storage: A dialogue between genes and synapses. Science, 294(5544), 1030–1038. https://doi.org/10.1126/science.1067020
Vygotsky, L. S. (1978). Mind in society: The development of higher psychological processes (M. Cole, V. John-Steiner, S. Scribner, & E. Souberman, Eds.). Harvard University Press.
National Research Council. (2000). How people learn: Brain, mind, experience, and school (Expanded ed., J. D. Bransford, A. L. Brown, & R. R. Cocking, Eds.). National Academies Press. https://doi.org/10.17226/9853
Argyris, C., & Schön, D. A. (1978). Organizational learning: A theory of action perspective. Addison-Wesley.
Huber, G. P. (1991). Organizational learning: The contributing processes and the literatures. Organization Science, 2(1), 88–115. https://doi.org/10.1287/orsc.2.1.88
March, J. G. (1991). Exploration and exploitation in organizational learning. Organization Science, 2(1), 71–87. https://doi.org/10.1287/orsc.2.1.71
Wright, T. P. (1936). Factors affecting the cost of airplanes. Journal of the Aeronautical Sciences, 3(4), 122–128. https://doi.org/10.2514/8.155
Arrow, K. J. (1962). The economic implications of learning by doing. Review of Economic Studies, 29(3), 155–173. https://doi.org/10.2307/2295952