Psychology

Embracing multiple definitions of learning

Andrew B. Barron1*, Eileen A. Hebets2*, Thomas A. Cleland3, Courtney L. Fitzpatrick4, Mark E. Hauber5, and Jeffrey R. Stevens6,7

1 Macquarie University, Department of Biological Sciences, North Ryde, NSW 2109, Australia

2 School of Biological Sciences, University of Nebraska-Lincoln, NE 68588, USA

3 Department of Psychology, Cornell University, Ithaca, NY 14853, USA

4 National Evolutionary Synthesis Center, Durham, NC, 27708, USA

5 Department of Psychology, Hunter College and the Graduate Center, City University of New York, NY 10065, USA

6 Department of Psychology, University of Nebraska-Lincoln, NE 68588, USA

7 Center for Brain, Biology and Behavior, University of Nebraska-Lincoln, NE 68588, USA

Forum

Definitions of learning vary widely across disciplines, driven largely by different approaches used to assess its occurrence. These definitions can be better reconciled with each other if each is recognized as coherent with a common conceptualization of learning, while appreciat- ing the practical utility of different learning definitions in different contexts.

The challenges of defining learning Learning is a major focus of research in psychology, neuro- science, behavioral ecology, evolutionary theory, and com- puter science, as well as in many other disciplines. Despite its conceptual prevalence, definitions of learning differ enor- mously both within and between these disciplines, and new definitions continue to be proposed [1]. Ongoing disputes over the definition of learning generate uncertainty regard- ing the boundaries of the learning concept and confuse assessments about which phenomena genuinely constitute learning. These disputes impair transdisciplinary collabo- ration and synthesis between conceptually related fields. Many of the definitions in use by these different disciplines, however, can be aligned with a common ‘umbrella concept’ of learning that can be applied across disciplines by consider- ing learning simply as the processing of information derived from experience to update system properties [2–5]. Many of the definitions also have clear practical utility in that they reflect a variety of approaches to determine whether or how learning has occurred. We argue that embracing the multi- ple definitions defined by individual subfields (Table S1 in the supplementary material online) – while simultaneously recognizing their shared relationship to this umbrella con- cept – will facilitate the integration of neurophysiological, psychological, computational, and evolutionary approaches to learning.

The difficulty of establishing a single satisfactory scien- tific definition for learning has long been recognized [6]. Perhaps owing to this difficulty, many contemporary psychology and neuroscience textbooks avoid defining

0166-2236/

� 2015 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.tins.2015.04.008

Corresponding author: Hauber, M.E. ([email protected]). Keywords: definitions; experience dependence; function; learning; mechanism; plasticity. *Shared first authorship.

learning altogether, preferring instead to explain specific experimental subtypes of learning (such as operant condi- tioning or habituation) for which it is easier to offer an experimentally supported definition (Table S1). A weak- ness of this approach, of course, is that it discourages engagement with the complexity of the learning concept and its manifestations within different areas of study.

While the specific definitions of learning can vary sub- stantially among fields and even within fields (Table S1), most contemporary theoretical considerations of learning view it as a structured updating of system properties based on processing of new information [2–5]. This concept of learning can operate across disciplines. It does not neces- sarily imply specific mental states, cognitive processes, or processing by neurons. It does not limit learning to complex brains: learning can be instantiated in machines or reflex arcs. It emphasizes that learning is not behavioral change; however, changes in behavior, neural systems, or other elements of the performance of a system all can be useful and practical experimental methods to assess whether learning has occurred.

Despite this general underlying conceptual consensus, there is a wide range of highly specified definitions of learning that vary between disciplines. These variations often arise out of the endeavors of the experimental scien- tist. Because learning is a concept of information processing, it can rarely be measured directly: instead, it is often inferred to have taken place by changes in the (biological, artificial, or virtual/computational) system’s properties or performance. For this reason a range of pragmatic defini- tions of learning delimit the concept in such a way that it can be addressed experimentally [1,7]. Many define learning as a change in behavior, and some define learning as changes in the mechanisms that enable behavioral change (Table S1). These pragmatic definitions vary between dis- ciplines and have merit and utility in different experimental circumstances. By appreciating the situational advantages of these different perspectives, and by describing how the term is being employed in a specific context, scholars of learning can minimize confusion within fields of study and facilitate the meaningful translation of studies of learning across the disciplines.

Learning as a change in behavior Learning is commonly defined as behavioral change. Early on, Skinner [6,8], promoted this approach by arguing that,

Trends in Neurosciences, July 2015, Vol. 38, No. 7 405

Forum Trends in Neurosciences July 2015, Vol. 38, No. 7

because learning is usually determined by assessing be- havioral change, defining learning as the behavioral change or altered behavioral outcome per se eliminates the need for speculative inference about (hidden) underly- ing processes. Likewise, De Houwer [1,7] has more recently advocated for defining learning as behavioral change be- cause this ‘functional’ approach is more verifiable and generalizable than mechanistic definitions, which require direct knowledge of internal processes. Similar functional definitions of learning are most common in disciplines that focus on the evolution of behavioral outcomes and their consequences, including evolutionary and ecological re- search (Table S1). For instance, mathematical models of evolution that include changes in behavior due to learning most often take a functional approach and define learning as behavioral change, because – rather than being con- cerned with underlying physiological processes – they are concerned with the ultimate fitness effects of the pheno- typic changes caused by learning. Learning can be modeled simply as non-genetic inheritance (e.g., song learning from parents) [9] or as within-generation plasticity of a behav- ioral phenotype (e.g., song learning from peers) [10]. Nota- bly, while such models make few assumptions about mechanisms, they nonetheless contribute to mechanistic understandings of learning, its ecological distribution, and its evolutionary consequences.

However, defining learning as behavioral change suffers from significant limitations. Domjan [11], for example, has argued that when defining learning as altered behavior, it is both practically and philosophically difficult to disentangle how much of a given behavioral change results from learning and how much may result from other factors, such as altered motivation, physiological changes, or muscle fatigue, matu- ration, or damage [11,12]. For this reason, some definitions of learning require changes in specific physiological mecha- nisms that support learning to clarify the distinction be- tween learning and other possible causes of behavioral change (e.g., spraining an ankle and walking more slowly thereafter) [11]. The limitation of these mechanistic defini- tions is that they require identification and measurement of the underlying physiological mechanisms of learning. Ac- cordingly, such definitions of learning occur frequently in the psychological and neural sciences (Table S1) [5,11].

As an alternative strategy to distinguish the effects of learning from other factors that could affect behavior, authors often attach various riders to behavioral defini- tions of learning to constrain the definition. Many of these qualifiers are negative, yielding lengthy discussions of what forms of behavioral change do not reflect learning. However, the most common positive qualifier is that learn- ing depends on ‘experience’.

Learning and experience Experience is strongly linked to the learning concept be- cause experience is assumed to be the source of the infor- mation that is learned [4,5]. Whereas experience is part of most definitions of learning (Table S1), it is rare to find a scientific definition of experience, or a discussion of what experience is [13]. Furthermore, the definitions that do exist recapitulate the imprecisions of some learning defini- tions. For example, experience has been defined as an

406

environmental event that is perceived by an organism and that can alter behavior [12]. However, the experience of a startling noise may effect a behavioral response with- out this response being considered learning [1]. Thus, learning may depend on experience, but not all experiences will be learned.

Moreover, the requirement that the event must be perceived by the organism to be considered experience has been criticized on functional grounds because it blurs the line between the sensation of detectable environmental events and the inference of cognitive processing [14]. This is particularly problematic for animal behavior research, which frequently assumes, but does not test internal men- tal states and events for non-human animals. These pro- blems are reduced if experience is considered simply as a source of information. Viewed in this way, experience does not presuppose any particular mental events.

Is it necessary to know what has been experienced to claim that learning has occurred? As Rescorla [5,15] has clearly argued, it can be very misleading to assume, rather than test explicitly, what is being learned from any expe- rience. For example, classical conditioning theorists origi- nally considered learning to be a process by which a behavioral response transferred to a conditioned stimulus, whereas the contemporary perspective recognizes classical conditioning as learning the relationship between stimuli [5]: a radical change in perspective regarding what is learned in classical conditioning. For a small number of established laboratory neuroscience protocols with model systems and controlled stimulus presentation, there has been good experimental analysis of what is being learned. For ethological or ecological data about learning in the wild, however, it is often uncertain which environmental events are salient to the animal, which convey information, or precisely what has been learned. Although the terms ‘experience-dependence’, ‘behavioral plasticity’, and ‘in- duced behavioral change’ appear increasingly in place of ‘learning’, we believe this is not constructive. There is no compelling reason to limit the use of ‘learning’ to situations where the nature of the experience is known or assumed. To do so would invite serious errors of interpretation, and inhibit transdisciplinary syntheses of learning by frag- menting the discussion of clearly related phenomena.

An integrative perspective on learning As with other complex concepts such as ‘fitness’ and ‘gene’, there is no single definition of ‘learning’ that can best serve all scientific purposes, or satisfy all fields and researchers. Disciplines differ in their specific definitions of learning for pragmatic reasons, but it is possible to reconcile most of these definitions by reference to a common theoretical framework: learning as a structured updating of system properties based on the processing of new information. Accordingly, acknowledging the different meanings of learning and being clear on how the term is being used in specific studies are the most effective ways to facilitate transdisciplinary research.

Acknowledgments This study was part of a working group on decision making sponsored by the National Evolutionary Synthesis Center (NESCent), National Science

Forum Trends in Neurosciences July 2015, Vol. 38, No. 7

Foundation (NSF; grant EF-412 0905606). Additional funding was provided by NSF IOS-145624 to M.E.H. We thank the leaders and all members of the working group for stimulating discussions, and especially Kim Hoke, Maria Servedio, Rafael Rodriguez, and an anonymous referee for providing comments on this manuscript.

Appendix A. Supplementary data Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.tins.2015.04.008.

References 1 De Houwer, J. et al. (2013) What is learning? On the nature and merits

of a functional definition of learning. Psychon. Bull. Rev. 20, 631–642 2 Gallistel, C.R. (2008) Learning and representation. In Learning Theory

and Behavior (Learning and Memory: A Comprehensive Reference, Vol.1) (Menzel, R., ed.), pp. 227–242, Elsevier

3 Rudy, J.W. (2008) The Neurobiology of Learning and Memory, Sinauer Associates

4 Kandel, E.R. et al. (2000) Principles of Neural Science, McGraw-Hill 5 Rescorla, R.A. (1988) Behavioral studies of Pavlovian conditioning.

Annu. Rev. Neurosci. 11, 329–352

6 Skinner, B.F. (1950) Are theories of learning necessary? Psychol. Rev. 57, 193–216

7 De Houwer, J. (2011) Why the cognitive approach in psychology would profit from a functional approach and vice versa. Perspect. Psychol. Sci. 6, 202–209

8 Skinner, B.F. (1938) The Behavior of Organisms: An Experimental Analysis, Appleton-Century-Crofts

9 Olofsson, H. and Servedio, M.R. (2008) Sympatry affects the evolution of genetic versus cultural determination of song. Behav. Ecol. 19, 596–604

10 Cavalli-Sforza, L.L. and Feldman, M.W. (1981) Cultural Transmission and Evolution: A Quantitative Approach, Princeton University Press

11 Domjan, M. (2006) The Principles of Learning and Behavior. (5th edn), Thomson Wadsworth

12 Hall, G. (2003) Psychology of learning. In Encyclopedia of Cognitive Science (Nadel, L., ed.), pp. 837–845, Nature Publishing Group

13 Sitter, R.J. (1999) An Introduction to Animal Behaviour, Brooks/Cole 14 Lachman, S.J. (1997) Learning is a process: toward an improved

definition of learning. J. Psychol. 131, 477–480 15 Rescorla, R.A. (1988) Classical conditioning: it’s not what you think it

is. Am. Psychol. 43, 151–160

407