ReviewPerceptual learning and human expertise
Introduction
On a good day, the best human chess grandmaster can beat the world's best chess-playing computer. The computer program is no slouch; every second, it examines upwards of 200 million possible moves. Its makers incorporate sophisticated methods for evaluating positions, and they implement strategies based on advice from grandmaster consultants. Yet, not even this formidable array of techniques gives the computer a clear advantage over the best human player.
If chess performance were based on raw search, the human would not present the slightest problem for the computer. Estimates of human search through possible moves in chess suggest that even the best players examine on the order of 4 possible move sequences, each about 4 plies deep (where a ply is a pair of turns by the two sides). That estimate is per turn, not per second, and a single turn may take many seconds. Assuming the computer were limited to 10 sec of search per turn, the human would be at a disadvantage of about 1 999 999 984 moves searched per turn.
Given this disparity, how is it possible for the human to outplay the machine? The accomplishment suggests information processing abilities of remarkable power but mysterious nature. Whatever the human is doing, it is, at its best, roughly equivalent to 2 billion moves per sec of raw search. “Magical” would not seem too strong a description for such abilities.
We have not yet said what these abilities are, but before doing so, we add one more striking observation. Biological systems often display remarkable structures and capacities that have emerged as evolutionary adaptations to serve particular functions. Compared to machines that fly, for example, the capabilities of a dragonfly or hummingbird (or even the lowly mosquito) are astonishing. Yet the information processing capabilities we are considering may be seen as all the more remarkable because they do not appear to be adaptations specialized for one particular task. We did not evolve to play chess. In other words, it is likely that human attainments in chess are consequences of highly general abilities that contribute to learned expertise in many domains. Such abilities may have evolved for more ecological tasks, but they are of such power and generality that humans can become remarkably good in almost any domain involving complex structure.
What abilities are these? They are abilities of perceptual learning. The effects we are describing arise from experience-induced changes in the way perceivers pick up information. With practice in any domain, humans become attuned to the relevant features and structural relations that define important classifications, and over time we come to extract these with increasing selectivity and fluency. As a contrast, consider: Most artificial sensing devices that exist, or those we might envision, would have fixed characteristics. If they functioned properly, their performance on the 1000th trial of picking up some information would closely resemble their performance on the first trial. Not so in human perception. Rather, our extraction of information changes adaptively to optimize particular tasks. A large and growing research literature suggests that such changes are pervasive in perception and that they profoundly affect tasks from the pickup of minute sensory detail to the extraction of complex and abstract relations that underwrite symbolic thought. Perceptual learning thus furnishes a crucial basis of human expertise, from accomplishments as commonplace as skilled reading to those as rarified as expert air traffic control, radiological diagnosis, grandmaster chess, and creative scientific insight.
In this paper, we give an overview of perceptual learning, a long-neglected area of learning, both in scientific theory and research, as well as in educational practice. Our consideration of perceptual learning will proceed as follows. In the second section, we provide some brief historical background on perceptual learning and some taxonomic considerations, contrasting and relating it to other types of learning. In the third section, we consider some instructive examples of perceptual learning, indicating its influence in a range of levels and tasks, and arguing that the information processing changes brought about by perceptual learning can be usefully categorized as discovery and fluency effects. In the fourth section, we consider explanations and modeling concepts for perceptual learning, and we use this information to consider the scope of perceptual learning in the fifth section. As its role and scope in human expertise become clearer, its absence from conventional instructional settings becomes more paradoxical. In a final section, we discuss these issues and the potential for improving education by using perceptual learning techniques.
Section snippets
Perceptual learning and taxonomies of learning
Perceptual learning can be defined as “an increase in the ability to extract information from the environment, as a result of experience and practice with stimulation coming from it” [34, p. 3]. With sporadic exceptions, this kind of learning has been neglected in scientific research on learning. Researchers in animal learning have focused on conditioning – associative learning phenomena – involving connections between responses and stimuli. Most work on human learning and memory has focused on
Some perceptual learning phenomena
To give some idea of its scope and characteristics, we consider a few examples of perceptual learning phenomena. Our examples come primarily from visual perception. They span a range of levels, from simple sensory discriminations to higher-level perceptual learning effects more relevant to real-world expertise.
Highlighting these examples may also illustrate that perceptual learning phenomena can be organized into two general categories – discovery and fluency effects [58]. Discovery involves
Explaining and modeling perceptual learning phenomena
How have researchers sought to explain and model perceptual learning? These questions are important not only for understanding human performance but for artificial systems as well. Understanding how learners discover invariance among variable instances would have value for creating learning devices as well as explaining human abilities. We currently have no good machine learning algorithms that can learn from several examples to correctly classify new instances of dogs, cats, and squirrels the
The scope of perceptual learning
It is apparent from our discussion that perceptual learning encompasses several information processing changes. These are linked by our general definition of perceptual learning as comprising improvement in the pickup of information. There appear to be different ways to improve: coming to select relevant features for some classification, discovering higher-order invariants to which the perceiver is initially insensitive, and becoming more fluent or automatic in information pickup.
It would be
Perceptual learning and instruction
If one consults the educational literature about the cognitive bases of human learning, one would find ample treatment of fact and concept learning, conceptual understanding, procedure learning, constructing explanations, thinking and reasoning. Except for an occasional mention of pattern recognition, perceptual learning would be starkly absent. Yet it is arguably one of the most, possibly the most, important component of human expertise.
If one looks instead at research, not on education, but
Acknowledgements
We gratefully acknowledge support from the US Department of Education, Institute for Education Sciences, Cognition and Student Learning Program Grant R305H060070 and by National Science Foundation Grant REC-0231826 to PK. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the US Department of Education or the National Science Foundation. We thank Christine Massey for helpful discussions and
References (131)
- et al.
Learning pop-out detection: The spread of attention and learning in feature search: effects of target distribution and task difficulty
Vision Research
(2000) - et al.
The reverse hierarchy theory of visual perceptual learning
Trends in Cognitive Sciences
(2004) - et al.
Learning pop-out detection: Building representations for conflicting target-distractor relationships
Vision Research
(1998) - et al.
Perception in chess
Cognitive Psychology
(1973) - et al.
Categorization and representation of physics problems by experts and novices
Cognitive Science
(1981) - et al.
Long-term learning in vernier acuity: Effects of stimulus orientation, range and of feedback
Vision Research
(1993) - et al.
Fast perceptual learning in hyperacuity
Vision Research
(1995) - et al.
Infant artificial language learning and language acquisition
Trends in Cognitive Sciences
(2000) - et al.
Stochastic resonance in visual cortical neurons: Does the eye tremor actually improve visual acuity?
Neurocomputing
(2002) - et al.
Perception of partly occluded objects in infancy
Cognitive Psychology
(1983)