Connectionism (Edward L. Thorndike – 1898)
The prominent role of Aristotle’s laws of association in the 1900s may largely be due to the work of Edward L. Thorndike—the recognized founder of a “learning theory [that] dominated all others in America” for “nearly half a century” (Bower & Hilgard, 1981, p. 21). Thorndike’s theory was based initially on a series of puzzle box experiments that he used to plot learning curves of animals. In these experiments learning was defined as a function of the amount of time required for the animal to escape from the box. A full account of his experiments, including detailed descriptions of the puzzle boxes he used and examples of learning curves that were plotted, can be found in Animal intelligence (Thorndike, 1898).
In Thorndike’s view, learning is the process of forming associations or bonds, which he defined as “the connection of a certain act with a certain situation and resultant pleasure” (p. 8). His work leading up to 1898 provided “the beginning of an exact estimate of just what associations, simple and compound, an animal can form, how quickly he forms them, and how long he retains them” (p. 108).
Although his original experimental subjects were cats, dogs, and chicks, Thorndike clearly expressed his intention of applying his work to human learning when he said, “the main purpose of the study of the animal mind is to learn the development of mental life down through the phylum, to trace in particular the origin of human faculty” (1898, p. 2). From his work with animals he inferred “as necessary steps in the evolution of human faculty, a vast increase in the number of associations” (p. 108). A decade and a half later he expanded on the theme of human learning in a three volume series entitled, Educational psychology, with volume titles, The original nature of man (1913a), The psychology of learning (1913b), and Mental work and fatigue and individual differences and their causes (1914b). The material in these books was very comprehensive and targeted advanced students of psychology. He summarized the fundamental subject matter of the three volumes in a single, shorter textbook entitled, Educational psychology: briefer course (Thorndike, 1914a). In these volumes Thorndike provided a formative culmination of his theory of learning in the form of three laws of learning:
1. Law of Readiness – The law of readiness was intended to account for the motivational aspects of learning and was tightly coupled to the language of the science of neurology. It was defined in terms of the conduction unit, which term Thorndike (1914a) used to refer to “the neuron, neurons, synapse, synapses, part of a neuron, part of a synapse, parts of neurons or parts of synapses—whatever makes up the path which is ready for conduction” (p. 54). In its most concise form, the law of readiness was stated as follows, “for a conduction unit ready to conduct to do so is satisfying, and for it not to do so is annoying” (p. 54). The law of readiness is illustrated through two intuitive examples given by Thorndike:
The sight of the prey makes the animal run after it, and also puts the conductions and connections involved in jumping upon it when near into a state of excitability or readiness to be made….When a child sees an attractive object at a distance, his neurons may be said to prophetically prepare for the whole series of fixating it with the eyes, running toward it, seeing it within reach, grasping, feeling it in his hand, and curiously manipulating it. (p. 53)
2. Law of Exercise – The law of exercise had two parts: (a) the law of use and (b) the law of disuse. This law stated that connections grow stronger when used—where strength is defined as “vigor and duration as well as the frequency of its making” (p. 70)—and grow weaker when not used.
3. Law of Effect – The law of effect added to the law of exercise the notion that connections are strengthened only when the making of the connection results in a satisfying state of affairs and that they are weakened when the result is an annoying state of affairs.
These three laws were supplemented by five characteristics of learning “secondary in scope and importance only to the laws of readiness, exercise, and effect” (Thorndike, 1914a, p. 132). They are
1. Multiple response or varied reaction – When faced with a problem an animal will try one response after another until it finds success.
2. Set or attitude – The responses that an animal will try, and the results that it will find satisfying, depend largely on the animal’s attitude or state at the time.
The chick, according to his age, hunger, vitality, sleepiness, and the like, may be in one or another attitude toward the external situation. A sleepier and less hungry chick will, as a rule, be ‘set’ less toward escape-movements when confined; its neurons involved in roaming, perceiving companions and feeding will be less ready to act; it will not, in popular language, ‘try so hard to’ get out or ‘care so much about’ being out. (Thorndike, 1914a, p. 133)
3. Partial activity or prepotency of elements – Certain features of a situation may be prepotent in determining a response than others and an animal is able to attend to critical elements and ignore less important ones. This ability to attend to parts of a situation makes possible response by analogy and learning through insight.
Similarly, a cat that has learned to get out of a dozen boxes—in each case by pulling some loop, turning some bar, depressing a platform, or the like—will, in a new box, be, as we say, ‘more attentive to’ small objects on the sides of the box than it was before. The connections made may then be, not absolutely with the gross situation as a total, but predominantly with some element or elements of it. (Thorndike, 1914a, p. 134)
4. Assimilation – Due to the assimilation of analogous elements between two stimuli, an animal will respond to a novel stimulus in the way it has previously responded to a similar stimulus. In Thorndike’s words, “To any situations, which have no special original or acquired response of their own, the response made will be that which by original or acquired nature is connected with some situation which they resemble.” (Thorndike, 1914a, p. 135)
5. Associative shifting – Associative shifting refers to the transfer of a response evoked by a given stimulus to an entirely different stimulus.
The ordinary animal ‘tricks’ in response to verbal signals are convenient illustrations. One, for example, holds up before a cat a bit of fish, saying, “Stand up.” The cat, if hungry enough, and not of fixed contrary habit, will stand up in response to the fish. The response, however, contracts bonds also with the total situation, and hence to the human being in that position giving that signal as well as to the fish. After enough trials, by proper arrangement, the fish can be omitted, the other elements of the situation serving to evoke the response. Association may later be further shifted to the oral signal alone. (Thorndike, 1914a, p. 136)
Sixteen years after publishing his theory in the Educational Psychology series based on experiments with animals, Thorndike published twelve lectures that reported on experiments performed with human subjects between 1927 and 1930 (see Thorndike, 1931). The results of these experiments led Thorndike to make some modifications to his laws of connectionism.
The first change was to qualify the law of exercise. It was shown that the law of exercise, in and of itself, does not cause learning, but is dependent upon the law of effect. In an experiment in which subjects were blindfolded and repeatedly asked to draw a four-inch line with one quick movement Thorndike discovered that doing so 3,000 times “caused no learning” because the lines drawn in the eleventh or twelfth sittings were “not demonstrably better than or different from those drawn in the first or second” (Thorndike, 1931, p. 10). He summarized this finding by saying,
Our question is whether the mere repetition of a situation in and of itself causes learning, and in particular whether the more frequent connections tend, just because they are more frequent, to wax in strength at the expense of the less frequent. Our answer is No. (p. 13)
However, in drawing this conclusion, Thorndike was not disproving the law of exercise, but merely qualifying it (by saying that repetition must be guided by feedback):
It will be understood, of course, that repetition of a situation is ordinarily followed by learning, because ordinarily we reward certain of the connections leading from it and punish others by calling the responses to which they respectively lead right or wrong, or by otherwise favoring and thwarting them. Had I opened my eyes after each shove of the pencil during the second and later sittings and measured the lines and been desirous of accuracy in the task, the connections leading to 3.8, 3.9, 4.0, 4.1, and 4.2 would have become more frequent until I reached my limit of skill in the task. (p. 12-13)
The second change was to recast the relative importance of reward and punishment under the law of effect. Through a variety of experiments Thorndike concluded that satisfiers (reward) and annoyers (punishment) are not equal in their power to strengthen or weaken a connection, respectively. In one of these experiments students learned Spanish vocabulary by selecting for each Spanish word one of five possible English meanings followed by the rewarding feedback of being told “Right” or the punishing feedback of being told “Wrong.” From the results of this experiment Thorndike concluded that punishment does not diminish response as originally stated in the law of effect. In his own words,
Indeed the announcement of “Wrong” in our experiments does not weaken the connection at all, so far as we can see. Rather there is more gain in strength from the occurrence of the response than there is weakening by the attachment of “Wrong” to it. Whereas two occurrences of a right response followed by “Right” strengthen the connection much more than one does, two occurrences of a wrong response followed by “Wrong” weaken that connection less than one does. (p. 45)
In another experiment a series of words were read by the experimenter. The subject responded to each by stating a number between 1 and 10. If the subject picked the number the experimenter had predetermined to be “right” he was rewarded (the experimenter said “Right”), otherwise he was punished (the experimenter said “Wrong”). Other than the feedback received from the experimenter, the subject had no logical basis for selecting one number over another when choosing a response. Each series was repeated many times, however, the sequence of words was long, making it difficult for the subject to consciously remember any specific right and wrong word-number pairs. From the results of this and other similar experiments Thorndike demonstrated what he called the “spread of effect.” What he meant by this was that “punished connections do not behave alike, but that the ones that are nearest to a reward are strengthened” and that “the strengthening influence of a reward spreads to influence positively not only the connection which it directly follows…but also any connections which are near enough to it” (Thorndike, 1933, p. 174). More specifically,
A satisfying after-effect strengthens greatly the connection which it follows directly and to which it belongs, and also strengthens by a smaller amount the connections preceding and following that, and by a still smaller amount the preceding and succeeding connections two steps removed. (p. 174)
In addition to these two major changes to the law of exercise and the law of effect, Thorndike also began to explore four other factors of learning that might be viewed as precursors to cognitive learning research, which emerged in the decades that followed. They are summarized by Bower and Hilgard (1981):
- Belongingness – “a connection between two units or ideas is more readily established if the subject perceives the two as belonging or going together” (p. 35).
- Associative Polarity – “connections act more easily in the direction in which they were formed than in the opposite direction” (p. 35). For example, if when learning German vocabulary a person always tests themselves in the German-to-English direction it is more difficult for them to give the German equivalent when prompted with an English word than to give the English word when prompted with the German equivalent.
- Stimulus Identifiability – “a situation is easy to connect to a response to the extent that the situation is identifiable, distinctive, and distinguishable from others in a learning series” (p. 36).
- Response Availability – the ease of forming connections is directly proportional to the ease with which the response required by the situation is summoned or executed:
Some responses are overlearned as familiar acts (e.g., touching our nose, tapping our toes) which are readily executed upon command, whereas more finely skilled movements (e.g., drawing a line 4 inches as opposed to 5 inches long while blindfolded) may not be so readily summonable. (p.36-37)