Edward Thorndike

Introduction

Edward Lee Thorndike (31 August 1874 – 9 August 1949) ran the first controlled laboratory experiments on animal learning, and the theory that came out of them — that learning is the stamping in of connections between situations and responses, by trial and error, under the governance of consequences — supplied half the apparatus of behaviourism.

He built a puzzle box, put a hungry cat in it, put food outside, and timed how long the cat took to escape. The cat did not reason its way out. It clawed and scrambled until it happened to hit the mechanism, and on each subsequent trial it hit the mechanism a little sooner. From this Thorndike concluded that learning is not insight but the gradual strengthening of a connection, by the operation of what he called the law of effect: responses followed by satisfaction get stamped in.

He then did three things with it. He spent the rest of his career applying the same experimental logic to human learning and schooling, effectively founding educational psychology as a separate discipline. He used it to dismantle the dominant theory of what a curriculum is for, in one of the most consequential experiments in the history of American education. And he revised his own central law when the evidence required it — abandoning one of his three laws entirely and cutting the other down.

He was also a committed eugenicist who wrote, in his own words, about denying life to "inferior men." That is not incidental to his work and is dealt with below.

Early Life and Education

Thorndike was born in Williamsburg, Massachusetts, the son of a lawyer turned minister. He was a shy, solitary child who, by his own account, found satisfaction mainly in schoolwork and did well at it.

He entered Wesleyan University in 1891 to read English and graduated in 1895 with what was then the highest grade average in fifty years. He claimed not to have encountered the word psychology before his junior year, when a required course put William James's Principles of Psychology in front of him. It decided the direction of his life.

He went to Harvard in 1895, took a second bachelor's degree in 1896 and a master's in 1897, and — having been captivated by James's teaching — proposed a thesis on the instinctive and intelligent behaviour of chicks. James had stopped doing experimental work but agreed to the topic and gave Thorndike space in his own cellar, because the laboratory was not large enough to house animals.

Thorndike later said, with characteristic bluntness, that his motive had been "chiefly to meet the requirements for credits and a degree" and that he "had no special interest in animals" at the time. The most important experiment in the history of learning theory was begun as a way of getting a qualification.

Academic and Professional Career

In 1898 James McKeen Cattell, visiting Harvard, was impressed by the work and invited Thorndike to Columbia with a scholarship. He arrived in New York with two of his best-trained chickens and continued with cats and dogs, completing his doctorate in 1898 under Cattell with the dissertation that became Animal Intelligence (1911).

He joined Columbia Teachers College as an instructor in psychology in 1899 and was made professor in 1903. Apart from one year at Western Reserve University, he spent the rest of his career there, retiring in 1940.

He was elected president of the American Psychological Association in 1912 and to the National Academy of Sciences in 1917. He published around five hundred books, monographs, and papers — a volume of output exceeded among psychologists, by most counts, only by Piaget.

Major Contributions to Psychology

The puzzle box and against the anecdotal method

When Thorndike began, comparative psychology ran largely on the anecdotal method: collecting stories of animal behaviour and inferring the animal's conscious reasoning from them. Romanes's accounts of reasoning dogs and thoughtful cats were the standard fare.

Thorndike's objection was that the method had no controls, systematically overestimated animal intelligence, and interpreted any behaviour as evidence of thought. He replaced stories with measurement. He built apparatus — mazes, puzzle boxes — in which the animal's response could be timed and counted across trials, and from which one could tell whether improvement was gradual or sudden.

The result was the first laboratory paradigm for instrumental learning, and it is the direct ancestor of B.F. Skinner's operant conditioning. Pavlov, working at the same time, established the other paradigm. Between them they supplied behaviourism's methods.

Connectionism

Thorndike called his theory connectionism, deliberately avoiding "association," which carried the baggage of associating ideas. His unit was the connection between a situation and a response, established by trial and error: unsuccessful responses are gradually stamped out, the successful one is stamped in.

The point of the distinction was methodological. He wanted a psychology whose terms were observable — situations and responses rather than contents of mind — and this is why he is usually placed on the road from functionalism to behaviourism even though he never joined the behaviourist programme and disliked its refusal to discuss consciousness at all.

The three laws

The law of readiness held that conducting a neural unit ready to conduct is satisfying, and forcing one not ready to conduct is annoying. The law of exercise held that connections are strengthened by use and weakened by disuse. The law of effect — the one that mattered — held that a response followed by satisfaction strengthens the connection, and one followed by discomfort weakens it.

Thorndike emphasised that the satisfaction must follow the response closely, or the lesson does not take.

The 1930s revision

He abandoned the law of exercise. Repetition alone, he concluded, does not strengthen a connection; practice works only in combination with effect. This corrected a real error and anticipated the later finding that practice without feedback produces little or no learning.

He also cut the law of effect down. Punishment, he found, does not weaken a connection — it suppresses its expression, and only slightly. Reward genuinely strengthens; punishment does not symmetrically weaken. This is still the received view, and it is the reason modern behavioural practice leans on reinforcement rather than penalty.

The destruction of formal discipline

The most consequential thing Thorndike did to American schooling was not a positive proposal but a demolition. The prevailing doctrine, inherited from faculty psychology and defended well into the twentieth century, was formal discipline: the mind has faculties, difficult subjects train them as exercise trains a muscle, and so Latin and geometry make a person reason better generally. It was the intellectual justification for the classical curriculum.

Thorndike attacked it experimentally, first with Robert Woodworth in 1901 and then in a much larger study of his own. The first round was narrow and precise: practice estimating the areas of rectangles did not improve estimation of other shapes, however similar. Two decades later he went at the doctrine directly. Working with C. R. Broyler and Ella Woodyard, he gave standardised tests to high-school students in autumn and again in spring and asked whether the improvement a student showed depended on which subjects they had taken. If Latin makes you clever, then students who take Latin should gain more than comparable students who do not. They did not.

He replaced the doctrine with the theory of identical elements: transfer occurs only to the extent that the two situations share specific components, so the practical prescription is to teach things close to what the learner will actually have to do. Thorndike urged schools to design curricula around tasks resembling those outside school rather than around subjects prized for their presumed disciplinary effect.

The result was a revolution. Within a generation the classical curriculum in American schools had lost its scientific defence, and historians of education have credited Thorndike's findings with opening the way to the vocational and practical emphasis of twentieth-century American schooling. Whether that was entirely a gain is discussed below.

Educational psychology, measurement, and the counting of words

From 1903 onward Thorndike worked to make education an empirical discipline. His three-volume Educational Psychology (1913–14) separated the field from pedagogy and child study, and An Introduction to the Theory of Mental and Social Measurements (1904) pushed quantitative method into the social sciences at a point when they had almost none.

He combined learning theory, psychometrics, and applied research on school subjects in a way nobody had, and was among the first to argue that schooling should improve on nature rather than merely follow it — a position that put him against the developmentalists of the period.

A less visible but enormously consequential part of the same programme was his counting of words. The Teacher's Word Book (1921) and, with Irving Lorge, its much larger successor, tabulated how often English words actually occur, producing frequency counts on an industrial scale. On this foundation rested the readability formulae, the vocabulary-controlled readers and graded textbooks, and the design of standardised reading tests used throughout the English-speaking world for most of the century. It is a rare case of a psychologist's dataset becoming invisible infrastructure.

His 1920s work on adult learning found that the capacity to learn is determined mainly by individual factors rather than age, which helped establish adult education on a factual footing. He also introduced the term halo effect for the tendency of one salient trait to colour judgement of all the others.

Major Publications

Animal Intelligence: An Experimental Study of the Associative Processes in Animals (1898 dissertation; published 1911)
Educational Psychology (1903; expanded to three volumes, 1913–14)
An Introduction to the Theory of Mental and Social Measurements (1904)
The Principles of Teaching Based on Psychology (1906)
The Teacher's Word Book (1921)
Adult Learning (1928)
Human Learning (1931)
The Fundamentals of Learning (1932)
The Psychology of Wants, Interests, and Attitudes (1935)
Human Nature and the Social Order (1940)

Impact and Legacy

Instrumental conditioning — learning from the consequences of action — is Thorndike's paradigm, and every account of reinforcement since, from Hull's drive reduction to Skinner's operant behaviour to modern reinforcement learning in machine learning, is built on it. Educational psychology as a distinct field with its own methods is his creation, as is the word-frequency infrastructure behind a century of reading instruction.

Criticism. Five lines of objection, of differing weight.

The first was made by the behaviourists themselves. The law of effect explains learning by reference to "satisfaction," a subjective conscious state — exactly the sort of thing behaviourism existed to eliminate. Guthrie replaced it with contiguity, Clark Hull with drive reduction, and Skinner with a functional definition in which a reinforcer is simply an event that strengthens the response that produced it. Skinner's move made the law testable without reference to feelings, and it is the version that survived.

The second came later and is more damaging. Thorndike assumed rewards and punishments work through pleasure and pain. But reinforcers also carry information — they tell the organism whether a response was correct — a distinction not available to him. Experiments that separate the two, by making a painful stimulus signal that a response was right, show that learning depends more on the information value of a consequence than on how it feels. The law of effect is not wrong, but it locates the mechanism in the wrong place.

There is a related circularity worry. In Thorndike's cats "satisfaction" was produced by food after deprivation, so in practice a satisfier was identified by the strengthening it was posited to explain; the law risked becoming a restatement of the finding. Skinner's functional definition existed precisely to break that circle, and the fact that the field found it necessary is itself the criticism.

What the law of effect became is the sharpest test of that objection. The field that now carries it is machine learning. When the ACM gave the 2024 Turing Award to Andrew Barto and Richard Sutton for the conceptual and algorithmic foundations of reinforcement learning, the standard history of the field named Thorndike's 1911 law of effect as its starting point — the biological branch of reinforcement learning "extends back more than a century, perhaps to Thorndike's 'Law of Effect'". The neuroscience moved the same way: Schultz and colleagues' 1997 finding that phasic midbrain dopamine firing encodes a reward-prediction error is the law of effect implemented in a circuit, and the correspondence is close enough that computational models of dopamine and models of temporal-difference learning are now treated as describing the same quantity.

What the implementation actually reinforces, however, is not the satisfaction. In temporal-difference learning the update is driven by the discrepancy between the reward that was predicted and the reward that arrived: a fully expected reward teaches nothing, and a reward that fails to arrive teaches a great deal. That is exactly the distinction between a consequence's hedonic value and its information value that Thorndike's formulation could not make, arrived at independently from the direction of engineering. The law survives as the shape of the theory — behaviour followed by a good outcome is more likely to recur — while the mechanism it named, satisfaction, has been replaced by a prediction error.

The third concerns the method. Wolfgang Köhler found evidence of insight in chimpanzee problem-solving, and the objection is not that Thorndike's cats reasoned but that his apparatus could not have shown it if they had. A puzzle box whose mechanism is invisible to the animal, and whose solution has to be hit by accident, is a device for generating trial-and-error data. The gradual learning curves it produces are in part an artefact of its construction.

Connectionism also carried an assumption of equipotentiality — that any situation can in principle be connected to any response, given the right consequences. Later work on prepared and contraprepared learning showed this is false in a way Thorndike's framework could not accommodate: rats learn to associate nausea with a taste but not with a light, and animals trained to perform operant responses drift back into species-typical behaviour regardless of reinforcement. Which connections form readily depends on the organism's biology as well as its history.

The fourth concerns the curriculum. Thorndike's demolition of formal discipline was a genuine service to evidence, and the classical curriculum had no defensible basis in the terms its defenders claimed. But the empirical case against it was never as closed as it was taken to be: his transfer experiments were criticised on methodological grounds, and later work showed that transfer of principles and strategies does occur, particularly when learners are taught for transfer explicitly rather than drilled on surface features. What followed in American schooling was not simply the removal of a false doctrine but, in the judgement of historians of education, a turn toward vocationalism and a lowering of academic expectations for students judged unlikely to need more. A finding that showed a curriculum was unjustified is not the same as a finding that shows what should replace it, and Thorndike's authority was used as though it were.

The fifth is not a scientific objection. Thorndike was a proponent of eugenics and wrote about it in language that is difficult to quote without flinching: "selective breeding can alter man's capacity to learn, to keep sane, to cherish justice or to be happy. There is no more certain and economical a way to improve man's environment as to improve his nature." And, on the duty to act on it: "it would be a pitiable thing if man should forever make inferior men as a by-product of passion, and deny good men life in mistaken devotion to palliative and remedial philanthropy."

This was not a marginal interest held in private. It followed from the same premises as the rest of his work: that capacity is substantially fixed by original nature, that it can be measured, and that the measured facts have implications for how a society should be arranged. The programme of measuring human differences and the programme of ranking human worth were not, in his hands, separate enterprises. The same logic that said schooling should improve on nature rather than follow it also said that children differ in what they can be improved to, and the same instruments that were to diagnose those differences were available to sort them. His psychometrics fed directly into the American testing movement, and the testing movement into immigration restriction and sterilisation law.

It is worth saying plainly that his own revision of his laws shows the opposite disposition where evidence was concerned: he gave up a law he had spent thirty years defending because the data did not support it. That is why the eugenics is harder to account for than simple prejudice — it was held with the confidence of someone who believed, with reason in other domains, that measurement settles questions.

Learn More

B.F. Skinner — who made the law of effect operational and built operant conditioning on it
Robert S. Woodworth — his collaborator on the transfer experiments that broke formal discipline
Ivan Pavlov — who established the other great paradigm of learning at the same time
Edward Tolman — who argued the animals were learning where things are, not what to do
Leta Hollingworth — who took the measurement of human differences in the opposite direction

References

Thorndike, E. L. (1898). Animal intelligence: An experimental study of the associative processes in animals. Psychological Review Monograph Supplements, 2(4).