George A. Miller

Introduction

George Armitage Miller (1920–2012) wrote the paper that founded cognitive psychology, and then spent fifty years watching it be misunderstood.

"The Magical Number Seven, Plus or Minus Two" (1956) is among the most cited papers in the history of psychology, and its central number has escaped into popular culture as a rule: menus should have seven items, phone numbers should have seven digits, presentations should have seven points. Miller did not believe any of this. In an interview late in his life he said, with some exasperation, that he had been "persecuted by an integer."

What the paper actually argued is more interesting than the slogan it became: that human beings can expand a severely limited capacity by recoding information into chunks, and that the capacity limit applies to chunks rather than to bits. That reframing — the mind as an information channel with a capacity that can be worked around by recoding — was the founding move of the cognitive revolution.

Early Life and Education

Miller was born in Charleston, West Virginia, on 3 February 1920. His father worked in a foundry. He took his undergraduate degree at the University of Alabama in 1940 and a master's in speech there in 1941, before moving to Harvard, where he completed his doctorate in psychology in 1946.

The war years shaped his research direction. He worked on military communication problems — the intelligibility of speech in noise, the design of voice procedures for air-ground communication — and this is where he encountered information theory and the vocabulary of channels, capacity, and noise that would structure his most famous work. It is easy to underestimate how much of the 1956 paper is engineering: the questions it asks about a channel's capacity and its noise-limited performance were the questions being asked at the time about telephone lines and radar displays, and Miller's originality was to apply them to a person remembering digits.

He was trained as a behaviourist, in an American psychology that had little use for internal states. His turn away from that position was gradual and was completed by the mid-1950s, in parallel with the work of Noam Chomsky, Jerome Bruner, and Herbert Simon.

Academic and Professional Career

Miller held appointments at MIT, Harvard, Rockefeller University, and Princeton, where he spent the last phase of his career.

At MIT in the 1950s he worked in the Research Laboratory of Electronics on speech perception, and it was there that he wrote the 1956 paper. The laboratory's culture was engineering and information theory, not psychology, which is why the paper reads the way it does: it treats the mind as a channel with measurable capacity in bits, and asks what happens at the limits.

In 1960, with Jerome Bruner, he founded the Harvard Center for Cognitive Studies. The name is a manifesto. There was no field called cognitive science in 1960, and the Center was one of the institutions that made one — an explicitly interdisciplinary home for psychology, linguistics, anthropology, and computer science, at a moment when behaviourism still dominated American departments.

He served as president of the American Psychological Association in 1969 and received the National Medal of Science in 1991. In the 1980s he began work at Princeton on WordNet, a lexical database organised by meaning relations rather than alphabetically, which became a foundational resource for computational linguistics and natural language processing.

Major Contributions to Psychology

The magical number seven

The 1956 paper assembled three separate empirical literatures and argued that they reflect a single limit. Absolute judgment is limited: people can identify about seven categories along a single sensory dimension — pitch, loudness, saltiness — before performance collapses. Immediate memory span is limited: about seven unrelated items can be recalled in order. And both limits are about seven, which Miller took to be the same underlying constraint.

The crucial move was the second half of the paper. If capacity is fixed in bits but the amount of information per item can be increased by recoding, then the effective capacity depends on coding. Miller showed that a span of seven single binary digits could become a span of seven groups of binary digits — recoding a sequence of 42 bits into 7 chunks of 6 bits each, provided the recoding is known. He coined the term chunk for these units.

Miller had hedged in the paper itself. Its closing sentence runs: “And finally, what about the magical number seven? What about the seven wonders of the world, the seven seas, the seven deadly sins, the seven daughters of Atlas in the Pleiades, the seven ages of man, the seven levels of hell, the seven primary colors, the seven notes of the musical scale, and the seven days of the week? For the present I propose to withhold judgment. Perhaps there is something deep and profound behind all these sevens, something just calling out for us to discover it. But I suspect that it is only a pernicious, Pythagorean coincidence.”

This is why the paper mattered for the cognitive revolution. It treated the mind as an information-processing system with a specifiable architecture and a measurable limit, and it made the internal representation of information — rather than the observable stimulus or response — the object of study.

TOTE and the structure of plans

With Eugene Galanter and Karl Pribram, in Plans and the Structure of Behavior (1960), Miller attacked the behaviourist account of action as a chain of stimulus-response links. Their alternative unit was the TOTE: Test–Operate–Test–Exit. An organism tests the current state against a goal, operates to reduce the discrepancy, tests again, and exits when the test passes.

This is feedback control applied to behaviour, and it made goal-directedness and hierarchical organisation legitimate topics for a psychology that had spent forty years trying to explain them away. It is also, in substance, the same idea that Herbert Simon and Allen Newell were developing as means–ends analysis at the same time.

The scientific study of language

Miller's psycholinguistics work in the 1960s, much of it in collaboration with Chomsky's students, established that grammatical complexity predicts processing difficulty and that sentences are parsed incrementally. This was the empirical programme that made transformational grammar a psychological hypothesis rather than a formal exercise.

WordNet

From 1985, at Princeton, Miller led the construction of WordNet — a large lexical database in which words are grouped into sets of cognitive synonyms and linked by semantic relations such as hyponymy, antonymy, and meronymy.

The design reflected a psycholinguistic conviction: that the mental lexicon is organised by meaning relations rather than by spelling, and that a resource built that way would be useful both for modelling language processing and for practical computational work. WordNet became one of the most widely used resources in natural language processing and supported decades of research on semantic similarity.

Major Publications

  • Language and Communication (1951)
  • "The Magical Number Seven, Plus or Minus Two" (1956, Psychological Review) — the founding paper of cognitive psychology
  • Plans and the Structure of Behavior (1960, with Eugene Galanter and Karl Pribram)
  • The Psychology of Communication (1967)
  • Psychology: The Science of Mental Life (1962)
  • The Science of Words (1991)
  • WordNet (from 1985), as project director

Impact and Legacy

Miller's influence is unusual in being institutional as well as intellectual. The Harvard Center for Cognitive Studies trained and housed a generation of cognitive scientists. Plans and the Structure of Behavior and the 1956 paper between them dismantled the behaviourist prohibition on internal states, and the field that replaced behaviourism is the one he named.

Criticism — including his own. The number has not survived. Nelson Cowan's 2001 review concluded that the genuine capacity of working memory is closer to four chunks than seven, and that Miller's estimate was inflated by the methods he was reviewing: absolute-judgment experiments and immediate-memory-span experiments were pooled into a single limit, and the pooling is not clearly legitimate. Absolute judgment of a sensory dimension and serial recall of a list are different tasks with different constraints; adding their limits together and dividing was never a licensed operation, and Miller was candid that the convergence of "about seven" was a suggestive observation rather than a measurement. Later work has gone further and questioned whether there is a single fixed capacity at all, proposing instead that apparent limits reflect interference and retrieval processes rather than a store with a size.

Miller's own account of how he arrived at the number is more damaging than any published criticism. In a letter Nelson Cowan received in January 2000, after Miller had read Cowan's target article, he wrote: “Herb Simon used to say ‘George had the right idea, but the wrong number.’ I think Herb favored 3. I never argue with Herb Simon. I was aware, even in 1954 when I was writing my invited address, that the running memory span was only about 4, and my introspections convinced me that the 7 that was standard on the intelligence tests at that time must have been a hooking together of a rehearsed initial segment with a final segment from what Nancy Waugh used to call the ‘echo box.’ But I was stuck with 7 because of the absolute judgment results (the first half of the article that people forget).” He added that his interest had been in chunking as a way round the limit, that “everyone else wanted to fight about the size of the limit, or whether there was any limit at all — questions of little interest to me”, and that of his 1989 autobiographical chapter, “that is how I came to own the number 7”. Cowan published the exchange in 2015 (Psychological Review, 122(3), 536–541).

Cowan also documents the cost. For roughly forty years after 1956 almost nobody investigated capacity limits directly, partly because Miller's ironic framing made the question look unserious, and the field moved toward time-based and interference-based accounts instead. The paper that founded the study of working memory capacity is therefore also a large part of the reason that study was deferred for a generation.

The deeper limitation is that chunking was proposed as the mechanism and never specified. If capacity is measured in chunks, everything depends on how chunks are formed and how large they can be — and Miller had no account of either. That is where the subsequent field went, and the accounts that came out of it (Baddeley's working memory model, in particular) treat the 1956 paper as a starting point rather than a result.

The psycholinguistic programme has aged unevenly. The specific hypothesis that occupied much of it — the derivational theory of complexity, according to which the number of grammatical transformations in a sentence's derivation predicts how hard the sentence is to process — was tested extensively and failed: sentences that differ in derivational complexity do not differ correspondingly in difficulty once other factors are controlled. The programme that replaced it, incremental parsing with multiple interacting constraints, owes much to the questions Miller's group asked and little to the answer they first proposed. This is a normal outcome in a young field, but it is worth recording that the second half of his most productive decade rested on a hypothesis that did not hold.

WordNet has its own critics. It is, in the end, a very large thesaurus: its "cognitive synonyms" are stipulated rather than derived from evidence about how speakers actually treat words as equivalent, it represents meaning only for single words and so handles nothing compositional, and its coverage and category structure reflect the intuitions of its English-speaking compilers — which is why systems built on it have been shown to inherit those biases. Miller was explicit that it was a resource rather than a theory, and in that capacity it has been enormously productive; the criticism is that its success led a generation of researchers to treat its lexical relations as if they were findings about the mental lexicon.

Miller's own objection to what the paper became is the most quotable critique. He said he had been persecuted by an integer, and he objected specifically to the use of seven as a design constraint. The paper says that people can hold about seven chunks; it says nothing about how many menu items, bullet points, or slides an audience can absorb, because those involve comprehension and long-term memory rather than the span he measured. Applying it as a design rule is a category error of precisely the kind the paper's own argument about recoding should have prevented.

Learn More

References

  • Miller, G. A. (1956). The magical number seven, plus or minus two: Some limits on our capacity for processing information. Psychological Review, 63(2), 81-97.

Last updated: 2026-10-20

Menu
Loading