The magical number seven plus or minus two: some limits on our capacity for processing information.

By G. A. Miller. 1956

Subfields: Cognitive Science & Psychology

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Cognitive Science & Psychology

Citations in the corpus

Neural Turing Machines (2014, background): “Psychologists have extensively studied the capacity limitations of working memory, which is often quantified by the number of “chunks” of information that can be readily recalled (Miller, 1956).”
Interpretable Machine Learning: Fundamental Principles and 10 Grand Challenges (2021, background): “Sparsity is useful because humans can handle only 7±2 cognitive entities at the same time (Miller, 1956), and sparsity makes it easier to troubleshoot, check for typographical errors, and reason about counterfactuals (e.”
Stop Explaining Black Box Machine Learning Models for High Stakes Decisions and Use Interpretable Models Instead (2018, background): “table models can perform case-based reasoning for complex domains. Often for structured data, sparsity is a useful measure of interpretability, since humans can handle 7 2 cognitive entities at once [Miller, 1956]. Sparse models allow a view of how variables interact jointly rather than individually”
Please Stop Explaining Black Box Models for High Stakes Decisions (2018, background): “Often for structured data, sparsity is a useful measure of interpretability, since humans can handle 7±2 cognitive entities at once [Miller, 1956].”
Model-Agnostic Interpretability of Machine Learning (2016, background): “Interpretable models for such tasks remain unsatisfying; such models are inherently crippled by the need to be understandable, being susceptible to the limited “perception budget” (Miller, 1956) of the users.”
Interpretable classifiers using rules and Bayesian analysis: Building a better stroke prediction model (2015, background): “There is a subfield of artificial intelligence, Inductive Logic Programming [Muggleton and De Raedt (1994)], whose goal is to mine individual conjunctive rules.”
Interpretable classification models for recidivism prediction (2015, methodology): “…SLIM: Here, we constrained each coefﬁcient λ j to an integer between − 10 and 10 , we constrained the threshold λ 0 to an integer between − 100 and 100 , and we restricted the number of non-zero coefﬁcients to at most 8, similar to the number of cognitive entities humans can handle (Miller 1956).”
On the Binding Problem in Artificial Neural Networks (2020, background): “Indeed, the capacity of the human working memory is generally believed to only be around 3–9 objects (Fukuda et al., 2010; Miller, 1956).”
Learning customized and optimized lists of rules with mathematical programming (2018, background): “There is a body of work in psychology [32,46] indicating that humans cannot handle very many cognitive entities at once (hence the need for sparse models in a wide variety of settings).”
Interpretable Decision Sets: A Joint Framework for Description and Prediction (2016, background): “For example, research in psychology has demonstrated that an average human can pay attention to about 6 objects at any one time [28].”
Learning Optimized Or's of And's (2015, methodology): “In creating predictive models for healthcare, marketing, sociology, and in other domains, two aspects have long since been of interest: logical forms, and sparsity (see, e.g. Dawes, 1979; Johnson-Laird, Khemlani, and Goodwin, 2015; Miller, 1956).”
The Symbol Grounding Problem (1990, neutral): “Psychology became more like an empirical science when, with the gradual advent of cognitivism [17] [25] [29], it became acceptable to make inferences about the unobservable processes underlying behavior.”
Or's of And's for Interpretable Classification, with Application to Context-Aware Recommender Systems (2015, background): “The set of conditions within the model should be sparse, as humans can handle about 7 ± 2 cognitive entities at once [32].”
How do Humans Understand Explanations from Machine Learning Systems? An Evaluation of the Human-Interpretability of Explanation (2018, background): “Miller [1956] famously argued that humans can hold about seven items simultaneously in working memory, suggesting that human-interpretable explanations should obey some kind of capacity limit (importantly, these items can correspond to complex cognitive chunks —for example, ‘CIAFBINSA’ is easier to…”
Active Reward Learning (2014, background): “In a review of several experiments, Miller [25] concludes that humans have a general channel capacity somewhere between 1.6 and 3.9 bits for unidimensional stimuli.”
A survey of algorithmic recourse: definitions, formulations, solutions, and prospects (2020, background): “It is often argued that sparser solutions are desirable as they emphasize fewer changes (in explanations) or fewer variables to act upon (in recommendations) and are thus more inter-pretableforthe individual[123].”
Toward an Integration of Deep Learning and Neuroscience (2016, background): “Its capacity is somewhat limited, with the old idea being that verbal working memory has a capacity of “seven plus or minus two” (Miller, 1956), while visual working memory has a capacity of four (Luck and Vogel, 1997) (or, other authors defend, one).”
A consensus model for multiperson decision making with different preference structures (2002, methodology): “Miller’s study [16], Saaty suggests measuring using a ratio”