On Lines and Planes of Closest Fit to Systems of Points in Space
Karl Pearson's 1901 paper introducing principal component analysis, the foundational method of dimensionality reduction.
Landmark Papers
What the papers actually said - linked to the originals.
644 entries, all primary-sourced
A Symbolic Analysis of Relay and Switching Circuits
Shannon's 1937 MIT master's thesis showed Boolean algebra could design and simplify switching circuits - the logical foundation of digital computers.
A Logical Calculus of the Ideas Immanent in Nervous Activity
The 1943 paper by McCulloch and Pitts that modeled the brain's neurons as simple logical switches, founding the idea of the artificial neural network.
Computing Machinery and Intelligence
Alan Turing's 1950 paper that asked whether machines can think and replaced the question with a practical test - the imitation game, now called the Turing test.
On Information and Sufficiency
Kullback and Leibler's 1951 paper introducing the KL divergence, the standard measure of how one probability distribution differs from another.
A Stochastic Approximation Method
Robbins and Monro's 1951 paper introducing stochastic approximation, the mathematical ancestor of stochastic gradient descent.
Equation of State Calculations by Fast Computing Machines
The 1953 Los Alamos paper that introduced the Metropolis Monte Carlo method, the seed of modern Markov chain sampling.
The Magical Number Seven, Plus or Minus Two
George Miller's 1956 paper measuring the mind's limited capacity in bits, a founding text of the cognitive revolution.
Syntactic Structures
Chomsky's 1957 book argued syntax is independent of meaning and reshaped how computers model human language.
The Perceptron: A Probabilistic Model for Information Storage and Organization in the Brain
Frank Rosenblatt's 1958 paper introducing the perceptron, the first artificial neural network that could learn from examples by adjusting its own weights.
Programs with Common Sense
McCarthy's 1959 paper proposed the advice taker, a system that would reason from facts stated in formal logic, founding logic-based AI.
Man-Computer Symbiosis
Licklider's 1960 paper proposing that humans and computers form a close partnership, each doing what it does best.
Recursive Functions of Symbolic Expressions and Their Computation by Machine, Part I
John McCarthy's 1960 paper that defined Lisp, introducing symbolic list processing, recursion, and the idea that a language can be written in terms of itself.
Steps Toward Artificial Intelligence
Minsky's 1961 survey that mapped the young field of AI into five problem areas: search, pattern recognition, learning, planning, and induction.
Irreversibility and Heat Generation in the Computing Process
Landauer's 1961 paper that tied logic to thermodynamics, showing erasing one bit must dissipate at least kT ln 2 of energy as heat - the Landauer limit.
A Machine Program for Theorem-Proving
The 1962 paper by Davis, Logemann, and Loveland introduced the DPLL procedure, the backtracking search at the heart of modern SAT solvers.
A Machine-Oriented Logic Based on the Resolution Principle
J. A. Robinson's 1965 paper introduced resolution, the single inference rule that made automated theorem proving practical and later powered Prolog.
Alchemy and Artificial Intelligence
Hubert Dreyfus's 1965 RAND memo comparing AI researchers to alchemists, the opening shot of his decades-long philosophical critique.
Theory of Self-Reproducing Automata
Von Neumann's posthumous work proved a machine could reproduce itself, via a cellular automaton universal constructor, before molecular biology caught up.
Nearest Neighbor Pattern Classification
Cover and Hart's 1967 paper that put the k-nearest-neighbor rule on a firm theoretical footing for classification.
The Nature of Mental States
Putnam's paper introducing functionalism and multiple realizability: a mental state is defined by its role, not its physical material.
A Formal Basis for the Heuristic Determination of Minimum Cost Paths
The 1968 paper by Hart, Nilsson, and Raphael that introduced A*, the heuristic search algorithm still used for pathfinding and planning.
Monte Carlo Sampling Methods Using Markov Chains and Their Applications
Hastings' 1970 paper that generalized the Metropolis method into the broad Metropolis-Hastings algorithm used across statistics.
Artificial Paranoia (the PARRY model)
Colby, Weber and Hilf describe a computer program that simulates a paranoid patient, judged by indistinguishability tests.
On the Uniform Convergence of Relative Frequencies of Events to Their Probabilities
Vapnik and Chervonenkis's 1971 paper introducing VC dimension, the measure of model capacity at the heart of learning theory.
STRIPS: A New Approach to the Application of Theorem Proving to Problem Solving
The 1971 paper by Fikes and Nilsson that introduced STRIPS, the planner whose preconditions and add/delete lists still underpin AI planning.
A Statistical Interpretation of Term Specificity (IDF / TF-IDF)
Karen Sparck Jones's idea that rare words carry more weight, the basis of TF-IDF and decades of search.
A Bayesian Analysis of Some Nonparametric Problems
Ferguson's 1973 paper introduced the Dirichlet process, the prior that lets Bayesian models grow their complexity with the data.
A Framework for Representing Knowledge
Marvin Minsky's 1974 MIT memo introduced frames, structured templates of expectations that became a foundation of knowledge representation.
Design of Ion-Implanted MOSFET's with Very Small Physical Dimensions (Dennard Scaling)
The 1974 IBM paper defining Dennard scaling - the rule that shrinking transistors keeps power density flat - which powered chip progress until it broke down.
What Is It Like to Be a Bat?
Nagel's 1974 essay arguing that subjective experience cannot be captured by purely physical, objective descriptions of the mind.
An Analysis of Alpha-Beta Pruning
The 1975 Knuth and Moore paper that gave the first rigorous analysis of alpha-beta pruning, the technique that makes game-playing search tractable.
The Language of Thought (Fodor)
Jerry Fodor's 1975 book argued that thinking runs on an innate symbolic language of the mind, a touchstone for the symbolic view of intelligence.
Computer Science as Empirical Inquiry: Symbols and Search
Newell and Simon's 1975 Turing Award lecture stating the physical symbol system hypothesis at the heart of symbolic AI.
Computer Science as Empirical Inquiry: Symbols and Search
Newell and Simon's 1976 Turing Award lecture stating the physical symbol system and heuristic search hypotheses, the manifesto of symbolic AI.
Maximum Likelihood from Incomplete Data via the EM Algorithm
Dempster, Laird, and Rubin's 1977 paper that unified a class of methods into the Expectation-Maximization algorithm for fitting models with hidden data.
Scripts, Plans, Goals and Understanding
The 1977 Schank and Abelson book that proposed scripts, stereotyped event sequences, as the knowledge structures a machine needs to understand stories.
Modeling by Shortest Data Description
Rissanen's 1978 paper introduced the minimum description length principle: the best model is the one that compresses the data most.
Bootstrap Methods: Another Look at the Jackknife
Efron's 1979 paper introduced the bootstrap, estimating uncertainty by resampling the data itself instead of relying on formulas.
A Truth Maintenance System
Jon Doyle's 1979 paper introducing the truth maintenance system, which tracks the reasons behind a program's beliefs so it can cleanly revise them.
A Logic for Default Reasoning
Raymond Reiter's 1980 paper introducing default logic, a formal way to draw plausible conclusions like 'birds fly' that can be withdrawn on new evidence.
Circumscription: A Form of Non-Monotonic Reasoning
John McCarthy's 1980 paper formalizing how a reasoner can jump to the conclusion that the known objects are the only ones, a cornerstone of nonmonotonic logic.
The Hearsay-II Speech-Understanding System: Integrating Knowledge to Resolve Uncertainty
The 1980 paper on Hearsay-II introduced the blackboard architecture, where independent knowledge sources cooperate on a shared workspace.
Reverend Bayes on Inference Engines: A Distributed Hierarchical Approach
Pearl's 1982 paper introduced belief propagation, a message-passing scheme for updating probabilities across a network of variables.
Theory Formation by Heuristic Search (AM and EURISKO)
Douglas Lenat's 1983 paper describing AM and EURISKO, programs that used heuristics to discover new concepts and even invent new heuristics of their own.
A Theory of the Learnable (PAC Learning)
Valiant's 1984 paper founding computational learning theory with the Probably Approximately Correct model of learning.
Rule-Based Expert Systems: The MYCIN Experiments
The 1984 Buchanan and Shortliffe book collecting a decade of MYCIN research, the definitive record of how rule-based expert systems were built and tested.
The 2 Sigma Problem (Bloom, 1984)
Bloom's 1984 paper found one-to-one tutoring lifted average students two standard deviations - the benchmark AI tutoring chases.