Automaton
An automaton, from the Greek automatos meaning self-moving, is a machine that carries out a predetermined sequence of operations on its own, and in modern mathematics the word names an abstract model of computation defined by states and transition rules. The term therefore covers two long traditions that meet in the twentieth century: the craft of building mechanical figures that seem to act by themselves, and the theory of idealized machines that take input and change state according to fixed rules. Both ask the same underlying question, which is how much apparently purposeful behavior can be produced by mechanism alone.
The mechanical tradition begins in antiquity and flourishes in the medieval Islamic world. Homer's Iliad describes golden maidens fashioned by Hephaestus, a myth rather than engineering, but Hero of Alexandria in the first century CE wrote on pneumatic devices and on automaton-making, describing self-moving theatres driven by weights, ropes and cams. In 1206 the engineer Ismail al-Jazari completed his Book of Knowledge of Ingenious Mechanical Devices, documenting water clocks, hand-washing machines and musical automata with real mechanisms such as crankshafts and camshafts. These devices were programmed in the sense that their behavior was fixed in cam shapes and peg layouts, which can be read as an early form of stored instruction.
Eighteenth-century automata turned craftsmanship into public spectacle and into a debate about life. Jacques de Vaucanson exhibited a flute player in 1738 and, in 1739, a mechanical duck that appeared to eat grain, digest it and excrete. The duck's wings and neck were genuinely intricate, but later examination indicated that the digestion was staged rather than chemical, an early lesson that a convincing demonstration is not proof of the claimed process. The same age produced the Jaquet-Droz writing and drawing automata, and Wolfgang von Kempelen's chess-playing Turk of 1770, which concealed a human operator. The Turk is the standard cautionary tale: a machine that seems to think may hide its intelligence in the wrong place.
The Jacquard loom of 1804 is the hinge between figures that move and machines that are programmed. Joseph Marie Jacquard's loom used a chain of punched cards to select which warp threads were raised for each pass, so changing the cards changed the woven pattern with no change to the machine. Charles Babbage planned to feed his Analytical Engine with similar cards, and Ada Lovelace noted that the engine could weave algebraic patterns as the Jacquard loom wove flowers and leaves. The separation of a general mechanism from a replaceable program is the structural idea behind later stored-program computing.
Formal automata theory abstracts the mechanism away and keeps only states, inputs and transitions. Alan Turing's 1936 machine, an infinite tape read and written by a finite-state controller, set the limit of what any mechanical procedure can compute. At the bottom of the hierarchy sit finite automata, which have no memory beyond a fixed number of states; the 1943 neural model of McCulloch and Pitts is effectively one, Stephen Kleene showed in the 1950s that they recognize exactly the regular events, and Michael Rabin and Dana Scott gave the modern treatment in 1959. Noam Chomsky's 1956 hierarchy of grammars matched machine classes to language classes: finite automata for regular languages, pushdown automata for context-free ones, and Turing machines for the most general. The practical payoff is everywhere in software, from regular-expression engines and lexers to protocol state machines.
Cellular automata moved the idea from single machines to grids of identical cells. John von Neumann, prompted by a question of whether a machine could build a copy of itself, designed in the 1940s and 1950s a cellular automaton with many states per cell that could construct a copy of itself; his manuscript was edited by Arthur Burks and published in 1966 as Theory of Self-Reproducing Automata. John Conway's Game of Life, introduced to the public by Martin Gardner in Scientific American in October 1970, needs only two states and three simple neighbor rules, yet it supports gliders, oscillators and, as later shown, constructions equivalent to a universal computer. Matthew Cook demonstrated that Stephen Wolfram's one-dimensional Rule 110 is also universal, in work published in 2004.
The word invites several errors. An automaton in the theory sense is not a robot but a mathematical object, and most are far weaker than a Turing machine. The old figures were not intelligent: their behavior was fixed by cams, and the Turk was a fraud while the duck's digestion was staged. It is also a mistake to dismiss automata as quaint, because their descendants run through modern computing: state machines control game characters and network protocols, and cellular-automaton ideas resurface in simulation and in studies of emergence. Language models are not finite automata in any useful sense, yet theory still guides questions about which formal languages a given architecture can recognize.
Read automaton as the oldest answer to the question of how much behavior a fixed mechanism can hold: first in cams and punched cards, then in states and rules, and finally as the limit that every computer, however large, still obeys.