History of Computing: From the Abacus to Artificial Intelligence

In 1843, Ada Lovelace published a step-by-step method for calculating Bernoulli numbers on a machine that did not exist. She was writing for Charles Babbage's Analytical Engine, a mechanical computer that was designed but never built. More than a century would pass before any machine could run instructions of that kind. Yet in that note, printed in a scientific journal under the initials "A.A.L.", a human being had grasped that a machine manipulating numbers could, in principle, manipulate any symbols at all.

The history of computing is a story of compression. Each generation replaced room-sized machines with smaller, cheaper and more reliable ones, and each step changed who could use a computer and for what. A smartphone in an ordinary pocket today holds billions of transistors, far more switching elements than any computer built in 1960. This article walks through that story in order: counting tools, mechanical calculators, the theory of computation, the first electronic machines, the stored-program idea, the transistor and the chip, the personal computer, networks, and finally the machine-learning systems that dominate headlines now.

Key Takeaways

• Calculating aids such as counting boards and the abacus are thousands of years old; mechanical calculators appeared in Europe in the 17th century
• Charles Babbage designed the first general-purpose programmable mechanical computer, the Analytical Engine, in the 1830s; Ada Lovelace's 1843 notes contain what is usually called the first published program
• Alan Turing's 1936 paper defined what "computable" means, and John von Neumann's 1945 EDVAC draft described the stored-program design almost every computer still follows
• Electronic computing arrived during and just after World War II with Colossus (1944), ENIAC (unveiled 1946) and the Manchester Baby (1948)
• The transistor (1947), integrated circuit (1958–59) and microprocessor (1971) moved logic from rooms to fingernail-sized chips
• AI has been a research field since 1956; its current form, deep learning, took off after 2012

Ancient Roots: Counting Before Computers

Long before electricity, people needed to add up harvests, taxes and trade. The earliest aids were counting boards: lines drawn on a table or in sand, with pebbles moved between them to represent units, tens and hundreds. The Latin word calculus simply means "small stone", and it survives in both "calculate" and "calculus". The bead-frame abacus most people picture, such as the Chinese suanpan and the Japanese soroban, is a later refinement of the same idea. Exact origins are debated by historians, and dates for the earliest counting boards vary widely by source, so it is safest to say that physical calculating aids go back thousands of years across Mesopotamia, the Mediterranean and East Asia.

The abacus is not a computer in the modern sense. It stores numbers, but the human operator performs every step of the procedure. That distinction, between a device that holds data and a device that follows instructions by itself, is the thread that runs through the rest of this history.

The ancient Greek world produced a striking exception: the Antikythera mechanism, a geared bronze device recovered from a shipwreck off the Greek island of Antikythera in 1901. Researchers have shown that it modelled astronomical cycles, including the positions of the Sun and Moon and the timing of eclipses. It was an analog computer in the strict sense: its gear ratios physically embodied the mathematical relationships it calculated. Nothing of comparable mechanical complexity survives from the following centuries, which is one reason the object remains so famous.

Through the medieval and early modern periods, calculation stayed overwhelmingly manual. Astronomers, navigators and insurers relied on printed tables of logarithms, trigonometric functions and planetary positions. Those tables were produced by people called "computers", a job title that lasted well into the 20th century. The word described a person long before it described a machine. In the 1940s and 1950s, teams of human computers, many of them women, still did essential calculations for aviation and spaceflight; the story of one of them is told in our profile of Katherine Johnson, the human computer who sent John Glenn to orbit.

Mechanical Calculators (1623–1820s)

The first mechanical calculators appeared in 17th-century Europe:

• Wilhelm Schickard described a "calculating clock" in letters to Johannes Kepler in 1623. No original survives, but reconstructions have been built from his sketches.
• Blaise Pascal built the Pascaline in the 1640s to help his father, a tax official. It added and subtracted using geared wheels with an automatic carry.
• Gottfried Wilhelm Leibniz designed the stepped reckoner in the 1670s and 1690s, which could also multiply and divide through repeated addition. Leibniz also wrote about binary arithmetic, the base-2 system that would become central to digital computers three centuries later.

These machines were impressive, but they were fixed-function. Each did arithmetic and nothing else, and every calculation still needed a person to decide which operation came next. Mechanical calculators became commercially successful only in the 19th century, when the arithmometer of Thomas de Colmar went into series production.

A parallel development came from textiles. In 1804, Joseph-Marie Jacquard demonstrated a loom controlled by chains of punched cards. A hole or no hole decided whether a given thread was lifted. Changing the cards changed the pattern without rebuilding the loom. It was an early form of a stored, replaceable program, and it directly inspired Babbage.

Babbage, Lovelace, and the Dream of Programmable Machines

Charles Babbage was an English mathematician frustrated by the errors in printed mathematical tables. In 1822 he proposed the Difference Engine, a machine that would calculate polynomial tables by the method of finite differences and print the results, removing human copying mistakes. The British government funded the work, but after years of cost overruns and a dispute with his engineer, the project stopped in the 1830s with only a demonstration section complete. Much later, in 1991, the Science Museum in London completed a working Difference Engine No. 2 from Babbage's later drawings, showing that his design worked with the manufacturing tolerances of his era.

Babbage's second design was far more ambitious. The Analytical Engine, developed from the 1830s onward, separated the "store" (memory) from the "mill" (the arithmetic unit), read its instructions and data from punched cards borrowed from Jacquard's loom, and could branch: it could choose its next operation depending on a previous result. Those are the defining features of a general-purpose computer. It was never built, partly because of cost and partly because Babbage kept revising it.

Ada Lovelace, the mathematically trained daughter of the poet Lord Byron, translated an article on the Analytical Engine written in French by the Italian engineer Luigi Menabrea. She added seven notes, lettered A to G, that together ran longer than the original. The translation was published in August 1843 in Taylor's Scientific Memoirs. Note G set out, step by step, how the engine could compute Bernoulli numbers. Historians still debate how much of the mathematics came from Babbage and how much from Lovelace, but the notes go beyond arithmetic. They argue that the engine could act on any symbols whose relationships could be expressed formally, and speculate that it might one day compose music. They also contain a famous caution: the engine could only do what it was ordered to do and could not originate anything. Alan Turing later called this "Lady Lovelace's objection" and answered it in his 1950 paper on machine intelligence. Her life and work are covered in more depth in our profile of Ada Lovelace.

A further intellectual ingredient arrived in 1854, when George Boole published An Investigation of the Laws of Thought, which treated logical reasoning as algebra on true and false values. In 1937, Claude Shannon's master's thesis at MIT showed that Boole's algebra could describe, and help design, networks of electrical relays and switches. That link between logic and circuits is why every digital chip today is, at bottom, a vast arrangement of Boolean operations. See George Boole: The Logic That Became the Computer for that story.

The Theory Arrives: Turing, 1936

Before anyone built an electronic computer, mathematicians asked a deeper question: what, exactly, can be computed by following rules? In 1936, the 24-year-old Alan Turing wrote "On Computable Numbers, with an Application to the Entscheidungsproblem". It was received by the London Mathematical Society in May 1936 and printed in its Proceedings (Series 2, volume 42) in parts dated November and December 1936, which is why some references cite it as 1936 and others as 1937.

Turing imagined a machine that reads and writes symbols on an unbounded tape, one cell at a time, according to a finite table of rules. He then showed that a single "universal" machine could simulate any other such machine if given that machine's rule table as data on its tape. That is the theoretical root of software: one physical device that becomes a different machine depending on the program it is fed. Turing also proved that some problems cannot be solved by any such procedure, which answered a question posed by David Hilbert. Independently and at about the same time, Alonzo Church in the United States reached an equivalent result with a different formalism.

Turing's later wartime work at Bletchley Park, his 1950 paper "Computing Machinery and Intelligence", and his prosecution and death in 1954 are covered in our profile of Alan Turing.

Electromechanical and Wartime Machines

The late 1930s and World War II pushed several teams, mostly unaware of each other, to build real computing machines. The table below summarises the most cited ones, using dates from the Computer History Museum's timeline.

Machine	Where	Year	Technology	What made it notable
Atanasoff–Berry Computer	Iowa State College, USA	1939–1942	Vacuum tubes, capacitor drum memory	Electronic arithmetic for linear equations; not programmable
Zuse Z3	Berlin, Germany	1941	About 2,300 relays	Programmable, binary floating-point; destroyed in a 1943 air raid
Colossus	Bletchley Park, UK	1944	Up to 2,500 vacuum tubes per machine	Electronic codebreaking of the German Lorenz cipher; kept secret until the 1970s
Harvard Mark I	Harvard / IBM, USA	1944	About 3,500 relays	Room-sized electromechanical calculator producing mathematical tables
ENIAC	University of Pennsylvania, USA	Unveiled 1946	About 18,000 vacuum tubes, 30 tons	General-purpose electronic machine, programmed by rewiring
Manchester Baby	University of Manchester, UK	June 1948	Vacuum tubes, Williams-tube memory	First program run from electronic stored memory
EDSAC	University of Cambridge, UK	1949	Vacuum tubes, mercury delay lines	First practical stored-program machine offering a regular service

Konrad Zuse in Germany worked largely in isolation. His Z3, completed in 1941, used telephone relays and read its program from punched film. It is often described as the first working programmable, fully automatic digital computer, though it was electromechanical rather than electronic.

Colossus, designed by the Post Office engineer Tommy Flowers, became operational at Bletchley Park in 1944. Its job was to help break the Lorenz cipher used for German high command messages, and it cut the time needed from weeks to hours. Ten were delivered by the end of the war. Turing worked at Bletchley Park on the separate Enigma problem and the electromechanical Bombe machines; Colossus was Flowers's design. Because the project stayed secret until the 1970s, it was left out of computing histories for decades.

ENIAC, the Electronic Numerical Integrator and Computer, was built by J. Presper Eckert and John Mauchly at the University of Pennsylvania's Moore School for the US Army, which needed artillery firing tables. Work began in 1943 and the machine was unveiled in 1946. The Computer History Museum describes it as using about 18,000 vacuum tubes, weighing 30 tons and covering more than 1,000 square feet. It was more than a thousand times faster than the electromechanical machines before it. But programming meant setting switches and plugging cables, a job done largely by a group of six women mathematicians whose contribution was overlooked for decades. Reconfiguring ENIAC for a new problem could take days.

The question of who "invented the computer" was eventually tested in court. In 1973, a US federal ruling in Honeywell v. Sperry Rand found that Mauchly had seen Atanasoff's machine before designing ENIAC and declared the ENIAC patent invalid. The honest answer is that the electronic computer had many parents.

The Stored-Program Idea (1945–1949)

ENIAC's weakness was obvious to its own designers: the program lived in the wiring. The fix was to store instructions in the same memory as data, so a new program could be loaded as quickly as new numbers.

In 1945 John von Neumann, a mathematician consulting on the follow-up project called EDVAC, wrote the "First Draft of a Report on the EDVAC". The typed draft is dated June 30, 1945, and was circulated by Herman Goldstine. It described a machine with a processing unit, a control unit, a memory holding both instructions and data, and input and output. That layout became known as the "von Neumann architecture". The label is contested, because the draft carried only von Neumann's name while Eckert and Mauchly had developed many of the ideas, and the dispute over credit split the ENIAC team. Von Neumann's wider influence, from game theory to the hydrogen bomb, is covered in John von Neumann: The Man Who Built the Modern World.

The first machine to actually run a program from electronic memory was not EDVAC. It was the Manchester Small-Scale Experimental Machine, nicknamed the "Baby". On June 21, 1948, it ran a 17-instruction program written by Tom Kilburn. A year later, EDSAC at Cambridge, led by Maurice Wilkes, began offering a regular computing service to university researchers and popularised libraries of reusable subroutines. Commercial computing followed quickly: the first UNIVAC I was delivered to the US Census Bureau in 1951.

Transistors, Chips, and the Great Miniaturization

Vacuum-tube computers were large, hot, power-hungry and unreliable, because tubes burn out. The replacement came from Bell Telephone Laboratories, where John Bardeen and Walter Brattain, working in William Shockley's solid-state group, built the first point-contact transistor in December 1947. A transistor does the job of a tube, switching or amplifying a current, using a small piece of semiconductor with no heated filament and no vacuum. By the late 1950s transistorised computers were replacing tube machines. The full story of that device, from germanium and gold foil to today's three-dimensional transistors, is in our companion article How the Transistor Changed Everything.

Wiring thousands of individual transistors by hand created its own problem, sometimes called the "tyranny of numbers". Jack Kilby at Texas Instruments (1958) and Robert Noyce at Fairchild Semiconductor (1959) independently found ways to build complete circuits on a single piece of semiconductor. Noyce's version, using the planar process and deposited metal connections, was the one industry could mass-produce. Microchip Invention: Kilby, Noyce, and the Integrated Circuit covers that rivalry.

In April 1965, Gordon Moore, then at Fairchild, published "Cramming more components onto integrated circuits" in Electronics magazine. According to the Computer History Museum, he plotted the number of components per chip from 1959 to 1964 and projected a doubling every year, reaching about 65,000 components by 1975. In 1975 he revised the rate to a doubling every two years. That forecast, later called Moore's Law, became a planning target for the whole industry.

Meanwhile, the computer itself was becoming a product family. On April 7, 1964, IBM announced the System/360: a line of compatible machines across a wide range of speeds and prices that could largely run the same software. It established the idea that software was an investment worth protecting across hardware generations.

The next leap was to put an entire processor on one chip. Intel's 4004, designed for the Busicom calculator by Federico Faggin, Ted Hoff, Stanley Mazor and Busicom's Masatoshi Shima, went on sale on November 15, 1971. Guinness World Records lists it as the first commercially available microprocessor, with 2,300 transistors and a clock speed of 740 kHz. Fifty years later, chips with tens of billions of transistors are ordinary consumer products.

The PC Revolution and the Internet Age

Once a processor fitted on a chip, a computer could fit on a desk. The Altair 8800 appeared on the cover of Popular Electronics in January 1975 as a kit; the Computer History Museum lists it at $297, or $395 with a case. Bill Gates and Paul Allen wrote a BASIC interpreter for it, which became Microsoft's first product. The Apple II followed in 1977 as a ready-to-use machine with a keyboard and colour graphics.

IBM's Personal Computer, the Model 5150, was announced in 1981 with an Intel 8088 processor running at 4.77 MHz and Microsoft's MS-DOS. Because IBM built it largely from off-the-shelf parts, other companies could make compatible "clones", which turned the PC into a commodity and shifted power from hardware makers to the suppliers of the processor and operating system. Apple's Macintosh, launched in 1984, brought the graphical user interface, with windows, icons and a mouse, to a mass audience. Those ideas had been developed at Xerox PARC in the 1970s on the Alto. The arc from hobbyist kits to the beige box on every office desk is traced in The Personal Computer: From Kit to Commodity.

Networking ran in parallel. ARPANET, funded by the US Defense Department, connected its first nodes in 1969, and the TCP/IP protocols adopted on ARPANET on January 1, 1983 became the common language of the internet. In 1989, Tim Berners-Lee at CERN proposed the system that became the World Wide Web, and the web spread to the public through the 1990s. Our articles on the invention of the internet and Tim Berners-Lee cover those two stages separately, because they are often confused: the internet is the network, and the web is one service that runs on it.

The 2000s and 2010s moved computing in two directions at once. It went into pockets, as smartphones combined a computer, a camera and a radio, and into large remote data centres, as companies rented processing and storage instead of buying their own machines. The Smartphone: From IBM Simon to On-Device AI follows the first of those threads.

A Timeline of Computing Milestones

Year	Milestone
1623	Schickard describes his calculating clock
1804	Jacquard's punched-card loom
1822	Babbage proposes the Difference Engine
1843	Lovelace's notes on the Analytical Engine published
1854	Boole's Laws of Thought
1936	Turing's "On Computable Numbers"
1941	Zuse Z3 completed
1944	Colossus operational; Harvard Mark I completed
1945	Von Neumann's EDVAC draft (dated June 30)
1946	ENIAC unveiled
1947	Point-contact transistor at Bell Labs
1948	Manchester Baby runs the first stored program
1951	UNIVAC I delivered to the US Census Bureau
1958–59	Integrated circuit (Kilby, then Noyce)
1964	IBM System/360 announced
1965	Moore's "Cramming more components" article
1969	First ARPANET link
1971	Intel 4004 microprocessor
1975	Altair 8800
1981	IBM PC
1989	Web proposal at CERN
2012	AlexNet starts the deep-learning boom

Common Misconceptions About Computing History

"ENIAC was the first computer." It was the first well-publicised general-purpose electronic computer, but Colossus was electronic and earlier, the Z3 was programmable and earlier, and ENIAC could not store its own program. Which machine counts as "first" depends entirely on the definition.

"Von Neumann invented the stored-program computer alone." His 1945 draft was the first widely circulated written description, but Eckert and Mauchly had worked toward the same idea, and Turing had described a stored-program design of his own for the UK's National Physical Laboratory by early 1946.

"Moore's Law is a law of physics." It was an economic observation about how many components could be put on a chip at the lowest cost per component. It held for decades partly because the industry organised its research roadmaps around it.

"Lovelace's program ran on a computer." It could not, because the Analytical Engine was never built. Modern re-implementations of Note G have found a small error in the published table, which is unsurprising for a program that could never be tested.

Artificial Intelligence: The Latest Chapter

The term "artificial intelligence" was introduced in the 1955 proposal for a summer workshop held at Dartmouth College in 1956. Early AI relied on hand-written rules and symbolic logic. Expert systems had commercial success in the 1980s but proved brittle outside narrow domains, and funding collapsed more than once in periods later called "AI winters".

The current era rests on a different approach: artificial neural networks trained on large datasets, an idea with roots in the 1940s and 1950s that only became practical once computing hardware, data and training methods caught up. In 2012 a deep convolutional network called AlexNet won the ImageNet image-recognition competition by a wide margin, and the deep-learning boom followed. In 2016 DeepMind's AlphaGo beat the professional Go player Lee Sedol, and in 2017 Google researchers published the transformer architecture that underpins today's large language models. The graphics chips used to train those models are a direct descendant of the transistor and Moore's Law story above.

This article stops there on purpose. The full sequence of people, programs, booms and winters is laid out in The History of Artificial Intelligence: A Complete Timeline. For how the networks themselves compute, see How Do Neural Networks Work?.

Open Questions

Computing history is still being written, and several of its big questions remain open. Can transistor scaling continue as features approach a few atoms across, or will gains come mainly from three-dimensional stacking and specialised chips? Will quantum computers become practical tools for specific problems, as explained in How Do Quantum Computers Actually Work?, or stay largely in the lab? And how much of the energy of a modern economy will go to running computers at all? Each earlier generation faced a similar wall, whether tube failures, the tyranny of numbers or the cost of mainframes, and each time a new technology changed the terms of the problem.

FAQ

Q: Who invented the computer? A: No single person did. Babbage designed a general-purpose programmable mechanical computer in the 1830s, Turing defined computation mathematically in 1936, Zuse completed the programmable Z3 in 1941, Flowers's Colossus (1944) and Eckert and Mauchly's ENIAC (unveiled 1946) were early electronic machines, and the Manchester Baby ran the first stored program in 1948.

Q: What was the first stored-program computer? A: The Manchester Small-Scale Experimental Machine, or "Baby", ran the first program held in electronic memory on June 21, 1948. EDSAC at Cambridge, completed in 1949, was the first such machine to provide a regular computing service to users.

Q: What is the von Neumann architecture? A: It is the design described in John von Neumann's 1945 EDVAC draft: a processor and control unit working with a single memory that holds both instructions and data, plus input and output. Most general-purpose computers still follow it, although the credit for the idea is shared with Eckert, Mauchly and others.

Q: What is Moore's Law? A: It is Gordon Moore's 1965 observation, published in Electronics, that the number of components on an integrated circuit at minimum cost was doubling each year. He revised the rate to every two years in 1975. It describes an industry trend, not a law of nature.

Q: What was the first microprocessor? A: Intel's 4004, which went on sale on November 15, 1971. It held 2,300 transistors and ran at 740 kHz, and it was originally designed for a Busicom desktop calculator.

Q: How do computers represent information? A: In binary, as patterns of 0s and 1s. Text, images, sound and video are all encoded as binary numbers, and transistors acting as switches store and process those bits. Shannon's 1937 thesis showed how Boolean logic maps onto switching circuits, which is the basis of digital design.

Q: What comes after silicon chips? A: Candidates include quantum computers for certain classes of problems, neuromorphic chips modelled on brain architecture, photonic circuits that use light, and new transistor materials. None has displaced silicon for general-purpose computing so far.

Sources

• Computer History Museum, Timeline of Computer History (Z3, Colossus, Harvard Mark I, ENIAC, Manchester Baby, EDSAC, UNIVAC, System/360, Altair, Apple II, IBM PC)
• Computer History Museum, 1965: "Moore's Law" Predicts the Future of Integrated Circuits
• Computer History Museum, 1947: Invention of the Point-Contact Transistor
• Turing, A. M. (1936). "On Computable Numbers, with an Application to the Entscheidungsproblem." Proceedings of the London Mathematical Society, Series 2, 42, 230–265. Annotated copy
• von Neumann, J. (1945). First Draft of a Report on the EDVAC, dated June 30, 1945.
• Guinness World Records, First microprocessor (Intel 4004)

---

Read next:

• How the Transistor Changed Everything
• History of Aviation: From the Wright Brothers to the Jumbo Jet and Beyond
• History of Electricity: Faraday, Tesla, Edison, and the Force That Powers the World