Artificial intelligence can seem like something that appeared almost overnight.

ChatGPT, self-driving vehicles, image generators, voice assistants and computer systems capable of recognising faces or translating languages can give the impression that humanity suddenly crossed some technological threshold in the twenty-first century.

In reality, the foundations of artificial intelligence stretch back centuries.

And one of the most unlikely ancestors of the modern computer was a machine designed not to calculate numbers, but to weave cloth.

Its name was the Jacquard loom.

A Machine That Could Follow Instructions

In 1804–1805 the French inventor Joseph-Marie Jacquard demonstrated an improved mechanism for controlling looms.

Complex woven patterns traditionally required considerable skill and labour. In some weaving systems assistants had to manipulate individual threads according to the pattern being produced.

Jacquard helped automate this process.

His mechanism used a sequence of punched cards. Holes in each card determined which warp threads were raised and which remained lowered as the shuttle passed through them.

A hole represented one instruction.

The absence of a hole represented another.

By arranging cards in the correct sequence, extremely complicated patterns could be produced automatically and repeatedly.

The loom itself did not understand the design.

It simply followed encoded instructions.

Yet that seemingly simple idea contained something revolutionary:

information could be separated from the machine performing the work.

Change the cards and the same machine could perform a different sequence of operations.

In modern terminology, the Jacquard loom had become, in a rudimentary sense, programmable.

Punch Cards Leave the Weaving Shed

The significance of Jacquard’s cards did not remain confined to textiles.

Several decades later the English mathematician and inventor Charles Babbage began designing what he called the Analytical Engine.

Unlike ordinary calculating machines, Babbage imagined a general-purpose machine capable of performing different calculations depending upon the instructions supplied to it.

For inspiration, he turned to Jacquard.

Babbage proposed using punched cards to control the Analytical Engine.

There would be cards containing instructions and cards containing data.

The connection was remarkably direct: the same principle that allowed a loom to weave different patterns could allow a calculating machine to perform different mathematical operations.

Babbage famously owned a woven portrait of Jacquard produced on a Jacquard loom using thousands of punched cards.

The textile machine had helped inspire the architecture of computing.

Ada Lovelace Sees Something More

Babbage’s collaborator Ada Lovelace recognised an implication that went considerably beyond arithmetic.

In her famous notes on the Analytical Engine, published in 1843, Lovelace observed that a machine capable of manipulating symbols according to rules might potentially work with things other than numbers.

If musical relationships, for example, could be expressed mathematically, she suggested that such a machine might theoretically compose elaborate pieces of music.

This was an extraordinary intellectual leap.

Lovelace understood that computation was fundamentally about the manipulation of symbols and information, not simply numbers.

Modern computers operate on precisely that principle.

Text, photographs, music, films and computer programs are all ultimately converted into numerical representations that computers can manipulate.

Punch Cards Meet the Census

Another major step occurred in the United States during the late nineteenth century.

The 1880 US census had produced such an enormous quantity of information that officials feared the next census might take longer than ten years to tabulate.

Engineer Herman Hollerith developed an electromechanical system for processing information recorded on punched cards.

For the 1890 census, information about individuals could be encoded by punching holes into predetermined positions on cards.

Machines then electrically detected those holes and rapidly counted and sorted the information.

The system dramatically accelerated census processing.

Hollerith later helped establish a company that eventually became part of International Business Machines — IBM.

Punch cards subsequently became one of the defining technologies of twentieth-century computing.

For decades they were used by governments, universities, banks and businesses to enter both programs and data into computers.

A direct conceptual line now ran from:

loom → punched card → programmable calculating machine → data-processing machine → electronic computer.

From Mechanical Machines to Electronic Computers

During the twentieth century computation moved from mechanical and electromechanical devices to fully electronic machines.

The Second World War accelerated developments enormously.

Machines such as Britain’s Colossus, developed to assist in decrypting German communications, demonstrated the extraordinary speed of electronic computation.

In the United States, ENIAC, completed in the 1940s, became one of the first large-scale general-purpose electronic digital computers.

Early computers were enormous.

They filled rooms, consumed large amounts of electricity and possessed less computing power than devices people now carry in their pockets.

Programming them could involve switches, cables, paper tape or punched cards.

But something fundamental had changed.

Machines could now perform thousands — eventually millions and billions — of logical operations extremely rapidly.

Alan Turing and the Question of Machine Intelligence

One of the crucial intellectual figures linking computing with artificial intelligence was the British mathematician Alan Turing.

In 1936 Turing had described a theoretical computing device — now known as a Turing machine — demonstrating how a machine following sufficiently precise instructions could perform any calculation that was mathematically computable.

After the Second World War he became increasingly interested in machine intelligence.

In his influential 1950 paper Computing Machinery and Intelligence, Turing opened with a provocative question:

Can machines think?

Rather than trying to define thought itself, he proposed what became known as the Imitation Game, later popularly called the Turing Test.

If a human communicating through text could not reliably distinguish a computer from another human, Turing suggested that describing the machine as intelligent might be reasonable.

Computers were no longer simply being considered calculating devices.

They were becoming potential thinking machines.

Artificial Intelligence Gets a Name

The expression artificial intelligence was formally introduced during preparations for a research project held at Dartmouth College in the United States in 1956.

Computer scientist John McCarthy and colleagues proposed investigating whether aspects of intelligence could be described precisely enough for machines to simulate them.

Researchers initially possessed extraordinary optimism.

Some predicted that machines capable of human-like reasoning might appear within a generation.

Early programs demonstrated impressive abilities.

Computers could solve mathematical problems, play simple games and manipulate logical statements.

Yet human intelligence proved considerably more complicated than anticipated.

Computers could perform astonishing calculations while struggling with things a small child found effortless.

Recognising an object, understanding ordinary conversation or interpreting ambiguous language proved enormously difficult.

The AI Winters

Artificial intelligence subsequently experienced cycles of enthusiasm and disappointment.

Governments and universities invested heavily when breakthroughs appeared imminent.

When promised advances failed to materialise, funding declined.

These periods became known as AI winters.

The problem was not simply inadequate ideas.

Computers lacked sufficient processing power, memory and data.

Many concepts that researchers developed during the twentieth century could not become practical until computing technology caught up with them.

Machines That Learn

Traditional computer programming involves humans writing explicit instructions.

If this happens, do that.

If something else happens, perform another operation.

Machine learning takes a different approach.

Instead of specifying every rule, programmers provide systems with large quantities of examples from which statistical patterns can be identified.

Suppose a computer is required to recognise cats.

Rather than attempting to describe every possible cat — something extraordinarily difficult — researchers can provide thousands or millions of labelled images.

The system gradually identifies patterns associated with cats.

This approach became increasingly powerful as computers became faster and enormous digital datasets became available.

Neural Networks

One important branch of machine learning was inspired loosely by biological brains.

Artificial neural networks consist of interconnected mathematical units that process information through multiple layers.

The idea is not new.

Researchers experimented with artificial neural networks as early as the 1940s and 1950s.

For decades, however, their practical usefulness remained limited.

Eventually three developments transformed the field:

vast quantities of digital data;

enormously increased computing power;

and improved techniques for training large neural networks.

The result became known as deep learning.

Computers became remarkably effective at recognising images, transcribing speech, translating languages and detecting patterns within enormous collections of information.

From Recognising Information to Creating It

Another major step was generative artificial intelligence.

Instead of merely classifying information, these systems could produce new material based upon patterns learned during training.

AI systems began generating photographs, paintings, music, computer code and human-like language.

Large language models are trained using enormous collections of text.

They learn statistical relationships between words, phrases and concepts.

When responding to a prompt they generate sequences of language based upon patterns contained within their training and subsequent refinement.

The mechanism is vastly more complicated than a punched card.

Yet something about it would have been strangely familiar to Jacquard.

Information enters a machine.

The machine interprets patterns within that information.

The resulting instructions or probabilities determine what happens next.

The scale has changed almost beyond recognition.

The underlying human ambition has not.

Does AI Actually Think?

That remains one of the great unresolved philosophical questions.

Modern artificial intelligence can produce remarkably convincing language and solve complicated problems, but this does not necessarily mean it possesses consciousness, self-awareness or understanding comparable to a human mind.

A language model can describe grief without experiencing grief.

It can write about hunger without becoming hungry.

It can discuss fear without being afraid.

The distinction matters.

Artificial intelligence is extraordinarily capable pattern-processing technology.

Whether machines could someday possess genuine consciousness remains an unanswered scientific and philosophical problem.

Automation and Fear

There is another connection between Jacquard and artificial intelligence that has remarkably little to do with engineering.

It is fear of being replaced.

Mechanised textile production profoundly disrupted traditional employment.

The Luddites who attacked machinery in early nineteenth-century Britain were not simply ignorant people frightened by technology.

Many were skilled workers responding to machinery being introduced in ways that reduced wages, weakened traditional working practices and transferred economic power towards factory owners.

The same argument accompanies artificial intelligence today.

AI can automate tasks performed by translators, artists, administrators, programmers, journalists and many other workers.

The historical lesson is therefore not that technology should be stopped.

Technology rarely is.

The important question is who benefits from automation and who carries its costs.

That argument was being fought in weaving districts two centuries before anyone had heard of ChatGPT.

The Thread Running Through History

Look at the development of computing from a distance and the connections become striking.

Jacquard demonstrated that instructions could be encoded on cards.

Babbage realised those cards could control a calculating machine.

Ada Lovelace realised such machines might manipulate symbols rather than merely numbers.

Hollerith demonstrated that punched cards could process enormous quantities of human information.

Electronic computers made calculation unimaginably faster.

Turing asked whether those machines might ultimately imitate intelligence.

Computer scientists developed artificial intelligence.

Machine learning allowed computers to discover patterns rather than simply follow explicitly programmed instructions.

And modern generative AI learned to produce language, images, music and other forms of information.

The journey from a nineteenth-century weaving shed to artificial intelligence was therefore not a sudden technological revolution.

It was a chain of ideas stretching across more than two centuries.

Perhaps the most fitting metaphor is Jacquard’s own loom.

Every technological generation has added another thread.

Punched cards.

Mechanical computation.

Electronic circuits.

Programming languages.

Microprocessors.

The internet.

Machine learning.

Neural networks.

Artificial intelligence.

Individually they appear to be separate inventions.

Seen together, they form a single pattern.

And somewhere near the beginning of that enormous technological tapestry sits Joseph-Marie Jacquard, watching holes punched into pieces of cardboard tell a machine what to do.

And there is one delicious historical circle in it: the argument around AI and employment takes us straight back to the machinery, textile workers and Luddites we’ve just been wandering through. The technology has altered almost beyond comprehension; the social question — what happens to people when machines can do their work? — stubbornly refuses to go away.

Tim Alderman ©️2026