Evolution of Programmability: Encode, Automate, Transform

This is a story of an idea that survived every technology that was used to express it; a machine can be changed by altering its instructions, rather than rebuilding the machine.

Encoding Information

The industrialization of weaving may not, at first, seem like an important step in the history of computing. The Jacquard loom used stiff, rectangular paper cards to encode information. Holes punched into each card operated as a mechanical program to control fabric patterns. Simplified, a hole allowed a pin to pass through and raise a thread, while areas without a hole blocked the pin. These cards were laced together into a long chain, letting the loom run through an entire pattern automatically, one card at a time.

This use of punched cards was innovative in that it separated the pattern from the machine that executed it. The same loom could weave completely different designs just by swapping in a different card chain. This separation is strikingly similar to a core idea in modern computing: programmability. The same hardware can perform different tasks when it is given different instructions. The cards were also reusable and could be recombined, much as programmers reuse and combine components of code today.

Later in the 1800s, encoding information with punched holes also appeared in two other key inventions: Charles Babbage’s Analytical Engine and Herman Hollerith’s tabulating machines. The same physical idea (holes in cards) served different purposes in each. Jacquard’s cards represented weaving patterns, Babbage proposed cards that could control mathematical operations and supply data, and Hollerith’s cards represented information about people. Unlike the Jacquard loom and Hollerith’s tabulating machine, Babbage’s Analytical Engine was never actually built during his lifetime; it existed only as a design, though a remarkably influential one. What connected all three was not what the holes meant, but the larger idea that information could be encoded in a form a machine could act upon.

Together, Jacquard, Babbage, and Hollerith illustrate an important progression in what machines could do with encoded information. The Jacquard loom encoded patterns that a machine followed. Babbage’s proposed Analytical Engine encoded operations that a machine could compute. Hollerith’s tabulating machines encoded data that a machine processed. These ideas helped lead toward the modern computer, a general-purpose machine that can encode and process both programs and data.

Jacquard: Encode a Pattern That a Machine Follows

The Jacquard loom advanced the textile industry by making complex pattern manufacturing possible at scale.

Invented by Joseph Marie Charles dit Jacquard in the early 1800’s, the machine used punched cards to represent instructions for a design. A repeating pattern was produced by a set of cards, one per row. More complex designs, such as woven portraits, could require thousands of unique cards.

One of the Jacquard loom’s key features was the ability to combine cards to create different patterns. Using this modular method, machines could mass-produce textiles with far more intricate designs than manual weaving allowed, without redesigning the loom itself for each new pattern.

It’s important to note that the Jacquard loom grew out of earlier inventions and ideas. Jacquard had at least two earlier patents for machines that didn’t quite work that the Jacquard loom eventually replaced. He also built on nearly a century of prior experimentation. As outlined by D. Anderson and J. Delve in the IEEE Annals of the History of Computing, Jacquard combined and refined these earlier ideas into the machine that bears his name.

The Jacquard loom is a reminder that innovation isn’t necessarily inventing something entirely new, but about combining existing ideas in new ways; innovation is often recombination.

Punch Cards

The cards used to create a design sequence in textile weaving represented important ideas that computing would later reuse:

• Data can be encoded onto a physical medium. Information (the design) could be encoded onto a physical medium separate from the machine.
• Cards can be reused and interchanged. A sequence of cards represented a sequence of operations, meaning the same physical mechanism could execute entirely different “programs” just by swapping the deck.
• Two-state encoding. Each position on a card had one of two possible states: hole or no hole. Although Jacquard wasn’t encoding binary numbers, this is analogous to a fundamental idea in digital computing: information can be represented using two discrete states, which we now commonly represent as 1 and 0.

Modern computers no longer need holes punched into cardboard, but programs and data are still encoded into physical states that machines interpret.

Babbage: Encode Operations That a Machine Computes

These ideas weren’t confined to textiles. In the 1830s and 1840s, English mathematician Charles Babbage adapted a punch card system for his proposed Analytical Engine, which took the idea much further; cards could control the machine’s operations as well as provide information for calculations. This meant that the same machine could perform different tasks depending on the instructions it was given, an important step toward the modern idea of a programmable, general-purpose computer. Although the Analytical Engine was never completed, its design included many ideas associated with modern general-purpose computers, including memory, arithmetic operations, conditional branching, and repeated operations.

Ada Lovelace, who worked closely with Babbage (and wrote what’s often considered the first published algorithm intended to run on a machine), observed in 1843: “The Analytical Engine weaves algebraic patterns just as the Jacquard loom weaves flowers and leaves.” Her comparison wasn’t poetic embellishment. The Analytical Engine’s design borrowed the loom’s card-based logic to process abstract, symbolic information instead of thread.

Lovelace recognized that once something can be encoded symbolically, computation isn’t necessarily limited to arithmetic. Numbers, text, images, music, video and software all become things computers can manipulate because we devise systems for representing them as data.

Hollerith: Encode Data That a Machine Processes

The next major leap came in data processing rather than computation.

In the U.S. a census is taken every ten years. The data collected for the 1880 census took over seven years to process. Preparing for the 1890 census, there was a reasonable concern that, with the growth in population, it would take more than a decade to process the data using the same methods, and that it would be time for the next census before the 1890 one was completed.

In a competition for a more efficient solution, engineer Herman Hollerith built an electromechanical tabulating machine that used punch cards. He applied the already-established idea of encoding information with punched holes to an entirely different problem: processing census data. It cut the tabulation process down dramatically and demonstrated that punch cards could handle data processing on a national scale, not just textile design.

Each Hollerith card represented one person’s census data. Clerks punched holes corresponding to characteristics such as age, sex, marital status, and citizenship. The card was placed in a reader where, through an ingenious mechanism, a paper card became an electrical signal; spring-loaded pins passed through punched holes and into small cups of mercury, completing electrical circuits that advanced the appropriate counters.

Hollerith’s machines were an impressive improvement over the competition. The U.S. Census Bureau’s site notes that his data capture and preparation methods were notably more efficient; his machine captured data in about half of the time of the closest competitor, and prepared data for tabulation approximately 10x faster than the time needed for the closest competitor. Hollerith’s machine processed data for a population of nearly 63 million people during the 1890 census, demonstrating that machine-readable information could operate at an unprecedented scale.

1890 Census vs. Modern Digital Data

Comparing a modern multi-billion-row database query to Herman Hollerith’s 1890 punch card system highlights a massive evolutionary leap in structured data processing. The fundamental core concept of encoding human attributes into machine-readable formats to aggregate statistics is the same, but the execution has shifted from working with physical logistics to working with electronic manipulation.

In 1890 one Hollerith card held one person’s record; a card was a physical data record representing an individual. Millions of these records were processed and stored in cabinets for the census. To find data about a population in 1890, such as how many citizens are between the ages of 18 and 25, involved searching through physical cabinets, and a human operator manually feeding cards into a tabulating machine one by one. According to an article published in the Journal of American History, clerks processed an average of 7,000 to 8,000 cards a day.  Today, in contrast, billions of records can be stored electronically and queried without physically moving individual records through a machine.

Shift in Solving Large Scale Problems

The 1890 census was a foundational moment in the evolution of computing history because it proved that this type of technology could solve real-world, large-scale problems, and at a dramatic scale. The 1880 census, tabulated entirely by hand, had taken nearly seven years to complete. The 1890 census covered a larger population of nearly 63 million people, and Hollerith’s machines completed the count in about two years, saving the government millions of dollars in the process. That success didn’t just solve a government problem. It created a business opportunity, and it marked a shift in how organizations, especially governments and companies, began to think about handling data at scale.

Hollerith’s company later merged with others to form a company that was renamed International Business Machines (IBM) in 1924. IBM has contributed to the foundational infrastructure for modern data processing, mainframes, and personal computing. The IBM Personal Computer, released in 1981, established an architecture that was widely copied by manufacturers of ‘IBM-compatible’ computers and became enormously influential in the personal-computer market. The substantial market for IBM-compatible computers helped establish the hardware lineage that continues in many Windows PCs today.

The “IBM Computer Card” evolved and became standardized at 80 columns, dramatically increasing how much information a card could hold. Punched cards then became central to business data processing and programming for decades. Although we no longer rely on card-based computing, the Computer History Museum describes punched cards as a primary data-storage system for roughly 80 years. This could be a testament to how durable the technique used by Jacquard’s loom was. It could also be an indication that technologies often survive because entire systems grow around them.

Modern Computing: Encode Programs and Data

The more direct precursors of modern computers emerged in the mid-1900s. These were machines that were digital, programmable, and discrete, rather than analog.

The next major shift was eliminating the distinction between the medium holding instructions and the machine’s internal memory. Early electronic computers like ENIAC, completed in 1945, were programmed by physically rewiring circuits (connecting cables, flipping switches, and resetting plugboards) a process that could take days for a single change.

In 1945, mathematician John von Neumann proposed a different approach: store the program’s instructions in electronic memory alongside the data itself, so the machine could read and execute them without any physical rewiring. Machines built on this idea, like the Manchester Baby in 1948, became the first to run a program stored entirely in electronic memory.

The physical card or cable was no longer necessary for a machine to retain its instructions. But the idea introduced much earlier remained: the same hardware could perform different tasks depending on the instructions it received.

Summary

From weaving patterns, to counting people, to powering computers, the throughline is the same: encode, automate, transform. Jacquard, Babbage, and Hollerith each put holes in cards to represent something different (a pattern, a calculation, a person) but in every case, the specific meaning of the holes mattered less than the underlying idea: that information could be encoded in a form a machine could act upon. That idea is what carried forward into the modern computer, a general-purpose machine that can encode and process both programs and data.

ENCODE: Represent information in a form a machine can interpret.
AUTOMATE: Give the machine rules for acting on that representation.
TRANSFORM: Change what the machine can accomplish without necessarily changing the machine itself.

The punch card eventually disappeared. The idea behind it didn’t.

References

• Anderson, D., and J. Delve. “Biographies [F.C. Williams; J. Vaucanson; J.M. Jacquard].” IEEE Annals of the History of Computing 29, no. 4 (2007): 90–102. https://doi.org/10.1109/MAHC.2007.4407450.
• Barlow, Alfred. The History and Principles of Weaving by Hand and by Power. London: S. Low, Marston, Searle & Rivington, 1878. https://archive.org/details/historyandprinc00barlgoog/page/n158/mode/2up.
• Computer History Museum. “Babbage Engine History.” Accessed August 28, 2026. https://www.computerhistory.org/babbage/history.
• Computer History Museum. “The Punched Card’s Pedigree.” Accessed August 28, 2026. https://www.computerhistory.org/revolution/punched-cards/2/4.
• Computer History Museum. “Punched Cards: Control for the Jacquard Loom.” Accessed August 28, 2026. https://www.computerhistory.org/storageengine/punched-cards-control-jacquard-loom.
• Cruz, Frank da. “Herman Hollerith.” Columbia University, 2023. https://www.columbia.edu/cu/computinghistory/hollerith.html.
• Cruz, Frank da. “The Jacquard Loom.” Columbia University, 2015. https://www.columbia.edu/cu/computinghistory/jacquard.html.
• Ruggles, Steven, and Daniel L. Magnuson. “Census Technology, Politics, and Institutional Change, 1790–2020.” Journal of American History 107, no. 1 (2020): 19–51. https://doi.org/10.1093/jahist/jaaa007.
• Science and Industry Museum. “Programming Patterns: The Story of the Jacquard Loom.” https://www.scienceandindustrymuseum.org.uk/objects-and-stories/jacquard-loom.
• U.S. Census Bureau. “1890 Census.” Accessed August 28, 2026. https://www.census.gov/programs-surveys/decennial-census/decade/1890/about-1890.html.
• U.S. Census Bureau. “Hollerith Machine.” Accessed August 28, 2026. https://www.census.gov/about/history/bureau-history/census-innovations/technology/hollerith-machine.html.
• U.S. Census Bureau. “Tabulation and Processing.” Accessed August 28, 2026. https://www.census.gov/about/history/bureau-history/census-innovations/technology/tabulation-and-processing.html.

Primary Source

• Lovelace, Ada. “Notes on the Analytical Engine.” 1843. Reproduced online by Fourmilab. https://www.fourmilab.ch/babbage/sketch.html.

Additional Online Source

• Google Arts & Culture. “Punched Card Machines: The National Museum of Computing.” https://artsandculture.google.com/story/punched-card-machines-the-national-museum-of-computing/bwWBrooyeGKPiA.

Expand Your Knowledge

• Computational Thinking: A Framework for Problem Solving
• Numeral Systems and Data Representation
• Binary Numeral System (Base-2)
• Decimal Numeral System (Base-10)
• Hexadecimal Numeral System (Base-16)
• Bits and Binary: The Language of Computers
• Decimal and Binary: Numeral System Conversions