How computers began to slowly replace humans | David Alan Grier: Full Interview

Key Concepts

• Computing is not a recent invention, but has roots in the systematization of labor during the Industrial Revolution.
• The development of computing has been driven by the need to manage large datasets, ensure accuracy, and reduce costs.
• Standardization and the division of labor were crucial to the growth of both industrialization and early computing.
• The ARPANET’s initial focus was on building a community of computer scientists and facilitating human-to-human communication, with search functionality as a core goal.
• Humans adapt to systems, continuously negotiating between their preferences and computational outputs.
• The introduction of automation has historically sparked conflict over the ownership of skills and expertise.
• AI development follows a cyclical pattern of excitement, disillusionment, and eventual progress, and requires ongoing human adaptation.

The Historical Roots of Computing

David Alan Grier frames computing not as a late 20th-century phenomenon, but as integral to the Industrial Revolution, beginning in the 18th century. The core principle of computing – reducing complex tasks to fixed, repeatable steps – was present from the outset of industrialization, as exemplified by Adam Smith’s analysis of the division of labor in The Wealth of Nations (1776). The need for efficient calculation arose from the massive datasets generated by 18th-century astronomical observations, leading to the employment of “human computers” performing repetitive calculations.

The calculation of Halley’s Comet’s return provided a key example of this division of labor, with French astronomers dividing tasks among teams focused on Earth’s, Jupiter’s, and the comet’s locations. This, coupled with error checking, became a foundational principle. Accurate navigation and mapping further fueled this development, requiring precise astronomical data codified in nautical almanacs, necessitating systems for both calculation and error detection – exemplified by Babbage’s Rule, which recognizes the tendency for individuals to make the same errors when performing the same calculation in the same way.

Mechanization and Standardization

Charles Babbage, a 19th-century inventor, attempted to build mechanical calculating machines, recognizing the potential to systematize calculations and reduce costs. While his Difference Engine faced engineering limitations, his work laid the groundwork for future developments, representing an attempt to “industrialize calculation.” George Boole’s Boolean algebra provided a crucial theoretical foundation, though its connection to Babbage’s practical approach was less direct.

The importance of standardization, linked to the rise of mass production, is highlighted. Prior to the 20th century, manufacturing relied on bespoke parts. Herbert Hoover, during World War I, championed standardization to increase industrial scale and reduce costs, leading to the creation of institutions like the National Bureau of Standards and organizations like the IEEE. Standardization allowed for less-skilled workers to participate in large-scale data processing, driving efficiency.

From Census to Electronic Computers

The US Census and the Hollerith Tabulating Company (later IBM) exemplify this trend. The 1890 census, overwhelmed by data, adopted punched card technology, dramatically speeding up tabulation. This demonstrated the power of large-scale data processing and the potential to reveal societal insights. Grier notes a tendency to anthropomorphize the data and the machines processing it.

The development of electronic computers, specifically the ENIAC project during World War II, followed. John von Neumann’s contribution was crucial, abstracting the principles of the ENIAC into a generalized computer architecture consisting of a memory unit, a processing unit, and a program decoder – the fundamental components of modern computers. The ARPANET, the precursor to the Internet, was also developed, emphasizing human-to-human communication, data repositories, and the importance of search functionality. Its goal was to establish computing as a recognized discipline.

The ARPANET and Human-System Interaction

The ARPANET originated from the Department of Defense’s desire to establish computer science as a discipline in the 1960s. Early applications prioritized human-to-human communication, with email emerging as a key feature after establishing reliable file and message transfer (a process taking approximately two years). Efficient searching and access to information repositories were core concepts from the beginning, stemming from early computational challenges in algorithms for multiplication and division.

A computerized dating experiment at Stanford in the early 1960s serves as a cautionary tale, echoing earlier reactions to technologies like Hermann Hallruth’s machines, where people projected meaning onto the data. This pattern continued with the rise of the personal computer (PC) in the 1970s, where customization became an expression of identity. A fundamental principle is that humans adapt to systems, adjusting their thoughts and behaviors to align with the system’s design, illustrated by the evolution of phone usage and the use of the A* search algorithm in modern navigation apps. Users continuously negotiate between their preferences and computational output, guided by questions of utility, effort, and irritation.

Labor, Automation, and the Cyclical Nature of AI

The transition from human “computers” to machine computation involved a division of labor. The Math Tables Project in New York City during the 1930s employed approximately 450 individuals to perform repetitive calculations, prioritizing accuracy through redundant computation (each number calculated 6-7 times). This project contributed to military efforts, including D-Day mine clearing and, unknowingly, to the Manhattan Project. It highlights the tension between labor and automation, and the question of who owns the skills and knowledge generated through work.

The introduction of automated machine tools in the 1950s sparked conflict between manufacturers and skilled machinists over the ownership of expertise. This foreshadows contemporary debates surrounding data ownership and AI, where algorithms are trained on human data. The speaker cautions against overestimating current AI capabilities, noting that initial excitement often gives way to practical limitations, as seen with early machine translation failures. AI is not a monolithic entity but a collection of technologies, with current progress largely due to increased computing power and data availability.

Conclusion

The history of computing is deeply intertwined with the history of industrialization, driven by the need to manage complexity, ensure accuracy, and reduce costs. From the division of labor in 18th-century astronomy to the development of the ARPANET and modern AI, the story is one of continuous adaptation – both of technology to human needs and of humans to the systems they create. The ongoing development of AI necessitates a continuous adjustment of human thought and behavior, and a critical understanding of the historical context in which these technologies emerge.